Fwd: ThreadLocal in SegmentReader

classic Classic list List threaded Threaded
22 messages Options
12
Reply | Threaded
Open this post in threaded view
|

Fwd: ThreadLocal in SegmentReader

Michael McCandless-2

ThreadLocal, which we use in several places in Lucene, causes a leak  
in app servers because the classloader never fully deallocates  
Lucene's classes because the ThreadLocal is holding strong references.

Yet, ThreadLocal is very convenient for avoiding synchronization.

Does anyone have any ideas on how to solve this w/o falling back to  
"normal" synchronization?

Mike

Begin forwarded message:

> From: "Yonik Seeley" <[hidden email]>
> Date: July 7, 2008 3:30:28 PM EDT
> To: [hidden email]
> Subject: Re: ThreadLocal in SegmentReader
> Reply-To: [hidden email]
>
> On Mon, Jul 7, 2008 at 2:43 PM, Michael McCandless
> <[hidden email]> wrote:
>> So now I'm confused: the SegmentReader itself should no longer be  
>> reachable,
>> assuming you are not holding any references to your IndexReader.
>>
>> Which means the ThreadLocal instance should no longer be reachable.
>
> It will still be referenced from the Thread(s) ThreadLocalMap
> The key (the ThreadLocal) will be weakly referenced, but the values
> (now stale) are strongly referenced and won't be actually removed
> until the table is resized (under the Java6 impl at least).
> Nice huh?
>
> -Yonik
>
> ---------------------------------------------------------------------
> To unsubscribe, e-mail: [hidden email]
> For additional commands, e-mail: [hidden email]
>


---------------------------------------------------------------------
To unsubscribe, e-mail: [hidden email]
For additional commands, e-mail: [hidden email]

Reply | Threaded
Open this post in threaded view
|

Re: Fwd: ThreadLocal in SegmentReader

Adrian Tarau
Usually ThreadLocal.remove() should be called at the end(in a finally
block), before the current call leaves your code.

Ex : if during searching ThreadLocal is used, every search(..) method
should cleanup any ThreadLocal variables, or even deeper in the
implementation. When the call leaves Lucene any used ThreadLocal should
be cleaned up.

Michael McCandless wrote:

>
> ThreadLocal, which we use in several places in Lucene, causes a leak
> in app servers because the classloader never fully deallocates
> Lucene's classes because the ThreadLocal is holding strong references.
>
> Yet, ThreadLocal is very convenient for avoiding synchronization.
>
> Does anyone have any ideas on how to solve this w/o falling back to
> "normal" synchronization?
>
> Mike
>
> Begin forwarded message:
>
>> From: "Yonik Seeley" <[hidden email]>
>> Date: July 7, 2008 3:30:28 PM EDT
>> To: [hidden email]
>> Subject: Re: ThreadLocal in SegmentReader
>> Reply-To: [hidden email]
>>
>> On Mon, Jul 7, 2008 at 2:43 PM, Michael McCandless
>> <[hidden email]> wrote:
>>> So now I'm confused: the SegmentReader itself should no longer be
>>> reachable,
>>> assuming you are not holding any references to your IndexReader.
>>>
>>> Which means the ThreadLocal instance should no longer be reachable.
>>
>> It will still be referenced from the Thread(s) ThreadLocalMap
>> The key (the ThreadLocal) will be weakly referenced, but the values
>> (now stale) are strongly referenced and won't be actually removed
>> until the table is resized (under the Java6 impl at least).
>> Nice huh?
>>
>> -Yonik
>>
>> ---------------------------------------------------------------------
>> To unsubscribe, e-mail: [hidden email]
>> For additional commands, e-mail: [hidden email]
>>
>
>
> ---------------------------------------------------------------------
> To unsubscribe, e-mail: [hidden email]
> For additional commands, e-mail: [hidden email]
>


---------------------------------------------------------------------
To unsubscribe, e-mail: [hidden email]
For additional commands, e-mail: [hidden email]

Reply | Threaded
Open this post in threaded view
|

Re: ThreadLocal in SegmentReader

Michael McCandless-2

Well ... SegmentReader uses ThreadLocal to hold a thread-private  
instance of TermVectorsReader, to avoid synchronizing per-document  
when loading term vectors.

Clearing this ThreadLocal value per call to SegmentReader's methods  
that load term vectors would defeat its purpose.

Though, of course, we then synchronize on the underlying file (when  
using FSDirectory), so perhaps we are really not saving much by using  
ThreadLocal here.  But we are looking to relax that low level  
synchronization with LUCENE-753.

Maybe we could make our own ThreadLocal that just uses a HashMap,  
which we'd have to synchronize on when getting the per-thread  
instances.  Or, go back to sharing a single TermVectorsReader and  
synchronize per-document.

Jason has suggested moving to a model where you ask the IndexReader  
for an object that can return term vectors / stored fields / etc, and  
then you interact with that many times to retrieve each doc.  We could  
then synchronize only on retrieving that object, and provide a thread-
private instance.

It seems like we should move away from using ThreadLocal in Lucene and  
do "normal" synchronization instead.

Mike

Adrian Tarau wrote:

> Usually ThreadLocal.remove() should be called at the end(in a  
> finally block), before the current call leaves your code.
>
> Ex : if during searching ThreadLocal is used, every search(..)  
> method should cleanup any ThreadLocal variables, or even deeper in  
> the implementation. When the call leaves Lucene any used ThreadLocal  
> should be cleaned up.
>
> Michael McCandless wrote:
>>
>> ThreadLocal, which we use in several places in Lucene, causes a  
>> leak in app servers because the classloader never fully deallocates  
>> Lucene's classes because the ThreadLocal is holding strong  
>> references.
>>
>> Yet, ThreadLocal is very convenient for avoiding synchronization.
>>
>> Does anyone have any ideas on how to solve this w/o falling back to  
>> "normal" synchronization?
>>
>> Mike
>>
>> Begin forwarded message:
>>
>>> From: "Yonik Seeley" <[hidden email]>
>>> Date: July 7, 2008 3:30:28 PM EDT
>>> To: [hidden email]
>>> Subject: Re: ThreadLocal in SegmentReader
>>> Reply-To: [hidden email]
>>>
>>> On Mon, Jul 7, 2008 at 2:43 PM, Michael McCandless
>>> <[hidden email]> wrote:
>>>> So now I'm confused: the SegmentReader itself should no longer be  
>>>> reachable,
>>>> assuming you are not holding any references to your IndexReader.
>>>>
>>>> Which means the ThreadLocal instance should no longer be reachable.
>>>
>>> It will still be referenced from the Thread(s) ThreadLocalMap
>>> The key (the ThreadLocal) will be weakly referenced, but the values
>>> (now stale) are strongly referenced and won't be actually removed
>>> until the table is resized (under the Java6 impl at least).
>>> Nice huh?
>>>
>>> -Yonik
>>>
>>> ---------------------------------------------------------------------
>>> To unsubscribe, e-mail: [hidden email]
>>> For additional commands, e-mail: [hidden email]
>>>
>>
>>
>> ---------------------------------------------------------------------
>> To unsubscribe, e-mail: [hidden email]
>> For additional commands, e-mail: [hidden email]
>>
>
>
> ---------------------------------------------------------------------
> To unsubscribe, e-mail: [hidden email]
> For additional commands, e-mail: [hidden email]
>


---------------------------------------------------------------------
To unsubscribe, e-mail: [hidden email]
For additional commands, e-mail: [hidden email]

Reply | Threaded
Open this post in threaded view
|

Re: ThreadLocal in SegmentReader

Jason Rutherglen
A custom thread local (synchronized hashmap with thread as the key) that uses a global static manager may work.  The global manager can have a close method for explicit dereferencing.  The thread local registers itself with the global thread local manager.  The global manager is static and tied to the classloader which means it would be garbage collected in the case of a reloading webapp. 

On Tue, Jul 8, 2008 at 6:12 AM, Michael McCandless <[hidden email]> wrote:

Well ... SegmentReader uses ThreadLocal to hold a thread-private instance of TermVectorsReader, to avoid synchronizing per-document when loading term vectors.

Clearing this ThreadLocal value per call to SegmentReader's methods that load term vectors would defeat its purpose.

Though, of course, we then synchronize on the underlying file (when using FSDirectory), so perhaps we are really not saving much by using ThreadLocal here.  But we are looking to relax that low level synchronization with LUCENE-753.

Maybe we could make our own ThreadLocal that just uses a HashMap, which we'd have to synchronize on when getting the per-thread instances.  Or, go back to sharing a single TermVectorsReader and synchronize per-document.

Jason has suggested moving to a model where you ask the IndexReader for an object that can return term vectors / stored fields / etc, and then you interact with that many times to retrieve each doc.  We could then synchronize only on retrieving that object, and provide a thread-private instance.

It seems like we should move away from using ThreadLocal in Lucene and do "normal" synchronization instead.

Mike


Adrian Tarau wrote:

Usually ThreadLocal.remove() should be called at the end(in a finally block), before the current call leaves your code.

Ex : if during searching ThreadLocal is used, every search(..) method should cleanup any ThreadLocal variables, or even deeper in the implementation. When the call leaves Lucene any used ThreadLocal should be cleaned up.

Michael McCandless wrote:

ThreadLocal, which we use in several places in Lucene, causes a leak in app servers because the classloader never fully deallocates Lucene's classes because the ThreadLocal is holding strong references.

Yet, ThreadLocal is very convenient for avoiding synchronization.

Does anyone have any ideas on how to solve this w/o falling back to "normal" synchronization?

Mike

Begin forwarded message:

From: "Yonik Seeley" <[hidden email]>
Date: July 7, 2008 3:30:28 PM EDT
To: [hidden email]
Subject: Re: ThreadLocal in SegmentReader
Reply-To: [hidden email]

On Mon, Jul 7, 2008 at 2:43 PM, Michael McCandless
<[hidden email]> wrote:
So now I'm confused: the SegmentReader itself should no longer be reachable,
assuming you are not holding any references to your IndexReader.

Which means the ThreadLocal instance should no longer be reachable.

It will still be referenced from the Thread(s) ThreadLocalMap
The key (the ThreadLocal) will be weakly referenced, but the values
(now stale) are strongly referenced and won't be actually removed
until the table is resized (under the Java6 impl at least).
Nice huh?

-Yonik

---------------------------------------------------------------------
To unsubscribe, e-mail: [hidden email]
For additional commands, e-mail: [hidden email]



---------------------------------------------------------------------
To unsubscribe, e-mail: [hidden email]
For additional commands, e-mail: [hidden email]



---------------------------------------------------------------------
To unsubscribe, e-mail: [hidden email]
For additional commands, e-mail: [hidden email]



---------------------------------------------------------------------
To unsubscribe, e-mail: [hidden email]
For additional commands, e-mail: [hidden email]


Reply | Threaded
Open this post in threaded view
|

Re: ThreadLocal in SegmentReader

Robert Engels
In reply to this post by Michael McCandless-2
Using synchronization is a poor/invalid substitute for thread locals  
in many cases.

The point of the thread local in these referenced cases is too allow  
streaming reads on a file descriptor. if you use a shared file  
descriptor/buffer you are going to continually invalidate the buffer.

On Jul 8, 2008, at 5:12 AM, Michael McCandless wrote:

>
> Well ... SegmentReader uses ThreadLocal to hold a thread-private  
> instance of TermVectorsReader, to avoid synchronizing per-document  
> when loading term vectors.
>
> Clearing this ThreadLocal value per call to SegmentReader's methods  
> that load term vectors would defeat its purpose.
>
> Though, of course, we then synchronize on the underlying file (when  
> using FSDirectory), so perhaps we are really not saving much by  
> using ThreadLocal here.  But we are looking to relax that low level  
> synchronization with LUCENE-753.
>
> Maybe we could make our own ThreadLocal that just uses a HashMap,  
> which we'd have to synchronize on when getting the per-thread  
> instances.  Or, go back to sharing a single TermVectorsReader and  
> synchronize per-document.
>
> Jason has suggested moving to a model where you ask the IndexReader  
> for an object that can return term vectors / stored fields / etc,  
> and then you interact with that many times to retrieve each doc.  
> We could then synchronize only on retrieving that object, and  
> provide a thread-private instance.
>
> It seems like we should move away from using ThreadLocal in Lucene  
> and do "normal" synchronization instead.
>
> Mike
>
> Adrian Tarau wrote:
>
>> Usually ThreadLocal.remove() should be called at the end(in a  
>> finally block), before the current call leaves your code.
>>
>> Ex : if during searching ThreadLocal is used, every search(..)  
>> method should cleanup any ThreadLocal variables, or even deeper in  
>> the implementation. When the call leaves Lucene any used  
>> ThreadLocal should be cleaned up.
>>
>> Michael McCandless wrote:
>>>
>>> ThreadLocal, which we use in several places in Lucene, causes a  
>>> leak in app servers because the classloader never fully  
>>> deallocates Lucene's classes because the ThreadLocal is holding  
>>> strong references.
>>>
>>> Yet, ThreadLocal is very convenient for avoiding synchronization.
>>>
>>> Does anyone have any ideas on how to solve this w/o falling back  
>>> to "normal" synchronization?
>>>
>>> Mike
>>>
>>> Begin forwarded message:
>>>
>>>> From: "Yonik Seeley" <[hidden email]>
>>>> Date: July 7, 2008 3:30:28 PM EDT
>>>> To: [hidden email]
>>>> Subject: Re: ThreadLocal in SegmentReader
>>>> Reply-To: [hidden email]
>>>>
>>>> On Mon, Jul 7, 2008 at 2:43 PM, Michael McCandless
>>>> <[hidden email]> wrote:
>>>>> So now I'm confused: the SegmentReader itself should no longer  
>>>>> be reachable,
>>>>> assuming you are not holding any references to your IndexReader.
>>>>>
>>>>> Which means the ThreadLocal instance should no longer be  
>>>>> reachable.
>>>>
>>>> It will still be referenced from the Thread(s) ThreadLocalMap
>>>> The key (the ThreadLocal) will be weakly referenced, but the values
>>>> (now stale) are strongly referenced and won't be actually removed
>>>> until the table is resized (under the Java6 impl at least).
>>>> Nice huh?
>>>>
>>>> -Yonik
>>>>
>>>> -------------------------------------------------------------------
>>>> --
>>>> To unsubscribe, e-mail: [hidden email]
>>>> For additional commands, e-mail: [hidden email]
>>>>
>>>
>>>
>>> --------------------------------------------------------------------
>>> -
>>> To unsubscribe, e-mail: [hidden email]
>>> For additional commands, e-mail: [hidden email]
>>>
>>
>>
>> ---------------------------------------------------------------------
>> To unsubscribe, e-mail: [hidden email]
>> For additional commands, e-mail: [hidden email]
>>
>
>
> ---------------------------------------------------------------------
> To unsubscribe, e-mail: [hidden email]
> For additional commands, e-mail: [hidden email]
>


---------------------------------------------------------------------
To unsubscribe, e-mail: [hidden email]
For additional commands, e-mail: [hidden email]

Reply | Threaded
Open this post in threaded view
|

Re: ThreadLocal in SegmentReader

Adrian Tarau
 From what I know, storing objects in ThreadLocal is safe as long as you
release the object within a try {} finall {} block or store objects
which are independent of the rest of the code(no dependencies).Otherwise
it can get pretty tricky(memory leaks, classloader problems) after awhile.

It is pretty convenient to pass HTTP request information with a
ThreadLocal in a servlet(but you should cleanup the variable before
leaving the servlet) but I'm not sure how safe it is in this case.

robert engels wrote:

> Using synchronization is a poor/invalid substitute for thread locals
> in many cases.
>
> The point of the thread local in these referenced cases is too allow
> streaming reads on a file descriptor. if you use a shared file
> descriptor/buffer you are going to continually invalidate the buffer.
>
> On Jul 8, 2008, at 5:12 AM, Michael McCandless wrote:
>
>>
>> Well ... SegmentReader uses ThreadLocal to hold a thread-private
>> instance of TermVectorsReader, to avoid synchronizing per-document
>> when loading term vectors.
>>
>> Clearing this ThreadLocal value per call to SegmentReader's methods
>> that load term vectors would defeat its purpose.
>>
>> Though, of course, we then synchronize on the underlying file (when
>> using FSDirectory), so perhaps we are really not saving much by using
>> ThreadLocal here.  But we are looking to relax that low level
>> synchronization with LUCENE-753.
>>
>> Maybe we could make our own ThreadLocal that just uses a HashMap,
>> which we'd have to synchronize on when getting the per-thread
>> instances.  Or, go back to sharing a single TermVectorsReader and
>> synchronize per-document.
>>
>> Jason has suggested moving to a model where you ask the IndexReader
>> for an object that can return term vectors / stored fields / etc, and
>> then you interact with that many times to retrieve each doc.  We
>> could then synchronize only on retrieving that object, and provide a
>> thread-private instance.
>>
>> It seems like we should move away from using ThreadLocal in Lucene
>> and do "normal" synchronization instead.
>>
>> Mike
>>
>> Adrian Tarau wrote:
>>
>>> Usually ThreadLocal.remove() should be called at the end(in a
>>> finally block), before the current call leaves your code.
>>>
>>> Ex : if during searching ThreadLocal is used, every search(..)
>>> method should cleanup any ThreadLocal variables, or even deeper in
>>> the implementation. When the call leaves Lucene any used ThreadLocal
>>> should be cleaned up.
>>>
>>> Michael McCandless wrote:
>>>>
>>>> ThreadLocal, which we use in several places in Lucene, causes a
>>>> leak in app servers because the classloader never fully deallocates
>>>> Lucene's classes because the ThreadLocal is holding strong references.
>>>>
>>>> Yet, ThreadLocal is very convenient for avoiding synchronization.
>>>>
>>>> Does anyone have any ideas on how to solve this w/o falling back to
>>>> "normal" synchronization?
>>>>
>>>> Mike
>>>>
>>>> Begin forwarded message:
>>>>
>>>>> From: "Yonik Seeley" <[hidden email]>
>>>>> Date: July 7, 2008 3:30:28 PM EDT
>>>>> To: [hidden email]
>>>>> Subject: Re: ThreadLocal in SegmentReader
>>>>> Reply-To: [hidden email]
>>>>>
>>>>> On Mon, Jul 7, 2008 at 2:43 PM, Michael McCandless
>>>>> <[hidden email]> wrote:
>>>>>> So now I'm confused: the SegmentReader itself should no longer be
>>>>>> reachable,
>>>>>> assuming you are not holding any references to your IndexReader.
>>>>>>
>>>>>> Which means the ThreadLocal instance should no longer be reachable.
>>>>>
>>>>> It will still be referenced from the Thread(s) ThreadLocalMap
>>>>> The key (the ThreadLocal) will be weakly referenced, but the values
>>>>> (now stale) are strongly referenced and won't be actually removed
>>>>> until the table is resized (under the Java6 impl at least).
>>>>> Nice huh?
>>>>>
>>>>> -Yonik
>>>>>
>>>>> ---------------------------------------------------------------------
>>>>> To unsubscribe, e-mail: [hidden email]
>>>>> For additional commands, e-mail: [hidden email]
>>>>>
>>>>
>>>>
>>>> ---------------------------------------------------------------------
>>>> To unsubscribe, e-mail: [hidden email]
>>>> For additional commands, e-mail: [hidden email]
>>>>
>>>
>>>
>>> ---------------------------------------------------------------------
>>> To unsubscribe, e-mail: [hidden email]
>>> For additional commands, e-mail: [hidden email]
>>>
>>
>>
>> ---------------------------------------------------------------------
>> To unsubscribe, e-mail: [hidden email]
>> For additional commands, e-mail: [hidden email]
>>
>
>
> ---------------------------------------------------------------------
> To unsubscribe, e-mail: [hidden email]
> For additional commands, e-mail: [hidden email]
>


---------------------------------------------------------------------
To unsubscribe, e-mail: [hidden email]
For additional commands, e-mail: [hidden email]

Reply | Threaded
Open this post in threaded view
|

Re: ThreadLocal in SegmentReader

Robert Engels
Aside from the pre-1.5 thread local "perceived leak", there are no issues with ThreadLocals if used properly.

There is no need for try/finally blocks, unless you MUST release resources immediately, usually this is not the case, which is why a ThreadLocal is used in the first place.

From the ThreadLocalMap javadoc...

 /**
     * ThreadLocalMap is a customized hash map suitable only for
     * maintaining thread local values. No operations are exported
     * outside of the ThreadLocal class. The class is package private to
     * allow declaration of fields in class Thread.  To help deal with
     * very large and long-lived usages, the hash table entries use
     * WeakReferences for keys. However, since reference queues are not
     * used, stale entries are guaranteed to be removed only when
     * the table starts running out of space.
     */

/**
         * Heuristically scan some cells looking for stale entries.
         * This is invoked when either a new element is added, or
         * another stale one has been expunged. It performs a
         * logarithmic number of scans, as a balance between no
         * scanning (fast but retains garbage) and a number of scans
         * proportional to number of elements, that would find all
         * garbage but would cause some insertions to take O(n) time.
         *
         * @param i a position known NOT to hold a stale entry. The
         * scan starts at the element after i.
         *
         * @param n scan control: <tt>log2(n)</tt> cells are scanned,
         * unless a stale entry one is found, in which case
         * <tt>log2(table.length)-1</tt> additional cells are scanned.
         * When called from insertions, this parameter is the number
         * of elements, but when from replaceStaleEntry, it is the
         * table length. (Note: all this could be changed to be either
         * more or less aggressive by weighting n instead of just
         * using straight log n. But this version is simple, fast, and
         * seems to work well.)
         *
         * @return true if any stale entries have been removed.
         */


The instance ThreadLocals (and what the refer to) will be GC'd when the containing Object is GC'd.

There IS NO MEMORY LEAK in ThreadLocal. If the ThreadLocal refers to an object that has native resources (e.g. file handles), it may not be released until other thread locals are created by the thread (or the thread terminates).

You can avoid this "delay" by calling remove(), but in most applications it should never be necessary - unless a very strange usage...

On Jul 9, 2008, at 2:37 PM, Adrian Tarau wrote:

From what I know, storing objects in ThreadLocal is safe as long as you release the object within a try {} finall {} block or store objects which are independent of the rest of the code(no dependencies).Otherwise it can get pretty tricky(memory leaks, classloader problems) after awhile.

It is pretty convenient to pass HTTP request information with a ThreadLocal in a servlet(but you should cleanup the variable before leaving the servlet) but I'm not sure how safe it is in this case.

robert engels wrote:
Using synchronization is a poor/invalid substitute for thread locals in many cases.

The point of the thread local in these referenced cases is too allow streaming reads on a file descriptor. if you use a shared file descriptor/buffer you are going to continually invalidate the buffer.

On Jul 8, 2008, at 5:12 AM, Michael McCandless wrote:


Well ... SegmentReader uses ThreadLocal to hold a thread-private instance of TermVectorsReader, to avoid synchronizing per-document when loading term vectors.

Clearing this ThreadLocal value per call to SegmentReader's methods that load term vectors would defeat its purpose.

Though, of course, we then synchronize on the underlying file (when using FSDirectory), so perhaps we are really not saving much by using ThreadLocal here.  But we are looking to relax that low level synchronization with LUCENE-753.

Maybe we could make our own ThreadLocal that just uses a HashMap, which we'd have to synchronize on when getting the per-thread instances.  Or, go back to sharing a single TermVectorsReader and synchronize per-document.

Jason has suggested moving to a model where you ask the IndexReader for an object that can return term vectors / stored fields / etc, and then you interact with that many times to retrieve each doc.  We could then synchronize only on retrieving that object, and provide a thread-private instance.

It seems like we should move away from using ThreadLocal in Lucene and do "normal" synchronization instead.

Mike

Adrian Tarau wrote:

Usually ThreadLocal.remove() should be called at the end(in a finally block), before the current call leaves your code.

Ex : if during searching ThreadLocal is used, every search(..) method should cleanup any ThreadLocal variables, or even deeper in the implementation. When the call leaves Lucene any used ThreadLocal should be cleaned up.

Michael McCandless wrote:

ThreadLocal, which we use in several places in Lucene, causes a leak in app servers because the classloader never fully deallocates Lucene's classes because the ThreadLocal is holding strong references.

Yet, ThreadLocal is very convenient for avoiding synchronization.

Does anyone have any ideas on how to solve this w/o falling back to "normal" synchronization?

Mike

Begin forwarded message:

From: "Yonik Seeley" <[hidden email]>
Date: July 7, 2008 3:30:28 PM EDT
Subject: Re: ThreadLocal in SegmentReader
Reply-To: [hidden email]

On Mon, Jul 7, 2008 at 2:43 PM, Michael McCandless
So now I'm confused: the SegmentReader itself should no longer be reachable,
assuming you are not holding any references to your IndexReader.

Which means the ThreadLocal instance should no longer be reachable.

It will still be referenced from the Thread(s) ThreadLocalMap
The key (the ThreadLocal) will be weakly referenced, but the values
(now stale) are strongly referenced and won't be actually removed
until the table is resized (under the Java6 impl at least).
Nice huh?

-Yonik

---------------------------------------------------------------------
To unsubscribe, e-mail: [hidden email]
For additional commands, e-mail: [hidden email]



---------------------------------------------------------------------
To unsubscribe, e-mail: [hidden email]
For additional commands, e-mail: [hidden email]



---------------------------------------------------------------------
To unsubscribe, e-mail: [hidden email]
For additional commands, e-mail: [hidden email]



---------------------------------------------------------------------
To unsubscribe, e-mail: [hidden email]
For additional commands, e-mail: [hidden email]



---------------------------------------------------------------------
To unsubscribe, e-mail: [hidden email]
For additional commands, e-mail: [hidden email]



---------------------------------------------------------------------
To unsubscribe, e-mail: [hidden email]
For additional commands, e-mail: [hidden email]


Reply | Threaded
Open this post in threaded view
|

Re: ThreadLocal in SegmentReader

Adrian Tarau
As soon as the object can be collected...which doesn't happen in case of  (some) thread pools.

Here is an example : http://www.opensubscriber.com/message/ojb-user@.../1500135.html (it's pretty old but you can find newer examples googling around). There are some replies with workarounds.

I'm not saying there is something wrong using ThreadLocal in Lucene, but it could cause problems when classloaders have to be released, as Michael noticed.

As Jason suggested, a global manager for ThreadLocals could be the solution.
I prefer to use ThreadLocal to pass "context" objects but I didn't use them to store long-lived objects. I'm always releasing the object after it was used. I'm not sure if is too preventive, but I had some similar problems with web applications(redeploy), ThreadLocals  and long-lived objects.

robert engels wrote:
Aside from the pre-1.5 thread local "perceived leak", there are no issues with ThreadLocals if used properly.

There is no need for try/finally blocks, unless you MUST release resources immediately, usually this is not the case, which is why a ThreadLocal is used in the first place.

From the ThreadLocalMap javadoc...

 /**
     * ThreadLocalMap is a customized hash map suitable only for
     * maintaining thread local values. No operations are exported
     * outside of the ThreadLocal class. The class is package private to
     * allow declaration of fields in class Thread.  To help deal with
     * very large and long-lived usages, the hash table entries use
     * WeakReferences for keys. However, since reference queues are not
     * used, stale entries are guaranteed to be removed only when
     * the table starts running out of space.
     */

/**
         * Heuristically scan some cells looking for stale entries.
         * This is invoked when either a new element is added, or
         * another stale one has been expunged. It performs a
         * logarithmic number of scans, as a balance between no
         * scanning (fast but retains garbage) and a number of scans
         * proportional to number of elements, that would find all
         * garbage but would cause some insertions to take O(n) time.
         *
         * @param i a position known NOT to hold a stale entry. The
         * scan starts at the element after i.
         *
         * @param n scan control: <tt>log2(n)</tt> cells are scanned,
         * unless a stale entry one is found, in which case
         * <tt>log2(table.length)-1</tt> additional cells are scanned.
         * When called from insertions, this parameter is the number
         * of elements, but when from replaceStaleEntry, it is the
         * table length. (Note: all this could be changed to be either
         * more or less aggressive by weighting n instead of just
         * using straight log n. But this version is simple, fast, and
         * seems to work well.)
         *
         * @return true if any stale entries have been removed.
         */


The instance ThreadLocals (and what the refer to) will be GC'd when the containing Object is GC'd.

There IS NO MEMORY LEAK in ThreadLocal. If the ThreadLocal refers to an object that has native resources (e.g. file handles), it may not be released until other thread locals are created by the thread (or the thread terminates).

You can avoid this "delay" by calling remove(), but in most applications it should never be necessary - unless a very strange usage...

On Jul 9, 2008, at 2:37 PM, Adrian Tarau wrote:

From what I know, storing objects in ThreadLocal is safe as long as you release the object within a try {} finall {} block or store objects which are independent of the rest of the code(no dependencies).Otherwise it can get pretty tricky(memory leaks, classloader problems) after awhile.

It is pretty convenient to pass HTTP request information with a ThreadLocal in a servlet(but you should cleanup the variable before leaving the servlet) but I'm not sure how safe it is in this case.

robert engels wrote:
Using synchronization is a poor/invalid substitute for thread locals in many cases.

The point of the thread local in these referenced cases is too allow streaming reads on a file descriptor. if you use a shared file descriptor/buffer you are going to continually invalidate the buffer.

On Jul 8, 2008, at 5:12 AM, Michael McCandless wrote:


Well ... SegmentReader uses ThreadLocal to hold a thread-private instance of TermVectorsReader, to avoid synchronizing per-document when loading term vectors.

Clearing this ThreadLocal value per call to SegmentReader's methods that load term vectors would defeat its purpose.

Though, of course, we then synchronize on the underlying file (when using FSDirectory), so perhaps we are really not saving much by using ThreadLocal here.  But we are looking to relax that low level synchronization with LUCENE-753.

Maybe we could make our own ThreadLocal that just uses a HashMap, which we'd have to synchronize on when getting the per-thread instances.  Or, go back to sharing a single TermVectorsReader and synchronize per-document.

Jason has suggested moving to a model where you ask the IndexReader for an object that can return term vectors / stored fields / etc, and then you interact with that many times to retrieve each doc.  We could then synchronize only on retrieving that object, and provide a thread-private instance.

It seems like we should move away from using ThreadLocal in Lucene and do "normal" synchronization instead.

Mike

Adrian Tarau wrote:

Usually ThreadLocal.remove() should be called at the end(in a finally block), before the current call leaves your code.

Ex : if during searching ThreadLocal is used, every search(..) method should cleanup any ThreadLocal variables, or even deeper in the implementation. When the call leaves Lucene any used ThreadLocal should be cleaned up.

Michael McCandless wrote:

ThreadLocal, which we use in several places in Lucene, causes a leak in app servers because the classloader never fully deallocates Lucene's classes because the ThreadLocal is holding strong references.

Yet, ThreadLocal is very convenient for avoiding synchronization.

Does anyone have any ideas on how to solve this w/o falling back to "normal" synchronization?

Mike

Begin forwarded message:

From: "Yonik Seeley" <[hidden email]>
Date: July 7, 2008 3:30:28 PM EDT
Subject: Re: ThreadLocal in SegmentReader
Reply-To: [hidden email]

On Mon, Jul 7, 2008 at 2:43 PM, Michael McCandless
So now I'm confused: the SegmentReader itself should no longer be reachable,
assuming you are not holding any references to your IndexReader.

Which means the ThreadLocal instance should no longer be reachable.

It will still be referenced from the Thread(s) ThreadLocalMap
The key (the ThreadLocal) will be weakly referenced, but the values
(now stale) are strongly referenced and won't be actually removed
until the table is resized (under the Java6 impl at least).
Nice huh?

-Yonik

---------------------------------------------------------------------
To unsubscribe, e-mail: [hidden email]
For additional commands, e-mail: [hidden email]



---------------------------------------------------------------------
To unsubscribe, e-mail: [hidden email]
For additional commands, e-mail: [hidden email]



---------------------------------------------------------------------
To unsubscribe, e-mail: [hidden email]
For additional commands, e-mail: [hidden email]



---------------------------------------------------------------------
To unsubscribe, e-mail: [hidden email]
For additional commands, e-mail: [hidden email]



---------------------------------------------------------------------
To unsubscribe, e-mail: [hidden email]
For additional commands, e-mail: [hidden email]



---------------------------------------------------------------------
To unsubscribe, e-mail: [hidden email]
For additional commands, e-mail: [hidden email]



Reply | Threaded
Open this post in threaded view
|

Re: ThreadLocal in SegmentReader

Adrian Tarau
In reply to this post by Robert Engels
Just a few examples of "problems" using ThreadLocals.

http://opensource.atlassian.com/projects/hibernate/browse/HHH-2481
http://www.theserverside.com/news/thread.tss?thread_id=41473

Once again, I'm not pointing to Lucene SegmentReader as a "bad" implementation, and maybe the current "problems" of ThreadLocals are not a problem for SegmentReader but it seems safer to use ThreadLocals to pass context information which is cleared when the call exits instead of storing long-lived objects.


robert engels wrote:
Aside from the pre-1.5 thread local "perceived leak", there are no issues with ThreadLocals if used properly.

There is no need for try/finally blocks, unless you MUST release resources immediately, usually this is not the case, which is why a ThreadLocal is used in the first place.

From the ThreadLocalMap javadoc...

 /**
     * ThreadLocalMap is a customized hash map suitable only for
     * maintaining thread local values. No operations are exported
     * outside of the ThreadLocal class. The class is package private to
     * allow declaration of fields in class Thread.  To help deal with
     * very large and long-lived usages, the hash table entries use
     * WeakReferences for keys. However, since reference queues are not
     * used, stale entries are guaranteed to be removed only when
     * the table starts running out of space.
     */

/**
         * Heuristically scan some cells looking for stale entries.
         * This is invoked when either a new element is added, or
         * another stale one has been expunged. It performs a
         * logarithmic number of scans, as a balance between no
         * scanning (fast but retains garbage) and a number of scans
         * proportional to number of elements, that would find all
         * garbage but would cause some insertions to take O(n) time.
         *
         * @param i a position known NOT to hold a stale entry. The
         * scan starts at the element after i.
         *
         * @param n scan control: <tt>log2(n)</tt> cells are scanned,
         * unless a stale entry one is found, in which case
         * <tt>log2(table.length)-1</tt> additional cells are scanned.
         * When called from insertions, this parameter is the number
         * of elements, but when from replaceStaleEntry, it is the
         * table length. (Note: all this could be changed to be either
         * more or less aggressive by weighting n instead of just
         * using straight log n. But this version is simple, fast, and
         * seems to work well.)
         *
         * @return true if any stale entries have been removed.
         */


The instance ThreadLocals (and what the refer to) will be GC'd when the containing Object is GC'd.

There IS NO MEMORY LEAK in ThreadLocal. If the ThreadLocal refers to an object that has native resources (e.g. file handles), it may not be released until other thread locals are created by the thread (or the thread terminates).

You can avoid this "delay" by calling remove(), but in most applications it should never be necessary - unless a very strange usage...

On Jul 9, 2008, at 2:37 PM, Adrian Tarau wrote:

From what I know, storing objects in ThreadLocal is safe as long as you release the object within a try {} finall {} block or store objects which are independent of the rest of the code(no dependencies).Otherwise it can get pretty tricky(memory leaks, classloader problems) after awhile.

It is pretty convenient to pass HTTP request information with a ThreadLocal in a servlet(but you should cleanup the variable before leaving the servlet) but I'm not sure how safe it is in this case.

robert engels wrote:
Using synchronization is a poor/invalid substitute for thread locals in many cases.

The point of the thread local in these referenced cases is too allow streaming reads on a file descriptor. if you use a shared file descriptor/buffer you are going to continually invalidate the buffer.

On Jul 8, 2008, at 5:12 AM, Michael McCandless wrote:


Well ... SegmentReader uses ThreadLocal to hold a thread-private instance of TermVectorsReader, to avoid synchronizing per-document when loading term vectors.

Clearing this ThreadLocal value per call to SegmentReader's methods that load term vectors would defeat its purpose.

Though, of course, we then synchronize on the underlying file (when using FSDirectory), so perhaps we are really not saving much by using ThreadLocal here.  But we are looking to relax that low level synchronization with LUCENE-753.

Maybe we could make our own ThreadLocal that just uses a HashMap, which we'd have to synchronize on when getting the per-thread instances.  Or, go back to sharing a single TermVectorsReader and synchronize per-document.

Jason has suggested moving to a model where you ask the IndexReader for an object that can return term vectors / stored fields / etc, and then you interact with that many times to retrieve each doc.  We could then synchronize only on retrieving that object, and provide a thread-private instance.

It seems like we should move away from using ThreadLocal in Lucene and do "normal" synchronization instead.

Mike

Adrian Tarau wrote:

Usually ThreadLocal.remove() should be called at the end(in a finally block), before the current call leaves your code.

Ex : if during searching ThreadLocal is used, every search(..) method should cleanup any ThreadLocal variables, or even deeper in the implementation. When the call leaves Lucene any used ThreadLocal should be cleaned up.

Michael McCandless wrote:

ThreadLocal, which we use in several places in Lucene, causes a leak in app servers because the classloader never fully deallocates Lucene's classes because the ThreadLocal is holding strong references.

Yet, ThreadLocal is very convenient for avoiding synchronization.

Does anyone have any ideas on how to solve this w/o falling back to "normal" synchronization?

Mike

Begin forwarded message:

From: "Yonik Seeley" <[hidden email]>
Date: July 7, 2008 3:30:28 PM EDT
Subject: Re: ThreadLocal in SegmentReader
Reply-To: [hidden email]

On Mon, Jul 7, 2008 at 2:43 PM, Michael McCandless
So now I'm confused: the SegmentReader itself should no longer be reachable,
assuming you are not holding any references to your IndexReader.

Which means the ThreadLocal instance should no longer be reachable.

It will still be referenced from the Thread(s) ThreadLocalMap
The key (the ThreadLocal) will be weakly referenced, but the values
(now stale) are strongly referenced and won't be actually removed
until the table is resized (under the Java6 impl at least).
Nice huh?

-Yonik

---------------------------------------------------------------------
To unsubscribe, e-mail: [hidden email]
For additional commands, e-mail: [hidden email]



---------------------------------------------------------------------
To unsubscribe, e-mail: [hidden email]
For additional commands, e-mail: [hidden email]



---------------------------------------------------------------------
To unsubscribe, e-mail: [hidden email]
For additional commands, e-mail: [hidden email]



---------------------------------------------------------------------
To unsubscribe, e-mail: [hidden email]
For additional commands, e-mail: [hidden email]



---------------------------------------------------------------------
To unsubscribe, e-mail: [hidden email]
For additional commands, e-mail: [hidden email]



---------------------------------------------------------------------
To unsubscribe, e-mail: [hidden email]
For additional commands, e-mail: [hidden email]



Reply | Threaded
Open this post in threaded view
|

Re: ThreadLocal in SegmentReader

Robert Engels
In reply to this post by Adrian Tarau
As stated in the cited email, they use a static ThreadLocal - this is a different issue (and I don't think Lucene uses any of these).

It is still not a bug in this case, but just the same "delayed" release behavior.  The problem is that the some class loaders probably are never released as the Threads are never terminated, yet they do not create any more ThreadLocals (thus causing some stale slots to never be cleared).

I am not sure that Lucene has many long-lived objects... maybe SegmentReaders if the index rarely changes/updated.

We have done extensive memory profiling of Lucene - there are no inherent leaks (as of the 1.9 code set). I don't think anything has changed since then related to this (expect using some FixedThreadLocal to improve the situation in some cases).

If you search the archives for FixedThreadLocal you can find a simple implementation that solves the perceived problem.


On Jul 9, 2008, at 3:59 PM, Adrian Tarau wrote:

As soon as the object can be collected...which doesn't happen in case of  (some) thread pools.

Here is an example : http://www.opensubscriber.com/message/ojb-user@.../1500135.html (it's pretty old but you can find newer examples googling around). There are some replies with workarounds.

I'm not saying there is something wrong using ThreadLocal in Lucene, but it could cause problems when classloaders have to be released, as Michael noticed.

As Jason suggested, a global manager for ThreadLocals could be the solution.
I prefer to use ThreadLocal to pass "context" objects but I didn't use them to store long-lived objects. I'm always releasing the object after it was used. I'm not sure if is too preventive, but I had some similar problems with web applications(redeploy), ThreadLocals  and long-lived objects.

robert engels wrote:
Aside from the pre-1.5 thread local "perceived leak", there are no issues with ThreadLocals if used properly.

There is no need for try/finally blocks, unless you MUST release resources immediately, usually this is not the case, which is why a ThreadLocal is used in the first place.

From the ThreadLocalMap javadoc...

 /**
     * ThreadLocalMap is a customized hash map suitable only for
     * maintaining thread local values. No operations are exported
     * outside of the ThreadLocal class. The class is package private to
     * allow declaration of fields in class Thread.  To help deal with
     * very large and long-lived usages, the hash table entries use
     * WeakReferences for keys. However, since reference queues are not
     * used, stale entries are guaranteed to be removed only when
     * the table starts running out of space.
     */

/**
         * Heuristically scan some cells looking for stale entries.
         * This is invoked when either a new element is added, or
         * another stale one has been expunged. It performs a
         * logarithmic number of scans, as a balance between no
         * scanning (fast but retains garbage) and a number of scans
         * proportional to number of elements, that would find all
         * garbage but would cause some insertions to take O(n) time.
         *
         * @param i a position known NOT to hold a stale entry. The
         * scan starts at the element after i.
         *
         * @param n scan control: <tt>log2(n)</tt> cells are scanned,
         * unless a stale entry one is found, in which case
         * <tt>log2(table.length)-1</tt> additional cells are scanned.
         * When called from insertions, this parameter is the number
         * of elements, but when from replaceStaleEntry, it is the
         * table length. (Note: all this could be changed to be either
         * more or less aggressive by weighting n instead of just
         * using straight log n. But this version is simple, fast, and
         * seems to work well.)
         *
         * @return true if any stale entries have been removed.
         */


The instance ThreadLocals (and what the refer to) will be GC'd when the containing Object is GC'd.

There IS NO MEMORY LEAK in ThreadLocal. If the ThreadLocal refers to an object that has native resources (e.g. file handles), it may not be released until other thread locals are created by the thread (or the thread terminates).

You can avoid this "delay" by calling remove(), but in most applications it should never be necessary - unless a very strange usage...

On Jul 9, 2008, at 2:37 PM, Adrian Tarau wrote:

From what I know, storing objects in ThreadLocal is safe as long as you release the object within a try {} finall {} block or store objects which are independent of the rest of the code(no dependencies).Otherwise it can get pretty tricky(memory leaks, classloader problems) after awhile.

It is pretty convenient to pass HTTP request information with a ThreadLocal in a servlet(but you should cleanup the variable before leaving the servlet) but I'm not sure how safe it is in this case.

robert engels wrote:
Using synchronization is a poor/invalid substitute for thread locals in many cases.

The point of the thread local in these referenced cases is too allow streaming reads on a file descriptor. if you use a shared file descriptor/buffer you are going to continually invalidate the buffer.

On Jul 8, 2008, at 5:12 AM, Michael McCandless wrote:


Well ... SegmentReader uses ThreadLocal to hold a thread-private instance of TermVectorsReader, to avoid synchronizing per-document when loading term vectors.

Clearing this ThreadLocal value per call to SegmentReader's methods that load term vectors would defeat its purpose.

Though, of course, we then synchronize on the underlying file (when using FSDirectory), so perhaps we are really not saving much by using ThreadLocal here.  But we are looking to relax that low level synchronization with LUCENE-753.

Maybe we could make our own ThreadLocal that just uses a HashMap, which we'd have to synchronize on when getting the per-thread instances.  Or, go back to sharing a single TermVectorsReader and synchronize per-document.

Jason has suggested moving to a model where you ask the IndexReader for an object that can return term vectors / stored fields / etc, and then you interact with that many times to retrieve each doc.  We could then synchronize only on retrieving that object, and provide a thread-private instance.

It seems like we should move away from using ThreadLocal in Lucene and do "normal" synchronization instead.

Mike

Adrian Tarau wrote:

Usually ThreadLocal.remove() should be called at the end(in a finally block), before the current call leaves your code.

Ex : if during searching ThreadLocal is used, every search(..) method should cleanup any ThreadLocal variables, or even deeper in the implementation. When the call leaves Lucene any used ThreadLocal should be cleaned up.

Michael McCandless wrote:

ThreadLocal, which we use in several places in Lucene, causes a leak in app servers because the classloader never fully deallocates Lucene's classes because the ThreadLocal is holding strong references.

Yet, ThreadLocal is very convenient for avoiding synchronization.

Does anyone have any ideas on how to solve this w/o falling back to "normal" synchronization?

Mike

Begin forwarded message:

From: "Yonik Seeley" <[hidden email]>
Date: July 7, 2008 3:30:28 PM EDT
Subject: Re: ThreadLocal in SegmentReader
Reply-To: [hidden email]

On Mon, Jul 7, 2008 at 2:43 PM, Michael McCandless
So now I'm confused: the SegmentReader itself should no longer be reachable,
assuming you are not holding any references to your IndexReader.

Which means the ThreadLocal instance should no longer be reachable.

It will still be referenced from the Thread(s) ThreadLocalMap
The key (the ThreadLocal) will be weakly referenced, but the values
(now stale) are strongly referenced and won't be actually removed
until the table is resized (under the Java6 impl at least).
Nice huh?

-Yonik

---------------------------------------------------------------------
To unsubscribe, e-mail: [hidden email]
For additional commands, e-mail: [hidden email]



---------------------------------------------------------------------
To unsubscribe, e-mail: [hidden email]
For additional commands, e-mail: [hidden email]



---------------------------------------------------------------------
To unsubscribe, e-mail: [hidden email]
For additional commands, e-mail: [hidden email]



---------------------------------------------------------------------
To unsubscribe, e-mail: [hidden email]
For additional commands, e-mail: [hidden email]



---------------------------------------------------------------------
To unsubscribe, e-mail: [hidden email]
For additional commands, e-mail: [hidden email]



---------------------------------------------------------------------
To unsubscribe, e-mail: [hidden email]
For additional commands, e-mail: [hidden email]




Reply | Threaded
Open this post in threaded view
|

Re: ThreadLocal in SegmentReader

Robert Engels
In reply to this post by Adrian Tarau
Once again, these are "static" thread locals. A completely different issue. Since the object is available statically, the weak reference cannot be cleared so stale entries will never be cleared as long as the thread is alive.

On Jul 9, 2008, at 4:46 PM, Adrian Tarau wrote:

Just a few examples of "problems" using ThreadLocals.

http://opensource.atlassian.com/projects/hibernate/browse/HHH-2481
http://www.theserverside.com/news/thread.tss?thread_id=41473

Once again, I'm not pointing to Lucene SegmentReader as a "bad" implementation, and maybe the current "problems" of ThreadLocals are not a problem for SegmentReader but it seems safer to use ThreadLocals to pass context information which is cleared when the call exits instead of storing long-lived objects.


robert engels wrote:
Aside from the pre-1.5 thread local "perceived leak", there are no issues with ThreadLocals if used properly.

There is no need for try/finally blocks, unless you MUST release resources immediately, usually this is not the case, which is why a ThreadLocal is used in the first place.

From the ThreadLocalMap javadoc...

 /**
     * ThreadLocalMap is a customized hash map suitable only for
     * maintaining thread local values. No operations are exported
     * outside of the ThreadLocal class. The class is package private to
     * allow declaration of fields in class Thread.  To help deal with
     * very large and long-lived usages, the hash table entries use
     * WeakReferences for keys. However, since reference queues are not
     * used, stale entries are guaranteed to be removed only when
     * the table starts running out of space.
     */

/**
         * Heuristically scan some cells looking for stale entries.
         * This is invoked when either a new element is added, or
         * another stale one has been expunged. It performs a
         * logarithmic number of scans, as a balance between no
         * scanning (fast but retains garbage) and a number of scans
         * proportional to number of elements, that would find all
         * garbage but would cause some insertions to take O(n) time.
         *
         * @param i a position known NOT to hold a stale entry. The
         * scan starts at the element after i.
         *
         * @param n scan control: <tt>log2(n)</tt> cells are scanned,
         * unless a stale entry one is found, in which case
         * <tt>log2(table.length)-1</tt> additional cells are scanned.
         * When called from insertions, this parameter is the number
         * of elements, but when from replaceStaleEntry, it is the
         * table length. (Note: all this could be changed to be either
         * more or less aggressive by weighting n instead of just
         * using straight log n. But this version is simple, fast, and
         * seems to work well.)
         *
         * @return true if any stale entries have been removed.
         */


The instance ThreadLocals (and what the refer to) will be GC'd when the containing Object is GC'd.

There IS NO MEMORY LEAK in ThreadLocal. If the ThreadLocal refers to an object that has native resources (e.g. file handles), it may not be released until other thread locals are created by the thread (or the thread terminates).

You can avoid this "delay" by calling remove(), but in most applications it should never be necessary - unless a very strange usage...

On Jul 9, 2008, at 2:37 PM, Adrian Tarau wrote:

From what I know, storing objects in ThreadLocal is safe as long as you release the object within a try {} finall {} block or store objects which are independent of the rest of the code(no dependencies).Otherwise it can get pretty tricky(memory leaks, classloader problems) after awhile.

It is pretty convenient to pass HTTP request information with a ThreadLocal in a servlet(but you should cleanup the variable before leaving the servlet) but I'm not sure how safe it is in this case.

robert engels wrote:
Using synchronization is a poor/invalid substitute for thread locals in many cases.

The point of the thread local in these referenced cases is too allow streaming reads on a file descriptor. if you use a shared file descriptor/buffer you are going to continually invalidate the buffer.

On Jul 8, 2008, at 5:12 AM, Michael McCandless wrote:


Well ... SegmentReader uses ThreadLocal to hold a thread-private instance of TermVectorsReader, to avoid synchronizing per-document when loading term vectors.

Clearing this ThreadLocal value per call to SegmentReader's methods that load term vectors would defeat its purpose.

Though, of course, we then synchronize on the underlying file (when using FSDirectory), so perhaps we are really not saving much by using ThreadLocal here.  But we are looking to relax that low level synchronization with LUCENE-753.

Maybe we could make our own ThreadLocal that just uses a HashMap, which we'd have to synchronize on when getting the per-thread instances.  Or, go back to sharing a single TermVectorsReader and synchronize per-document.

Jason has suggested moving to a model where you ask the IndexReader for an object that can return term vectors / stored fields / etc, and then you interact with that many times to retrieve each doc.  We could then synchronize only on retrieving that object, and provide a thread-private instance.

It seems like we should move away from using ThreadLocal in Lucene and do "normal" synchronization instead.

Mike

Adrian Tarau wrote:

Usually ThreadLocal.remove() should be called at the end(in a finally block), before the current call leaves your code.

Ex : if during searching ThreadLocal is used, every search(..) method should cleanup any ThreadLocal variables, or even deeper in the implementation. When the call leaves Lucene any used ThreadLocal should be cleaned up.

Michael McCandless wrote:

ThreadLocal, which we use in several places in Lucene, causes a leak in app servers because the classloader never fully deallocates Lucene's classes because the ThreadLocal is holding strong references.

Yet, ThreadLocal is very convenient for avoiding synchronization.

Does anyone have any ideas on how to solve this w/o falling back to "normal" synchronization?

Mike

Begin forwarded message:

From: "Yonik Seeley" <[hidden email]>
Date: July 7, 2008 3:30:28 PM EDT
Subject: Re: ThreadLocal in SegmentReader
Reply-To: [hidden email]

On Mon, Jul 7, 2008 at 2:43 PM, Michael McCandless
So now I'm confused: the SegmentReader itself should no longer be reachable,
assuming you are not holding any references to your IndexReader.

Which means the ThreadLocal instance should no longer be reachable.

It will still be referenced from the Thread(s) ThreadLocalMap
The key (the ThreadLocal) will be weakly referenced, but the values
(now stale) are strongly referenced and won't be actually removed
until the table is resized (under the Java6 impl at least).
Nice huh?

-Yonik

---------------------------------------------------------------------
To unsubscribe, e-mail: [hidden email]
For additional commands, e-mail: [hidden email]



---------------------------------------------------------------------
To unsubscribe, e-mail: [hidden email]
For additional commands, e-mail: [hidden email]



---------------------------------------------------------------------
To unsubscribe, e-mail: [hidden email]
For additional commands, e-mail: [hidden email]



---------------------------------------------------------------------
To unsubscribe, e-mail: [hidden email]
For additional commands, e-mail: [hidden email]



---------------------------------------------------------------------
To unsubscribe, e-mail: [hidden email]
For additional commands, e-mail: [hidden email]



---------------------------------------------------------------------
To unsubscribe, e-mail: [hidden email]
For additional commands, e-mail: [hidden email]




Reply | Threaded
Open this post in threaded view
|

Re: ThreadLocal in SegmentReader

Michael McCandless-2
OK, I created a simple test to test this (attached).  The test just  
runs 10 threads, each one creating a 100 KB byte array which is stored  
into a ThreadLocal, and then periodically the ThreadLocal is replaced  
with a new one.  This is to test whether GC of a ThreadLocal, even  
though the thread is still alive, in fact leads to GC of the objects  
held in the ThreadLocal.

Indeed on Sun JRE 1.4, 1.5 and 1.6 it appears that the objects are in  
fact properly collected.

So this is not a leak but rather a "delayed collection" issue.  Java's  
GC is never guaranteed to be immediate, and apparently when using  
ThreadLocals it's even less immediate than "normal".  In the original  
issue, if other things create ThreadLocals, then eventually Lucene's  
unreferenced ThreadLocals would be properly collected.

So I think we continue to use non-static ThreadLocals in Lucene...

Mike





robert engels wrote:

> Once again, these are "static" thread locals. A completely different  
> issue. Since the object is available statically, the weak reference  
> cannot be cleared so stale entries will never be cleared as long as  
> the thread is alive.
>
> On Jul 9, 2008, at 4:46 PM, Adrian Tarau wrote:
>
>> Just a few examples of "problems" using ThreadLocals.
>>
>> http://opensource.atlassian.com/projects/hibernate/browse/HHH-2481
>> http://www.theserverside.com/news/thread.tss?thread_id=41473
>>
>> Once again, I'm not pointing to Lucene SegmentReader as a "bad"  
>> implementation, and maybe the current "problems" of ThreadLocals  
>> are not a problem for SegmentReader but it seems safer to use  
>> ThreadLocals to pass context information which is cleared when the  
>> call exits instead of storing long-lived objects.
>>
>>
>> robert engels wrote:
>>>
>>> Aside from the pre-1.5 thread local "perceived leak", there are no  
>>> issues with ThreadLocals if used properly.
>>>
>>> There is no need for try/finally blocks, unless you MUST release  
>>> resources immediately, usually this is not the case, which is why  
>>> a ThreadLocal is used in the first place.
>>>
>>> From the ThreadLocalMap javadoc...
>>>
>>>  /**
>>>      * ThreadLocalMap is a customized hash map suitable only for
>>>      * maintaining thread local values. No operations are exported
>>>      * outside of the ThreadLocal class. The class is package  
>>> private to
>>>      * allow declaration of fields in class Thread.  To help deal  
>>> with
>>>      * very large and long-lived usages, the hash table entries use
>>>      * WeakReferences for keys. However, since reference queues  
>>> are not
>>>      * used, stale entries are guaranteed to be removed only when
>>>      * the table starts running out of space.
>>>      */
>>>
>>> /**
>>>          * Heuristically scan some cells looking for stale entries.
>>>          * This is invoked when either a new element is added, or
>>>          * another stale one has been expunged. It performs a
>>>          * logarithmic number of scans, as a balance between no
>>>          * scanning (fast but retains garbage) and a number of scans
>>>          * proportional to number of elements, that would find all
>>>          * garbage but would cause some insertions to take O(n)  
>>> time.
>>>          *
>>>          * @param i a position known NOT to hold a stale entry. The
>>>          * scan starts at the element after i.
>>>          *
>>>          * @param n scan control: <tt>log2(n)</tt> cells are  
>>> scanned,
>>>          * unless a stale entry one is found, in which case
>>>          * <tt>log2(table.length)-1</tt> additional cells are  
>>> scanned.
>>>          * When called from insertions, this parameter is the number
>>>          * of elements, but when from replaceStaleEntry, it is the
>>>          * table length. (Note: all this could be changed to be  
>>> either
>>>          * more or less aggressive by weighting n instead of just
>>>          * using straight log n. But this version is simple, fast,  
>>> and
>>>          * seems to work well.)
>>>          *
>>>          * @return true if any stale entries have been removed.
>>>          */
>>>
>>>
>>> The instance ThreadLocals (and what the refer to) will be GC'd  
>>> when the containing Object is GC'd.
>>>
>>> There IS NO MEMORY LEAK in ThreadLocal. If the ThreadLocal refers  
>>> to an object that has native resources (e.g. file handles), it may  
>>> not be released until other thread locals are created by the  
>>> thread (or the thread terminates).
>>>
>>> You can avoid this "delay" by calling remove(), but in most  
>>> applications it should never be necessary - unless a very strange  
>>> usage...
>>>
>>> On Jul 9, 2008, at 2:37 PM, Adrian Tarau wrote:
>>>
>>>> From what I know, storing objects in ThreadLocal is safe as long  
>>>> as you release the object within a try {} finall {} block or  
>>>> store objects which are independent of the rest of the code(no  
>>>> dependencies).Otherwise it can get pretty tricky(memory leaks,  
>>>> classloader problems) after awhile.
>>>>
>>>> It is pretty convenient to pass HTTP request information with a  
>>>> ThreadLocal in a servlet(but you should cleanup the variable  
>>>> before leaving the servlet) but I'm not sure how safe it is in  
>>>> this case.
>>>>
>>>> robert engels wrote:
>>>>> Using synchronization is a poor/invalid substitute for thread  
>>>>> locals in many cases.
>>>>>
>>>>> The point of the thread local in these referenced cases is too  
>>>>> allow streaming reads on a file descriptor. if you use a shared  
>>>>> file descriptor/buffer you are going to continually invalidate  
>>>>> the buffer.
>>>>>
>>>>> On Jul 8, 2008, at 5:12 AM, Michael McCandless wrote:
>>>>>
>>>>>>
>>>>>> Well ... SegmentReader uses ThreadLocal to hold a thread-
>>>>>> private instance of TermVectorsReader, to avoid synchronizing  
>>>>>> per-document when loading term vectors.
>>>>>>
>>>>>> Clearing this ThreadLocal value per call to SegmentReader's  
>>>>>> methods that load term vectors would defeat its purpose.
>>>>>>
>>>>>> Though, of course, we then synchronize on the underlying file  
>>>>>> (when using FSDirectory), so perhaps we are really not saving  
>>>>>> much by using ThreadLocal here.  But we are looking to relax  
>>>>>> that low level synchronization with LUCENE-753.
>>>>>>
>>>>>> Maybe we could make our own ThreadLocal that just uses a  
>>>>>> HashMap, which we'd have to synchronize on when getting the per-
>>>>>> thread instances.  Or, go back to sharing a single  
>>>>>> TermVectorsReader and synchronize per-document.
>>>>>>
>>>>>> Jason has suggested moving to a model where you ask the  
>>>>>> IndexReader for an object that can return term vectors / stored  
>>>>>> fields / etc, and then you interact with that many times to  
>>>>>> retrieve each doc.  We could then synchronize only on  
>>>>>> retrieving that object, and provide a thread-private instance.
>>>>>>
>>>>>> It seems like we should move away from using ThreadLocal in  
>>>>>> Lucene and do "normal" synchronization instead.
>>>>>>
>>>>>> Mike
>>>>>>
>>>>>> Adrian Tarau wrote:
>>>>>>
>>>>>>> Usually ThreadLocal.remove() should be called at the end(in a  
>>>>>>> finally block), before the current call leaves your code.
>>>>>>>
>>>>>>> Ex : if during searching ThreadLocal is used, every search(..)  
>>>>>>> method should cleanup any ThreadLocal variables, or even  
>>>>>>> deeper in the implementation. When the call leaves Lucene any  
>>>>>>> used ThreadLocal should be cleaned up.
>>>>>>>
>>>>>>> Michael McCandless wrote:
>>>>>>>>
>>>>>>>> ThreadLocal, which we use in several places in Lucene, causes  
>>>>>>>> a leak in app servers because the classloader never fully  
>>>>>>>> deallocates Lucene's classes because the ThreadLocal is  
>>>>>>>> holding strong references.
>>>>>>>>
>>>>>>>> Yet, ThreadLocal is very convenient for avoiding  
>>>>>>>> synchronization.
>>>>>>>>
>>>>>>>> Does anyone have any ideas on how to solve this w/o falling  
>>>>>>>> back to "normal" synchronization?
>>>>>>>>
>>>>>>>> Mike
>>>>>>>>
>>>>>>>> Begin forwarded message:
>>>>>>>>
>>>>>>>>> From: "Yonik Seeley" <[hidden email]>
>>>>>>>>> Date: July 7, 2008 3:30:28 PM EDT
>>>>>>>>> To: [hidden email]
>>>>>>>>> Subject: Re: ThreadLocal in SegmentReader
>>>>>>>>> Reply-To: [hidden email]
>>>>>>>>>
>>>>>>>>> On Mon, Jul 7, 2008 at 2:43 PM, Michael McCandless
>>>>>>>>> <[hidden email]> wrote:
>>>>>>>>>> So now I'm confused: the SegmentReader itself should no  
>>>>>>>>>> longer be reachable,
>>>>>>>>>> assuming you are not holding any references to your  
>>>>>>>>>> IndexReader.
>>>>>>>>>>
>>>>>>>>>> Which means the ThreadLocal instance should no longer be  
>>>>>>>>>> reachable.
>>>>>>>>>
>>>>>>>>> It will still be referenced from the Thread(s) ThreadLocalMap
>>>>>>>>> The key (the ThreadLocal) will be weakly referenced, but the  
>>>>>>>>> values
>>>>>>>>> (now stale) are strongly referenced and won't be actually  
>>>>>>>>> removed
>>>>>>>>> until the table is resized (under the Java6 impl at least).
>>>>>>>>> Nice huh?
>>>>>>>>>
>>>>>>>>> -Yonik
>>>>>>>>>
>>>>>>>>> ---------------------------------------------------------------------
>>>>>>>>> To unsubscribe, e-mail: java-user-
>>>>>>>>> [hidden email]
>>>>>>>>> For additional commands, e-mail: [hidden email]
>>>>>>>>>
>>>>>>>>
>>>>>>>>
>>>>>>>> ---------------------------------------------------------------------
>>>>>>>> To unsubscribe, e-mail: [hidden email]
>>>>>>>> For additional commands, e-mail: java-dev-
>>>>>>>> [hidden email]
>>>>>>>>
>>>>>>>
>>>>>>>
>>>>>>> ---------------------------------------------------------------------
>>>>>>> To unsubscribe, e-mail: [hidden email]
>>>>>>> For additional commands, e-mail: [hidden email]
>>>>>>>
>>>>>>
>>>>>>
>>>>>> ---------------------------------------------------------------------
>>>>>> To unsubscribe, e-mail: [hidden email]
>>>>>> For additional commands, e-mail: [hidden email]
>>>>>>
>>>>>
>>>>>
>>>>> ---------------------------------------------------------------------
>>>>> To unsubscribe, e-mail: [hidden email]
>>>>> For additional commands, e-mail: [hidden email]
>>>>>
>>>>
>>>>
>>>> ---------------------------------------------------------------------
>>>> To unsubscribe, e-mail: [hidden email]
>>>> For additional commands, e-mail: [hidden email]
>>>>
>>>
>>
>


---------------------------------------------------------------------
To unsubscribe, e-mail: [hidden email]
For additional commands, e-mail: [hidden email]

ThreadTest.java (1K) Download Attachment
Reply | Threaded
Open this post in threaded view
|

Re: ThreadLocal in SegmentReader

Robert Engels
As always, you still have the issue that if the object in the  
ThreadLocal has a reference to a native resource (e.g. file handle),  
you might run out of file handles before any OOM which triggers the  
GC (to close the file handle if relying on finalization).

On Jul 11, 2008, at 4:54 AM, Michael McCandless wrote:

> OK, I created a simple test to test this (attached).  The test just  
> runs 10 threads, each one creating a 100 KB byte array which is  
> stored into a ThreadLocal, and then periodically the ThreadLocal is  
> replaced with a new one.  This is to test whether GC of a  
> ThreadLocal, even though the thread is still alive, in fact leads  
> to GC of the objects held in the ThreadLocal.
>
> Indeed on Sun JRE 1.4, 1.5 and 1.6 it appears that the objects are  
> in fact properly collected.
>
> So this is not a leak but rather a "delayed collection" issue.  
> Java's GC is never guaranteed to be immediate, and apparently when  
> using ThreadLocals it's even less immediate than "normal".  In the  
> original issue, if other things create ThreadLocals, then  
> eventually Lucene's unreferenced ThreadLocals would be properly  
> collected.
>
> So I think we continue to use non-static ThreadLocals in Lucene...
>
> Mike
>
> <ThreadTest.java>
>
>
> robert engels wrote:
>
>> Once again, these are "static" thread locals. A completely  
>> different issue. Since the object is available statically, the  
>> weak reference cannot be cleared so stale entries will never be  
>> cleared as long as the thread is alive.
>>
>> On Jul 9, 2008, at 4:46 PM, Adrian Tarau wrote:
>>
>>> Just a few examples of "problems" using ThreadLocals.
>>>
>>> http://opensource.atlassian.com/projects/hibernate/browse/HHH-2481
>>> http://www.theserverside.com/news/thread.tss?thread_id=41473
>>>
>>> Once again, I'm not pointing to Lucene SegmentReader as a "bad"  
>>> implementation, and maybe the current "problems" of ThreadLocals  
>>> are not a problem for SegmentReader but it seems safer to use  
>>> ThreadLocals to pass context information which is cleared when  
>>> the call exits instead of storing long-lived objects.
>>>
>>>
>>> robert engels wrote:
>>>>
>>>> Aside from the pre-1.5 thread local "perceived leak", there are  
>>>> no issues with ThreadLocals if used properly.
>>>>
>>>> There is no need for try/finally blocks, unless you MUST release  
>>>> resources immediately, usually this is not the case, which is  
>>>> why a ThreadLocal is used in the first place.
>>>>
>>>> From the ThreadLocalMap javadoc...
>>>>
>>>>  /**
>>>>      * ThreadLocalMap is a customized hash map suitable only for
>>>>      * maintaining thread local values. No operations are exported
>>>>      * outside of the ThreadLocal class. The class is package  
>>>> private to
>>>>      * allow declaration of fields in class Thread.  To help  
>>>> deal with
>>>>      * very large and long-lived usages, the hash table entries use
>>>>      * WeakReferences for keys. However, since reference queues  
>>>> are not
>>>>      * used, stale entries are guaranteed to be removed only when
>>>>      * the table starts running out of space.
>>>>      */
>>>>
>>>> /**
>>>>          * Heuristically scan some cells looking for stale entries.
>>>>          * This is invoked when either a new element is added, or
>>>>          * another stale one has been expunged. It performs a
>>>>          * logarithmic number of scans, as a balance between no
>>>>          * scanning (fast but retains garbage) and a number of  
>>>> scans
>>>>          * proportional to number of elements, that would find all
>>>>          * garbage but would cause some insertions to take O(n)  
>>>> time.
>>>>          *
>>>>          * @param i a position known NOT to hold a stale entry. The
>>>>          * scan starts at the element after i.
>>>>          *
>>>>          * @param n scan control: <tt>log2(n)</tt> cells are  
>>>> scanned,
>>>>          * unless a stale entry one is found, in which case
>>>>          * <tt>log2(table.length)-1</tt> additional cells are  
>>>> scanned.
>>>>          * When called from insertions, this parameter is the  
>>>> number
>>>>          * of elements, but when from replaceStaleEntry, it is the
>>>>          * table length. (Note: all this could be changed to be  
>>>> either
>>>>          * more or less aggressive by weighting n instead of just
>>>>          * using straight log n. But this version is simple,  
>>>> fast, and
>>>>          * seems to work well.)
>>>>          *
>>>>          * @return true if any stale entries have been removed.
>>>>          */
>>>>
>>>>
>>>> The instance ThreadLocals (and what the refer to) will be GC'd  
>>>> when the containing Object is GC'd.
>>>>
>>>> There IS NO MEMORY LEAK in ThreadLocal. If the ThreadLocal  
>>>> refers to an object that has native resources (e.g. file  
>>>> handles), it may not be released until other thread locals are  
>>>> created by the thread (or the thread terminates).
>>>>
>>>> You can avoid this "delay" by calling remove(), but in most  
>>>> applications it should never be necessary - unless a very  
>>>> strange usage...
>>>>
>>>> On Jul 9, 2008, at 2:37 PM, Adrian Tarau wrote:
>>>>
>>>>> From what I know, storing objects in ThreadLocal is safe as  
>>>>> long as you release the object within a try {} finall {} block  
>>>>> or store objects which are independent of the rest of the code
>>>>> (no dependencies).Otherwise it can get pretty tricky(memory  
>>>>> leaks, classloader problems) after awhile.
>>>>>
>>>>> It is pretty convenient to pass HTTP request information with a  
>>>>> ThreadLocal in a servlet(but you should cleanup the variable  
>>>>> before leaving the servlet) but I'm not sure how safe it is in  
>>>>> this case.
>>>>>
>>>>> robert engels wrote:
>>>>>> Using synchronization is a poor/invalid substitute for thread  
>>>>>> locals in many cases.
>>>>>>
>>>>>> The point of the thread local in these referenced cases is too  
>>>>>> allow streaming reads on a file descriptor. if you use a  
>>>>>> shared file descriptor/buffer you are going to continually  
>>>>>> invalidate the buffer.
>>>>>>
>>>>>> On Jul 8, 2008, at 5:12 AM, Michael McCandless wrote:
>>>>>>
>>>>>>>
>>>>>>> Well ... SegmentReader uses ThreadLocal to hold a thread-
>>>>>>> private instance of TermVectorsReader, to avoid synchronizing  
>>>>>>> per-document when loading term vectors.
>>>>>>>
>>>>>>> Clearing this ThreadLocal value per call to SegmentReader's  
>>>>>>> methods that load term vectors would defeat its purpose.
>>>>>>>
>>>>>>> Though, of course, we then synchronize on the underlying file  
>>>>>>> (when using FSDirectory), so perhaps we are really not saving  
>>>>>>> much by using ThreadLocal here.  But we are looking to relax  
>>>>>>> that low level synchronization with LUCENE-753.
>>>>>>>
>>>>>>> Maybe we could make our own ThreadLocal that just uses a  
>>>>>>> HashMap, which we'd have to synchronize on when getting the  
>>>>>>> per-thread instances.  Or, go back to sharing a single  
>>>>>>> TermVectorsReader and synchronize per-document.
>>>>>>>
>>>>>>> Jason has suggested moving to a model where you ask the  
>>>>>>> IndexReader for an object that can return term vectors /  
>>>>>>> stored fields / etc, and then you interact with that many  
>>>>>>> times to retrieve each doc.  We could then synchronize only  
>>>>>>> on retrieving that object, and provide a thread-private  
>>>>>>> instance.
>>>>>>>
>>>>>>> It seems like we should move away from using ThreadLocal in  
>>>>>>> Lucene and do "normal" synchronization instead.
>>>>>>>
>>>>>>> Mike
>>>>>>>
>>>>>>> Adrian Tarau wrote:
>>>>>>>
>>>>>>>> Usually ThreadLocal.remove() should be called at the end(in  
>>>>>>>> a finally block), before the current call leaves your code.
>>>>>>>>
>>>>>>>> Ex : if during searching ThreadLocal is used, every search
>>>>>>>> (..) method should cleanup any ThreadLocal variables, or  
>>>>>>>> even deeper in the implementation. When the call leaves  
>>>>>>>> Lucene any used ThreadLocal should be cleaned up.
>>>>>>>>
>>>>>>>> Michael McCandless wrote:
>>>>>>>>>
>>>>>>>>> ThreadLocal, which we use in several places in Lucene,  
>>>>>>>>> causes a leak in app servers because the classloader never  
>>>>>>>>> fully deallocates Lucene's classes because the ThreadLocal  
>>>>>>>>> is holding strong references.
>>>>>>>>>
>>>>>>>>> Yet, ThreadLocal is very convenient for avoiding  
>>>>>>>>> synchronization.
>>>>>>>>>
>>>>>>>>> Does anyone have any ideas on how to solve this w/o falling  
>>>>>>>>> back to "normal" synchronization?
>>>>>>>>>
>>>>>>>>> Mike
>>>>>>>>>
>>>>>>>>> Begin forwarded message:
>>>>>>>>>
>>>>>>>>>> From: "Yonik Seeley" <[hidden email]>
>>>>>>>>>> Date: July 7, 2008 3:30:28 PM EDT
>>>>>>>>>> To: [hidden email]
>>>>>>>>>> Subject: Re: ThreadLocal in SegmentReader
>>>>>>>>>> Reply-To: [hidden email]
>>>>>>>>>>
>>>>>>>>>> On Mon, Jul 7, 2008 at 2:43 PM, Michael McCandless
>>>>>>>>>> <[hidden email]> wrote:
>>>>>>>>>>> So now I'm confused: the SegmentReader itself should no  
>>>>>>>>>>> longer be reachable,
>>>>>>>>>>> assuming you are not holding any references to your  
>>>>>>>>>>> IndexReader.
>>>>>>>>>>>
>>>>>>>>>>> Which means the ThreadLocal instance should no longer be  
>>>>>>>>>>> reachable.
>>>>>>>>>>
>>>>>>>>>> It will still be referenced from the Thread(s) ThreadLocalMap
>>>>>>>>>> The key (the ThreadLocal) will be weakly referenced, but  
>>>>>>>>>> the values
>>>>>>>>>> (now stale) are strongly referenced and won't be actually  
>>>>>>>>>> removed
>>>>>>>>>> until the table is resized (under the Java6 impl at least).
>>>>>>>>>> Nice huh?
>>>>>>>>>>
>>>>>>>>>> -Yonik
>>>>>>>>>>
>>>>>>>>>> -------------------------------------------------------------
>>>>>>>>>> --------
>>>>>>>>>> To unsubscribe, e-mail: java-user-
>>>>>>>>>> [hidden email]
>>>>>>>>>> For additional commands, e-mail: java-user-
>>>>>>>>>> [hidden email]
>>>>>>>>>>
>>>>>>>>>
>>>>>>>>>
>>>>>>>>> --------------------------------------------------------------
>>>>>>>>> -------
>>>>>>>>> To unsubscribe, e-mail: [hidden email]
>>>>>>>>> For additional commands, e-mail: java-dev-
>>>>>>>>> [hidden email]
>>>>>>>>>
>>>>>>>>
>>>>>>>>
>>>>>>>> ---------------------------------------------------------------
>>>>>>>> ------
>>>>>>>> To unsubscribe, e-mail: [hidden email]
>>>>>>>> For additional commands, e-mail: java-dev-
>>>>>>>> [hidden email]
>>>>>>>>
>>>>>>>
>>>>>>>
>>>>>>> ----------------------------------------------------------------
>>>>>>> -----
>>>>>>> To unsubscribe, e-mail: [hidden email]
>>>>>>> For additional commands, e-mail: [hidden email]
>>>>>>>
>>>>>>
>>>>>>
>>>>>> -----------------------------------------------------------------
>>>>>> ----
>>>>>> To unsubscribe, e-mail: [hidden email]
>>>>>> For additional commands, e-mail: [hidden email]
>>>>>>
>>>>>
>>>>>
>>>>> ------------------------------------------------------------------
>>>>> ---
>>>>> To unsubscribe, e-mail: [hidden email]
>>>>> For additional commands, e-mail: [hidden email]
>>>>>
>>>>
>>>
>>
>
>
> ---------------------------------------------------------------------
> To unsubscribe, e-mail: [hidden email]
> For additional commands, e-mail: [hidden email]


---------------------------------------------------------------------
To unsubscribe, e-mail: [hidden email]
For additional commands, e-mail: [hidden email]

Reply | Threaded
Open this post in threaded view
|

Re: ThreadLocal in SegmentReader

Roman Puchkovskiy-2
In reply to this post by Michael McCandless-2
The problem here is not because ThreadLocal instances are not GC'd (they are GC'd, and your test shows this clearly).
But even one instance which is not removed from its Thread is enough to prevent the classloader from being unloaded, and that's the problem.

Michael McCandless-2 wrote
OK, I created a simple test to test this (attached).  The test just  
runs 10 threads, each one creating a 100 KB byte array which is stored  
into a ThreadLocal, and then periodically the ThreadLocal is replaced  
with a new one.  This is to test whether GC of a ThreadLocal, even  
though the thread is still alive, in fact leads to GC of the objects  
held in the ThreadLocal.

Indeed on Sun JRE 1.4, 1.5 and 1.6 it appears that the objects are in  
fact properly collected.

So this is not a leak but rather a "delayed collection" issue.  Java's  
GC is never guaranteed to be immediate, and apparently when using  
ThreadLocals it's even less immediate than "normal".  In the original  
issue, if other things create ThreadLocals, then eventually Lucene's  
unreferenced ThreadLocals would be properly collected.

So I think we continue to use non-static ThreadLocals in Lucene...

Mike


 



robert engels wrote:

> Once again, these are "static" thread locals. A completely different  
> issue. Since the object is available statically, the weak reference  
> cannot be cleared so stale entries will never be cleared as long as  
> the thread is alive.
>
> On Jul 9, 2008, at 4:46 PM, Adrian Tarau wrote:
>
>> Just a few examples of "problems" using ThreadLocals.
>>
>> http://opensource.atlassian.com/projects/hibernate/browse/HHH-2481
>> http://www.theserverside.com/news/thread.tss?thread_id=41473
>>
>> Once again, I'm not pointing to Lucene SegmentReader as a "bad"  
>> implementation, and maybe the current "problems" of ThreadLocals  
>> are not a problem for SegmentReader but it seems safer to use  
>> ThreadLocals to pass context information which is cleared when the  
>> call exits instead of storing long-lived objects.
>>
>>
>> robert engels wrote:
>>>
>>> Aside from the pre-1.5 thread local "perceived leak", there are no  
>>> issues with ThreadLocals if used properly.
>>>
>>> There is no need for try/finally blocks, unless you MUST release  
>>> resources immediately, usually this is not the case, which is why  
>>> a ThreadLocal is used in the first place.
>>>
>>> From the ThreadLocalMap javadoc...
>>>
>>>  /**
>>>      * ThreadLocalMap is a customized hash map suitable only for
>>>      * maintaining thread local values. No operations are exported
>>>      * outside of the ThreadLocal class. The class is package  
>>> private to
>>>      * allow declaration of fields in class Thread.  To help deal  
>>> with
>>>      * very large and long-lived usages, the hash table entries use
>>>      * WeakReferences for keys. However, since reference queues  
>>> are not
>>>      * used, stale entries are guaranteed to be removed only when
>>>      * the table starts running out of space.
>>>      */
>>>
>>> /**
>>>          * Heuristically scan some cells looking for stale entries.
>>>          * This is invoked when either a new element is added, or
>>>          * another stale one has been expunged. It performs a
>>>          * logarithmic number of scans, as a balance between no
>>>          * scanning (fast but retains garbage) and a number of scans
>>>          * proportional to number of elements, that would find all
>>>          * garbage but would cause some insertions to take O(n)  
>>> time.
>>>          *
>>>          * @param i a position known NOT to hold a stale entry. The
>>>          * scan starts at the element after i.
>>>          *
>>>          * @param n scan control: <tt>log2(n)</tt> cells are  
>>> scanned,
>>>          * unless a stale entry one is found, in which case
>>>          * <tt>log2(table.length)-1</tt> additional cells are  
>>> scanned.
>>>          * When called from insertions, this parameter is the number
>>>          * of elements, but when from replaceStaleEntry, it is the
>>>          * table length. (Note: all this could be changed to be  
>>> either
>>>          * more or less aggressive by weighting n instead of just
>>>          * using straight log n. But this version is simple, fast,  
>>> and
>>>          * seems to work well.)
>>>          *
>>>          * @return true if any stale entries have been removed.
>>>          */
>>>
>>>
>>> The instance ThreadLocals (and what the refer to) will be GC'd  
>>> when the containing Object is GC'd.
>>>
>>> There IS NO MEMORY LEAK in ThreadLocal. If the ThreadLocal refers  
>>> to an object that has native resources (e.g. file handles), it may  
>>> not be released until other thread locals are created by the  
>>> thread (or the thread terminates).
>>>
>>> You can avoid this "delay" by calling remove(), but in most  
>>> applications it should never be necessary - unless a very strange  
>>> usage...
>>>
>>> On Jul 9, 2008, at 2:37 PM, Adrian Tarau wrote:
>>>
>>>> From what I know, storing objects in ThreadLocal is safe as long  
>>>> as you release the object within a try {} finall {} block or  
>>>> store objects which are independent of the rest of the code(no  
>>>> dependencies).Otherwise it can get pretty tricky(memory leaks,  
>>>> classloader problems) after awhile.
>>>>
>>>> It is pretty convenient to pass HTTP request information with a  
>>>> ThreadLocal in a servlet(but you should cleanup the variable  
>>>> before leaving the servlet) but I'm not sure how safe it is in  
>>>> this case.
>>>>
>>>> robert engels wrote:
>>>>> Using synchronization is a poor/invalid substitute for thread  
>>>>> locals in many cases.
>>>>>
>>>>> The point of the thread local in these referenced cases is too  
>>>>> allow streaming reads on a file descriptor. if you use a shared  
>>>>> file descriptor/buffer you are going to continually invalidate  
>>>>> the buffer.
>>>>>
>>>>> On Jul 8, 2008, at 5:12 AM, Michael McCandless wrote:
>>>>>
>>>>>>
>>>>>> Well ... SegmentReader uses ThreadLocal to hold a thread-
>>>>>> private instance of TermVectorsReader, to avoid synchronizing  
>>>>>> per-document when loading term vectors.
>>>>>>
>>>>>> Clearing this ThreadLocal value per call to SegmentReader's  
>>>>>> methods that load term vectors would defeat its purpose.
>>>>>>
>>>>>> Though, of course, we then synchronize on the underlying file  
>>>>>> (when using FSDirectory), so perhaps we are really not saving  
>>>>>> much by using ThreadLocal here.  But we are looking to relax  
>>>>>> that low level synchronization with LUCENE-753.
>>>>>>
>>>>>> Maybe we could make our own ThreadLocal that just uses a  
>>>>>> HashMap, which we'd have to synchronize on when getting the per-
>>>>>> thread instances.  Or, go back to sharing a single  
>>>>>> TermVectorsReader and synchronize per-document.
>>>>>>
>>>>>> Jason has suggested moving to a model where you ask the  
>>>>>> IndexReader for an object that can return term vectors / stored  
>>>>>> fields / etc, and then you interact with that many times to  
>>>>>> retrieve each doc.  We could then synchronize only on  
>>>>>> retrieving that object, and provide a thread-private instance.
>>>>>>
>>>>>> It seems like we should move away from using ThreadLocal in  
>>>>>> Lucene and do "normal" synchronization instead.
>>>>>>
>>>>>> Mike
>>>>>>
>>>>>> Adrian Tarau wrote:
>>>>>>
>>>>>>> Usually ThreadLocal.remove() should be called at the end(in a  
>>>>>>> finally block), before the current call leaves your code.
>>>>>>>
>>>>>>> Ex : if during searching ThreadLocal is used, every search(..)  
>>>>>>> method should cleanup any ThreadLocal variables, or even  
>>>>>>> deeper in the implementation. When the call leaves Lucene any  
>>>>>>> used ThreadLocal should be cleaned up.
>>>>>>>
>>>>>>> Michael McCandless wrote:
>>>>>>>>
>>>>>>>> ThreadLocal, which we use in several places in Lucene, causes  
>>>>>>>> a leak in app servers because the classloader never fully  
>>>>>>>> deallocates Lucene's classes because the ThreadLocal is  
>>>>>>>> holding strong references.
>>>>>>>>
>>>>>>>> Yet, ThreadLocal is very convenient for avoiding  
>>>>>>>> synchronization.
>>>>>>>>
>>>>>>>> Does anyone have any ideas on how to solve this w/o falling  
>>>>>>>> back to "normal" synchronization?
>>>>>>>>
>>>>>>>> Mike
>>>>>>>>
>>>>>>>> Begin forwarded message:
>>>>>>>>
>>>>>>>>> From: "Yonik Seeley" <yonik@apache.org>
>>>>>>>>> Date: July 7, 2008 3:30:28 PM EDT
>>>>>>>>> To: java-user@lucene.apache.org
>>>>>>>>> Subject: Re: ThreadLocal in SegmentReader
>>>>>>>>> Reply-To: java-user@lucene.apache.org
>>>>>>>>>
>>>>>>>>> On Mon, Jul 7, 2008 at 2:43 PM, Michael McCandless
>>>>>>>>> <lucene@mikemccandless.com> wrote:
>>>>>>>>>> So now I'm confused: the SegmentReader itself should no  
>>>>>>>>>> longer be reachable,
>>>>>>>>>> assuming you are not holding any references to your  
>>>>>>>>>> IndexReader.
>>>>>>>>>>
>>>>>>>>>> Which means the ThreadLocal instance should no longer be  
>>>>>>>>>> reachable.
>>>>>>>>>
>>>>>>>>> It will still be referenced from the Thread(s) ThreadLocalMap
>>>>>>>>> The key (the ThreadLocal) will be weakly referenced, but the  
>>>>>>>>> values
>>>>>>>>> (now stale) are strongly referenced and won't be actually  
>>>>>>>>> removed
>>>>>>>>> until the table is resized (under the Java6 impl at least).
>>>>>>>>> Nice huh?
>>>>>>>>>
>>>>>>>>> -Yonik
>>>>>>>>>
>>>>>>>>> ---------------------------------------------------------------------
>>>>>>>>> To unsubscribe, e-mail: java-user-
>>>>>>>>> unsubscribe@lucene.apache.org
>>>>>>>>> For additional commands, e-mail: java-user-help@lucene.apache.org
>>>>>>>>>
>>>>>>>>
>>>>>>>>
>>>>>>>> ---------------------------------------------------------------------
>>>>>>>> To unsubscribe, e-mail: java-dev-unsubscribe@lucene.apache.org
>>>>>>>> For additional commands, e-mail: java-dev-
>>>>>>>> help@lucene.apache.org
>>>>>>>>
>>>>>>>
>>>>>>>
>>>>>>> ---------------------------------------------------------------------
>>>>>>> To unsubscribe, e-mail: java-dev-unsubscribe@lucene.apache.org
>>>>>>> For additional commands, e-mail: java-dev-help@lucene.apache.org
>>>>>>>
>>>>>>
>>>>>>
>>>>>> ---------------------------------------------------------------------
>>>>>> To unsubscribe, e-mail: java-dev-unsubscribe@lucene.apache.org
>>>>>> For additional commands, e-mail: java-dev-help@lucene.apache.org
>>>>>>
>>>>>
>>>>>
>>>>> ---------------------------------------------------------------------
>>>>> To unsubscribe, e-mail: java-dev-unsubscribe@lucene.apache.org
>>>>> For additional commands, e-mail: java-dev-help@lucene.apache.org
>>>>>
>>>>
>>>>
>>>> ---------------------------------------------------------------------
>>>> To unsubscribe, e-mail: java-dev-unsubscribe@lucene.apache.org
>>>> For additional commands, e-mail: java-dev-help@lucene.apache.org
>>>>
>>>
>>
>



---------------------------------------------------------------------
To unsubscribe, e-mail: java-dev-unsubscribe@lucene.apache.org
For additional commands, e-mail: java-dev-help@lucene.apache.org
Reply | Threaded
Open this post in threaded view
|

Re: ThreadLocal in SegmentReader

Robert Engels
This is only an issue for static ThreadLocals ...

On Jul 11, 2008, at 11:32 PM, Roman Puchkovskiy wrote:

>
> The problem here is not because ThreadLocal instances are not GC'd  
> (they are
> GC'd, and your test shows this clearly).
> But even one instance which is not removed from its Thread is  
> enough to
> prevent the classloader from being unloaded, and that's the problem.
>
>
> Michael McCandless-2 wrote:
>>
>> OK, I created a simple test to test this (attached).  The test just
>> runs 10 threads, each one creating a 100 KB byte array which is  
>> stored
>> into a ThreadLocal, and then periodically the ThreadLocal is replaced
>> with a new one.  This is to test whether GC of a ThreadLocal, even
>> though the thread is still alive, in fact leads to GC of the objects
>> held in the ThreadLocal.
>>
>> Indeed on Sun JRE 1.4, 1.5 and 1.6 it appears that the objects are in
>> fact properly collected.
>>
>> So this is not a leak but rather a "delayed collection" issue.  
>> Java's
>> GC is never guaranteed to be immediate, and apparently when using
>> ThreadLocals it's even less immediate than "normal".  In the original
>> issue, if other things create ThreadLocals, then eventually Lucene's
>> unreferenced ThreadLocals would be properly collected.
>>
>> So I think we continue to use non-static ThreadLocals in Lucene...
>>
>> Mike
>>
>>
>>
>>
>>
>>
>> robert engels wrote:
>>
>>> Once again, these are "static" thread locals. A completely different
>>> issue. Since the object is available statically, the weak reference
>>> cannot be cleared so stale entries will never be cleared as long as
>>> the thread is alive.
>>>
>>> On Jul 9, 2008, at 4:46 PM, Adrian Tarau wrote:
>>>
>>>> Just a few examples of "problems" using ThreadLocals.
>>>>
>>>> http://opensource.atlassian.com/projects/hibernate/browse/HHH-2481
>>>> http://www.theserverside.com/news/thread.tss?thread_id=41473
>>>>
>>>> Once again, I'm not pointing to Lucene SegmentReader as a "bad"
>>>> implementation, and maybe the current "problems" of ThreadLocals
>>>> are not a problem for SegmentReader but it seems safer to use
>>>> ThreadLocals to pass context information which is cleared when the
>>>> call exits instead of storing long-lived objects.
>>>>
>>>>
>>>> robert engels wrote:
>>>>>
>>>>> Aside from the pre-1.5 thread local "perceived leak", there are no
>>>>> issues with ThreadLocals if used properly.
>>>>>
>>>>> There is no need for try/finally blocks, unless you MUST release
>>>>> resources immediately, usually this is not the case, which is why
>>>>> a ThreadLocal is used in the first place.
>>>>>
>>>>> From the ThreadLocalMap javadoc...
>>>>>
>>>>>  /**
>>>>>      * ThreadLocalMap is a customized hash map suitable only for
>>>>>      * maintaining thread local values. No operations are exported
>>>>>      * outside of the ThreadLocal class. The class is package
>>>>> private to
>>>>>      * allow declaration of fields in class Thread.  To help deal
>>>>> with
>>>>>      * very large and long-lived usages, the hash table entries  
>>>>> use
>>>>>      * WeakReferences for keys. However, since reference queues
>>>>> are not
>>>>>      * used, stale entries are guaranteed to be removed only when
>>>>>      * the table starts running out of space.
>>>>>      */
>>>>>
>>>>> /**
>>>>>          * Heuristically scan some cells looking for stale  
>>>>> entries.
>>>>>          * This is invoked when either a new element is added, or
>>>>>          * another stale one has been expunged. It performs a
>>>>>          * logarithmic number of scans, as a balance between no
>>>>>          * scanning (fast but retains garbage) and a number of  
>>>>> scans
>>>>>          * proportional to number of elements, that would find all
>>>>>          * garbage but would cause some insertions to take O(n)
>>>>> time.
>>>>>          *
>>>>>          * @param i a position known NOT to hold a stale entry.  
>>>>> The
>>>>>          * scan starts at the element after i.
>>>>>          *
>>>>>          * @param n scan control: <tt>log2(n)</tt> cells are
>>>>> scanned,
>>>>>          * unless a stale entry one is found, in which case
>>>>>          * <tt>log2(table.length)-1</tt> additional cells are
>>>>> scanned.
>>>>>          * When called from insertions, this parameter is the  
>>>>> number
>>>>>          * of elements, but when from replaceStaleEntry, it is the
>>>>>          * table length. (Note: all this could be changed to be
>>>>> either
>>>>>          * more or less aggressive by weighting n instead of just
>>>>>          * using straight log n. But this version is simple, fast,
>>>>> and
>>>>>          * seems to work well.)
>>>>>          *
>>>>>          * @return true if any stale entries have been removed.
>>>>>          */
>>>>>
>>>>>
>>>>> The instance ThreadLocals (and what the refer to) will be GC'd
>>>>> when the containing Object is GC'd.
>>>>>
>>>>> There IS NO MEMORY LEAK in ThreadLocal. If the ThreadLocal refers
>>>>> to an object that has native resources (e.g. file handles), it may
>>>>> not be released until other thread locals are created by the
>>>>> thread (or the thread terminates).
>>>>>
>>>>> You can avoid this "delay" by calling remove(), but in most
>>>>> applications it should never be necessary - unless a very strange
>>>>> usage...
>>>>>
>>>>> On Jul 9, 2008, at 2:37 PM, Adrian Tarau wrote:
>>>>>
>>>>>> From what I know, storing objects in ThreadLocal is safe as long
>>>>>> as you release the object within a try {} finall {} block or
>>>>>> store objects which are independent of the rest of the code(no
>>>>>> dependencies).Otherwise it can get pretty tricky(memory leaks,
>>>>>> classloader problems) after awhile.
>>>>>>
>>>>>> It is pretty convenient to pass HTTP request information with a
>>>>>> ThreadLocal in a servlet(but you should cleanup the variable
>>>>>> before leaving the servlet) but I'm not sure how safe it is in
>>>>>> this case.
>>>>>>
>>>>>> robert engels wrote:
>>>>>>> Using synchronization is a poor/invalid substitute for thread
>>>>>>> locals in many cases.
>>>>>>>
>>>>>>> The point of the thread local in these referenced cases is too
>>>>>>> allow streaming reads on a file descriptor. if you use a shared
>>>>>>> file descriptor/buffer you are going to continually invalidate
>>>>>>> the buffer.
>>>>>>>
>>>>>>> On Jul 8, 2008, at 5:12 AM, Michael McCandless wrote:
>>>>>>>
>>>>>>>>
>>>>>>>> Well ... SegmentReader uses ThreadLocal to hold a thread-
>>>>>>>> private instance of TermVectorsReader, to avoid synchronizing
>>>>>>>> per-document when loading term vectors.
>>>>>>>>
>>>>>>>> Clearing this ThreadLocal value per call to SegmentReader's
>>>>>>>> methods that load term vectors would defeat its purpose.
>>>>>>>>
>>>>>>>> Though, of course, we then synchronize on the underlying file
>>>>>>>> (when using FSDirectory), so perhaps we are really not saving
>>>>>>>> much by using ThreadLocal here.  But we are looking to relax
>>>>>>>> that low level synchronization with LUCENE-753.
>>>>>>>>
>>>>>>>> Maybe we could make our own ThreadLocal that just uses a
>>>>>>>> HashMap, which we'd have to synchronize on when getting the  
>>>>>>>> per-
>>>>>>>> thread instances.  Or, go back to sharing a single
>>>>>>>> TermVectorsReader and synchronize per-document.
>>>>>>>>
>>>>>>>> Jason has suggested moving to a model where you ask the
>>>>>>>> IndexReader for an object that can return term vectors / stored
>>>>>>>> fields / etc, and then you interact with that many times to
>>>>>>>> retrieve each doc.  We could then synchronize only on
>>>>>>>> retrieving that object, and provide a thread-private instance.
>>>>>>>>
>>>>>>>> It seems like we should move away from using ThreadLocal in
>>>>>>>> Lucene and do "normal" synchronization instead.
>>>>>>>>
>>>>>>>> Mike
>>>>>>>>
>>>>>>>> Adrian Tarau wrote:
>>>>>>>>
>>>>>>>>> Usually ThreadLocal.remove() should be called at the end(in a
>>>>>>>>> finally block), before the current call leaves your code.
>>>>>>>>>
>>>>>>>>> Ex : if during searching ThreadLocal is used, every search(..)
>>>>>>>>> method should cleanup any ThreadLocal variables, or even
>>>>>>>>> deeper in the implementation. When the call leaves Lucene any
>>>>>>>>> used ThreadLocal should be cleaned up.
>>>>>>>>>
>>>>>>>>> Michael McCandless wrote:
>>>>>>>>>>
>>>>>>>>>> ThreadLocal, which we use in several places in Lucene, causes
>>>>>>>>>> a leak in app servers because the classloader never fully
>>>>>>>>>> deallocates Lucene's classes because the ThreadLocal is
>>>>>>>>>> holding strong references.
>>>>>>>>>>
>>>>>>>>>> Yet, ThreadLocal is very convenient for avoiding
>>>>>>>>>> synchronization.
>>>>>>>>>>
>>>>>>>>>> Does anyone have any ideas on how to solve this w/o falling
>>>>>>>>>> back to "normal" synchronization?
>>>>>>>>>>
>>>>>>>>>> Mike
>>>>>>>>>>
>>>>>>>>>> Begin forwarded message:
>>>>>>>>>>
>>>>>>>>>>> From: "Yonik Seeley" <[hidden email]>
>>>>>>>>>>> Date: July 7, 2008 3:30:28 PM EDT
>>>>>>>>>>> To: [hidden email]
>>>>>>>>>>> Subject: Re: ThreadLocal in SegmentReader
>>>>>>>>>>> Reply-To: [hidden email]
>>>>>>>>>>>
>>>>>>>>>>> On Mon, Jul 7, 2008 at 2:43 PM, Michael McCandless
>>>>>>>>>>> <[hidden email]> wrote:
>>>>>>>>>>>> So now I'm confused: the SegmentReader itself should no
>>>>>>>>>>>> longer be reachable,
>>>>>>>>>>>> assuming you are not holding any references to your
>>>>>>>>>>>> IndexReader.
>>>>>>>>>>>>
>>>>>>>>>>>> Which means the ThreadLocal instance should no longer be
>>>>>>>>>>>> reachable.
>>>>>>>>>>>
>>>>>>>>>>> It will still be referenced from the Thread(s)  
>>>>>>>>>>> ThreadLocalMap
>>>>>>>>>>> The key (the ThreadLocal) will be weakly referenced, but the
>>>>>>>>>>> values
>>>>>>>>>>> (now stale) are strongly referenced and won't be actually
>>>>>>>>>>> removed
>>>>>>>>>>> until the table is resized (under the Java6 impl at least).
>>>>>>>>>>> Nice huh?
>>>>>>>>>>>
>>>>>>>>>>> -Yonik
>>>>>>>>>>>
>>>>>>>>>>> ------------------------------------------------------------
>>>>>>>>>>> ---------
>>>>>>>>>>> To unsubscribe, e-mail: java-user-
>>>>>>>>>>> [hidden email]
>>>>>>>>>>> For additional commands, e-mail: java-user-
>>>>>>>>>>> [hidden email]
>>>>>>>>>>>
>>>>>>>>>>
>>>>>>>>>>
>>>>>>>>>> -------------------------------------------------------------
>>>>>>>>>> --------
>>>>>>>>>> To unsubscribe, e-mail: java-dev-
>>>>>>>>>> [hidden email]
>>>>>>>>>> For additional commands, e-mail: java-dev-
>>>>>>>>>> [hidden email]
>>>>>>>>>>
>>>>>>>>>
>>>>>>>>>
>>>>>>>>> --------------------------------------------------------------
>>>>>>>>> -------
>>>>>>>>> To unsubscribe, e-mail: [hidden email]
>>>>>>>>> For additional commands, e-mail: java-dev-
>>>>>>>>> [hidden email]
>>>>>>>>>
>>>>>>>>
>>>>>>>>
>>>>>>>> ---------------------------------------------------------------
>>>>>>>> ------
>>>>>>>> To unsubscribe, e-mail: [hidden email]
>>>>>>>> For additional commands, e-mail: java-dev-
>>>>>>>> [hidden email]
>>>>>>>>
>>>>>>>
>>>>>>>
>>>>>>> ----------------------------------------------------------------
>>>>>>> -----
>>>>>>> To unsubscribe, e-mail: [hidden email]
>>>>>>> For additional commands, e-mail: [hidden email]
>>>>>>>
>>>>>>
>>>>>>
>>>>>> -----------------------------------------------------------------
>>>>>> ----
>>>>>> To unsubscribe, e-mail: [hidden email]
>>>>>> For additional commands, e-mail: [hidden email]
>>>>>>
>>>>>
>>>>
>>>
>>
>>
>>
>> ---------------------------------------------------------------------
>> To unsubscribe, e-mail: [hidden email]
>> For additional commands, e-mail: [hidden email]
>>
>
> --
> View this message in context: http://www.nabble.com/Fwd%3A- 
> ThreadLocal-in-SegmentReader-tp18326720p18416022.html
> Sent from the Lucene - Java Developer mailing list archive at  
> Nabble.com.
>
>
> ---------------------------------------------------------------------
> To unsubscribe, e-mail: [hidden email]
> For additional commands, e-mail: [hidden email]
>


---------------------------------------------------------------------
To unsubscribe, e-mail: [hidden email]
For additional commands, e-mail: [hidden email]

Reply | Threaded
Open this post in threaded view
|

Re: ThreadLocal in SegmentReader

Roman Puchkovskiy-2
Well, possibly I'm mistaken, but it seems that this affects non-static fields too. Please see http://www.nabble.com/ThreadLocal-in-SegmentReader-to18306230.html where the use case is described in the details.
In short: it seems that the scope of ThreadLocals does not matter. What really matters is that they are referenced by ThreadLocals map in the thread which is still alive.

Robert Engels wrote
This is only an issue for static ThreadLocals ...

On Jul 11, 2008, at 11:32 PM, Roman Puchkovskiy wrote:

>
> The problem here is not because ThreadLocal instances are not GC'd  
> (they are
> GC'd, and your test shows this clearly).
> But even one instance which is not removed from its Thread is  
> enough to
> prevent the classloader from being unloaded, and that's the problem.
>
>
> Michael McCandless-2 wrote:
>>
>> OK, I created a simple test to test this (attached).  The test just
>> runs 10 threads, each one creating a 100 KB byte array which is  
>> stored
>> into a ThreadLocal, and then periodically the ThreadLocal is replaced
>> with a new one.  This is to test whether GC of a ThreadLocal, even
>> though the thread is still alive, in fact leads to GC of the objects
>> held in the ThreadLocal.
>>
>> Indeed on Sun JRE 1.4, 1.5 and 1.6 it appears that the objects are in
>> fact properly collected.
>>
>> So this is not a leak but rather a "delayed collection" issue.  
>> Java's
>> GC is never guaranteed to be immediate, and apparently when using
>> ThreadLocals it's even less immediate than "normal".  In the original
>> issue, if other things create ThreadLocals, then eventually Lucene's
>> unreferenced ThreadLocals would be properly collected.
>>
>> So I think we continue to use non-static ThreadLocals in Lucene...
>>
>> Mike
>>
>>
>>
>>
>>
>>
>> robert engels wrote:
>>
>>> Once again, these are "static" thread locals. A completely different
>>> issue. Since the object is available statically, the weak reference
>>> cannot be cleared so stale entries will never be cleared as long as
>>> the thread is alive.
>>>
>>> On Jul 9, 2008, at 4:46 PM, Adrian Tarau wrote:
>>>
>>>> Just a few examples of "problems" using ThreadLocals.
>>>>
>>>> http://opensource.atlassian.com/projects/hibernate/browse/HHH-2481
>>>> http://www.theserverside.com/news/thread.tss?thread_id=41473
>>>>
>>>> Once again, I'm not pointing to Lucene SegmentReader as a "bad"
>>>> implementation, and maybe the current "problems" of ThreadLocals
>>>> are not a problem for SegmentReader but it seems safer to use
>>>> ThreadLocals to pass context information which is cleared when the
>>>> call exits instead of storing long-lived objects.
>>>>
>>>>
>>>> robert engels wrote:
>>>>>
>>>>> Aside from the pre-1.5 thread local "perceived leak", there are no
>>>>> issues with ThreadLocals if used properly.
>>>>>
>>>>> There is no need for try/finally blocks, unless you MUST release
>>>>> resources immediately, usually this is not the case, which is why
>>>>> a ThreadLocal is used in the first place.
>>>>>
>>>>> From the ThreadLocalMap javadoc...
>>>>>
>>>>>  /**
>>>>>      * ThreadLocalMap is a customized hash map suitable only for
>>>>>      * maintaining thread local values. No operations are exported
>>>>>      * outside of the ThreadLocal class. The class is package
>>>>> private to
>>>>>      * allow declaration of fields in class Thread.  To help deal
>>>>> with
>>>>>      * very large and long-lived usages, the hash table entries  
>>>>> use
>>>>>      * WeakReferences for keys. However, since reference queues
>>>>> are not
>>>>>      * used, stale entries are guaranteed to be removed only when
>>>>>      * the table starts running out of space.
>>>>>      */
>>>>>
>>>>> /**
>>>>>          * Heuristically scan some cells looking for stale  
>>>>> entries.
>>>>>          * This is invoked when either a new element is added, or
>>>>>          * another stale one has been expunged. It performs a
>>>>>          * logarithmic number of scans, as a balance between no
>>>>>          * scanning (fast but retains garbage) and a number of  
>>>>> scans
>>>>>          * proportional to number of elements, that would find all
>>>>>          * garbage but would cause some insertions to take O(n)
>>>>> time.
>>>>>          *
>>>>>          * @param i a position known NOT to hold a stale entry.  
>>>>> The
>>>>>          * scan starts at the element after i.
>>>>>          *
>>>>>          * @param n scan control: <tt>log2(n)</tt> cells are
>>>>> scanned,
>>>>>          * unless a stale entry one is found, in which case
>>>>>          * <tt>log2(table.length)-1</tt> additional cells are
>>>>> scanned.
>>>>>          * When called from insertions, this parameter is the  
>>>>> number
>>>>>          * of elements, but when from replaceStaleEntry, it is the
>>>>>          * table length. (Note: all this could be changed to be
>>>>> either
>>>>>          * more or less aggressive by weighting n instead of just
>>>>>          * using straight log n. But this version is simple, fast,
>>>>> and
>>>>>          * seems to work well.)
>>>>>          *
>>>>>          * @return true if any stale entries have been removed.
>>>>>          */
>>>>>
>>>>>
>>>>> The instance ThreadLocals (and what the refer to) will be GC'd
>>>>> when the containing Object is GC'd.
>>>>>
>>>>> There IS NO MEMORY LEAK in ThreadLocal. If the ThreadLocal refers
>>>>> to an object that has native resources (e.g. file handles), it may
>>>>> not be released until other thread locals are created by the
>>>>> thread (or the thread terminates).
>>>>>
>>>>> You can avoid this "delay" by calling remove(), but in most
>>>>> applications it should never be necessary - unless a very strange
>>>>> usage...
>>>>>
>>>>> On Jul 9, 2008, at 2:37 PM, Adrian Tarau wrote:
>>>>>
>>>>>> From what I know, storing objects in ThreadLocal is safe as long
>>>>>> as you release the object within a try {} finall {} block or
>>>>>> store objects which are independent of the rest of the code(no
>>>>>> dependencies).Otherwise it can get pretty tricky(memory leaks,
>>>>>> classloader problems) after awhile.
>>>>>>
>>>>>> It is pretty convenient to pass HTTP request information with a
>>>>>> ThreadLocal in a servlet(but you should cleanup the variable
>>>>>> before leaving the servlet) but I'm not sure how safe it is in
>>>>>> this case.
>>>>>>
>>>>>> robert engels wrote:
>>>>>>> Using synchronization is a poor/invalid substitute for thread
>>>>>>> locals in many cases.
>>>>>>>
>>>>>>> The point of the thread local in these referenced cases is too
>>>>>>> allow streaming reads on a file descriptor. if you use a shared
>>>>>>> file descriptor/buffer you are going to continually invalidate
>>>>>>> the buffer.
>>>>>>>
>>>>>>> On Jul 8, 2008, at 5:12 AM, Michael McCandless wrote:
>>>>>>>
>>>>>>>>
>>>>>>>> Well ... SegmentReader uses ThreadLocal to hold a thread-
>>>>>>>> private instance of TermVectorsReader, to avoid synchronizing
>>>>>>>> per-document when loading term vectors.
>>>>>>>>
>>>>>>>> Clearing this ThreadLocal value per call to SegmentReader's
>>>>>>>> methods that load term vectors would defeat its purpose.
>>>>>>>>
>>>>>>>> Though, of course, we then synchronize on the underlying file
>>>>>>>> (when using FSDirectory), so perhaps we are really not saving
>>>>>>>> much by using ThreadLocal here.  But we are looking to relax
>>>>>>>> that low level synchronization with LUCENE-753.
>>>>>>>>
>>>>>>>> Maybe we could make our own ThreadLocal that just uses a
>>>>>>>> HashMap, which we'd have to synchronize on when getting the  
>>>>>>>> per-
>>>>>>>> thread instances.  Or, go back to sharing a single
>>>>>>>> TermVectorsReader and synchronize per-document.
>>>>>>>>
>>>>>>>> Jason has suggested moving to a model where you ask the
>>>>>>>> IndexReader for an object that can return term vectors / stored
>>>>>>>> fields / etc, and then you interact with that many times to
>>>>>>>> retrieve each doc.  We could then synchronize only on
>>>>>>>> retrieving that object, and provide a thread-private instance.
>>>>>>>>
>>>>>>>> It seems like we should move away from using ThreadLocal in
>>>>>>>> Lucene and do "normal" synchronization instead.
>>>>>>>>
>>>>>>>> Mike
>>>>>>>>
>>>>>>>> Adrian Tarau wrote:
>>>>>>>>
>>>>>>>>> Usually ThreadLocal.remove() should be called at the end(in a
>>>>>>>>> finally block), before the current call leaves your code.
>>>>>>>>>
>>>>>>>>> Ex : if during searching ThreadLocal is used, every search(..)
>>>>>>>>> method should cleanup any ThreadLocal variables, or even
>>>>>>>>> deeper in the implementation. When the call leaves Lucene any
>>>>>>>>> used ThreadLocal should be cleaned up.
>>>>>>>>>
>>>>>>>>> Michael McCandless wrote:
>>>>>>>>>>
>>>>>>>>>> ThreadLocal, which we use in several places in Lucene, causes
>>>>>>>>>> a leak in app servers because the classloader never fully
>>>>>>>>>> deallocates Lucene's classes because the ThreadLocal is
>>>>>>>>>> holding strong references.
>>>>>>>>>>
>>>>>>>>>> Yet, ThreadLocal is very convenient for avoiding
>>>>>>>>>> synchronization.
>>>>>>>>>>
>>>>>>>>>> Does anyone have any ideas on how to solve this w/o falling
>>>>>>>>>> back to "normal" synchronization?
>>>>>>>>>>
>>>>>>>>>> Mike
>>>>>>>>>>
>>>>>>>>>> Begin forwarded message:
>>>>>>>>>>
>>>>>>>>>>> From: "Yonik Seeley" <yonik@apache.org>
>>>>>>>>>>> Date: July 7, 2008 3:30:28 PM EDT
>>>>>>>>>>> To: java-user@lucene.apache.org
>>>>>>>>>>> Subject: Re: ThreadLocal in SegmentReader
>>>>>>>>>>> Reply-To: java-user@lucene.apache.org
>>>>>>>>>>>
>>>>>>>>>>> On Mon, Jul 7, 2008 at 2:43 PM, Michael McCandless
>>>>>>>>>>> <lucene@mikemccandless.com> wrote:
>>>>>>>>>>>> So now I'm confused: the SegmentReader itself should no
>>>>>>>>>>>> longer be reachable,
>>>>>>>>>>>> assuming you are not holding any references to your
>>>>>>>>>>>> IndexReader.
>>>>>>>>>>>>
>>>>>>>>>>>> Which means the ThreadLocal instance should no longer be
>>>>>>>>>>>> reachable.
>>>>>>>>>>>
>>>>>>>>>>> It will still be referenced from the Thread(s)  
>>>>>>>>>>> ThreadLocalMap
>>>>>>>>>>> The key (the ThreadLocal) will be weakly referenced, but the
>>>>>>>>>>> values
>>>>>>>>>>> (now stale) are strongly referenced and won't be actually
>>>>>>>>>>> removed
>>>>>>>>>>> until the table is resized (under the Java6 impl at least).
>>>>>>>>>>> Nice huh?
>>>>>>>>>>>
>>>>>>>>>>> -Yonik
>>>>>>>>>>>
>>>>>>>>>>> ------------------------------------------------------------
>>>>>>>>>>> ---------
>>>>>>>>>>> To unsubscribe, e-mail: java-user-
>>>>>>>>>>> unsubscribe@lucene.apache.org
>>>>>>>>>>> For additional commands, e-mail: java-user-
>>>>>>>>>>> help@lucene.apache.org
>>>>>>>>>>>
>>>>>>>>>>
>>>>>>>>>>
>>>>>>>>>> -------------------------------------------------------------
>>>>>>>>>> --------
>>>>>>>>>> To unsubscribe, e-mail: java-dev-
>>>>>>>>>> unsubscribe@lucene.apache.org
>>>>>>>>>> For additional commands, e-mail: java-dev-
>>>>>>>>>> help@lucene.apache.org
>>>>>>>>>>
>>>>>>>>>
>>>>>>>>>
>>>>>>>>> --------------------------------------------------------------
>>>>>>>>> -------
>>>>>>>>> To unsubscribe, e-mail: java-dev-unsubscribe@lucene.apache.org
>>>>>>>>> For additional commands, e-mail: java-dev-
>>>>>>>>> help@lucene.apache.org
>>>>>>>>>
>>>>>>>>
>>>>>>>>
>>>>>>>> ---------------------------------------------------------------
>>>>>>>> ------
>>>>>>>> To unsubscribe, e-mail: java-dev-unsubscribe@lucene.apache.org
>>>>>>>> For additional commands, e-mail: java-dev-
>>>>>>>> help@lucene.apache.org
>>>>>>>>
>>>>>>>
>>>>>>>
>>>>>>> ----------------------------------------------------------------
>>>>>>> -----
>>>>>>> To unsubscribe, e-mail: java-dev-unsubscribe@lucene.apache.org
>>>>>>> For additional commands, e-mail: java-dev-help@lucene.apache.org
>>>>>>>
>>>>>>
>>>>>>
>>>>>> -----------------------------------------------------------------
>>>>>> ----
>>>>>> To unsubscribe, e-mail: java-dev-unsubscribe@lucene.apache.org
>>>>>> For additional commands, e-mail: java-dev-help@lucene.apache.org
>>>>>>
>>>>>
>>>>
>>>
>>
>>
>>
>> ---------------------------------------------------------------------
>> To unsubscribe, e-mail: java-dev-unsubscribe@lucene.apache.org
>> For additional commands, e-mail: java-dev-help@lucene.apache.org
>>
>
> --
> View this message in context: http://www.nabble.com/Fwd%3A- 
> ThreadLocal-in-SegmentReader-tp18326720p18416022.html
> Sent from the Lucene - Java Developer mailing list archive at  
> Nabble.com.
>
>
> ---------------------------------------------------------------------
> To unsubscribe, e-mail: java-dev-unsubscribe@lucene.apache.org
> For additional commands, e-mail: java-dev-help@lucene.apache.org
>



---------------------------------------------------------------------
To unsubscribe, e-mail: java-dev-unsubscribe@lucene.apache.org
For additional commands, e-mail: java-dev-help@lucene.apache.org
Reply | Threaded
Open this post in threaded view
|

Re: ThreadLocal in SegmentReader

Robert Engels
You are mistaken - Yonik's comment in that thread is correct  
(although it is not just at table resize - any time a ThreadLocal is  
added, and any time the ThreadLocal is not found in its direct hash  
it might clear entries).

The ThreadLocals map only has a WeakReference to the ThreadLocal, so  
if that is the only reference, it will be GC'd - eventually, and then  
it will be cleared as new ThreadLocals are created.

With a static reference, the thread can reference the ThreadLocal at  
any time, and thus the WeakReference will not be cleared.

If the object is VERY large, and new ThreadLocals are not created it  
could cause a problem, but I don't think that is the case with  
Lucene, as  the objects stored in ThreadLocals are designed to live  
for the life of the SegmentReader/IndexReader and thread.


On Jul 12, 2008, at 2:12 AM, Roman Puchkovskiy wrote:

>
> Well, possibly I'm mistaken, but it seems that this affects non-
> static fields
> too. Please see
> http://www.nabble.com/ThreadLocal-in-SegmentReader-to18306230.html 
> where the
> use case is described in the details.
> In short: it seems that the scope of ThreadLocals does not matter.  
> What
> really matters is that they are referenced by ThreadLocals map in  
> the thread
> which is still alive.
>
>
> Robert Engels wrote:
>>
>> This is only an issue for static ThreadLocals ...
>>
>> On Jul 11, 2008, at 11:32 PM, Roman Puchkovskiy wrote:
>>
>>>
>>> The problem here is not because ThreadLocal instances are not GC'd
>>> (they are
>>> GC'd, and your test shows this clearly).
>>> But even one instance which is not removed from its Thread is
>>> enough to
>>> prevent the classloader from being unloaded, and that's the problem.
>>>
>>>
>>> Michael McCandless-2 wrote:
>>>>
>>>> OK, I created a simple test to test this (attached).  The test just
>>>> runs 10 threads, each one creating a 100 KB byte array which is
>>>> stored
>>>> into a ThreadLocal, and then periodically the ThreadLocal is  
>>>> replaced
>>>> with a new one.  This is to test whether GC of a ThreadLocal, even
>>>> though the thread is still alive, in fact leads to GC of the  
>>>> objects
>>>> held in the ThreadLocal.
>>>>
>>>> Indeed on Sun JRE 1.4, 1.5 and 1.6 it appears that the objects  
>>>> are in
>>>> fact properly collected.
>>>>
>>>> So this is not a leak but rather a "delayed collection" issue.
>>>> Java's
>>>> GC is never guaranteed to be immediate, and apparently when using
>>>> ThreadLocals it's even less immediate than "normal".  In the  
>>>> original
>>>> issue, if other things create ThreadLocals, then eventually  
>>>> Lucene's
>>>> unreferenced ThreadLocals would be properly collected.
>>>>
>>>> So I think we continue to use non-static ThreadLocals in Lucene...
>>>>
>>>> Mike
>>>>
>>>>
>>>>
>>>>
>>>>
>>>>
>>>> robert engels wrote:
>>>>
>>>>> Once again, these are "static" thread locals. A completely  
>>>>> different
>>>>> issue. Since the object is available statically, the weak  
>>>>> reference
>>>>> cannot be cleared so stale entries will never be cleared as  
>>>>> long as
>>>>> the thread is alive.
>>>>>
>>>>> On Jul 9, 2008, at 4:46 PM, Adrian Tarau wrote:
>>>>>
>>>>>> Just a few examples of "problems" using ThreadLocals.
>>>>>>
>>>>>> http://opensource.atlassian.com/projects/hibernate/browse/ 
>>>>>> HHH-2481
>>>>>> http://www.theserverside.com/news/thread.tss?thread_id=41473
>>>>>>
>>>>>> Once again, I'm not pointing to Lucene SegmentReader as a "bad"
>>>>>> implementation, and maybe the current "problems" of ThreadLocals
>>>>>> are not a problem for SegmentReader but it seems safer to use
>>>>>> ThreadLocals to pass context information which is cleared when  
>>>>>> the
>>>>>> call exits instead of storing long-lived objects.
>>>>>>
>>>>>>
>>>>>> robert engels wrote:
>>>>>>>
>>>>>>> Aside from the pre-1.5 thread local "perceived leak", there  
>>>>>>> are no
>>>>>>> issues with ThreadLocals if used properly.
>>>>>>>
>>>>>>> There is no need for try/finally blocks, unless you MUST release
>>>>>>> resources immediately, usually this is not the case, which is  
>>>>>>> why
>>>>>>> a ThreadLocal is used in the first place.
>>>>>>>
>>>>>>> From the ThreadLocalMap javadoc...
>>>>>>>
>>>>>>>  /**
>>>>>>>      * ThreadLocalMap is a customized hash map suitable only for
>>>>>>>      * maintaining thread local values. No operations are  
>>>>>>> exported
>>>>>>>      * outside of the ThreadLocal class. The class is package
>>>>>>> private to
>>>>>>>      * allow declaration of fields in class Thread.  To help  
>>>>>>> deal
>>>>>>> with
>>>>>>>      * very large and long-lived usages, the hash table entries
>>>>>>> use
>>>>>>>      * WeakReferences for keys. However, since reference queues
>>>>>>> are not
>>>>>>>      * used, stale entries are guaranteed to be removed only  
>>>>>>> when
>>>>>>>      * the table starts running out of space.
>>>>>>>      */
>>>>>>>
>>>>>>> /**
>>>>>>>          * Heuristically scan some cells looking for stale
>>>>>>> entries.
>>>>>>>          * This is invoked when either a new element is  
>>>>>>> added, or
>>>>>>>          * another stale one has been expunged. It performs a
>>>>>>>          * logarithmic number of scans, as a balance between no
>>>>>>>          * scanning (fast but retains garbage) and a number of
>>>>>>> scans
>>>>>>>          * proportional to number of elements, that would  
>>>>>>> find all
>>>>>>>          * garbage but would cause some insertions to take O(n)
>>>>>>> time.
>>>>>>>          *
>>>>>>>          * @param i a position known NOT to hold a stale entry.
>>>>>>> The
>>>>>>>          * scan starts at the element after i.
>>>>>>>          *
>>>>>>>          * @param n scan control: <tt>log2(n)</tt> cells are
>>>>>>> scanned,
>>>>>>>          * unless a stale entry one is found, in which case
>>>>>>>          * <tt>log2(table.length)-1</tt> additional cells are
>>>>>>> scanned.
>>>>>>>          * When called from insertions, this parameter is the
>>>>>>> number
>>>>>>>          * of elements, but when from replaceStaleEntry, it  
>>>>>>> is the
>>>>>>>          * table length. (Note: all this could be changed to be
>>>>>>> either
>>>>>>>          * more or less aggressive by weighting n instead of  
>>>>>>> just
>>>>>>>          * using straight log n. But this version is simple,  
>>>>>>> fast,
>>>>>>> and
>>>>>>>          * seems to work well.)
>>>>>>>          *
>>>>>>>          * @return true if any stale entries have been removed.
>>>>>>>          */
>>>>>>>
>>>>>>>
>>>>>>> The instance ThreadLocals (and what the refer to) will be GC'd
>>>>>>> when the containing Object is GC'd.
>>>>>>>
>>>>>>> There IS NO MEMORY LEAK in ThreadLocal. If the ThreadLocal  
>>>>>>> refers
>>>>>>> to an object that has native resources (e.g. file handles),  
>>>>>>> it may
>>>>>>> not be released until other thread locals are created by the
>>>>>>> thread (or the thread terminates).
>>>>>>>
>>>>>>> You can avoid this "delay" by calling remove(), but in most
>>>>>>> applications it should never be necessary - unless a very  
>>>>>>> strange
>>>>>>> usage...
>>>>>>>
>>>>>>> On Jul 9, 2008, at 2:37 PM, Adrian Tarau wrote:
>>>>>>>
>>>>>>>> From what I know, storing objects in ThreadLocal is safe as  
>>>>>>>> long
>>>>>>>> as you release the object within a try {} finall {} block or
>>>>>>>> store objects which are independent of the rest of the code(no
>>>>>>>> dependencies).Otherwise it can get pretty tricky(memory leaks,
>>>>>>>> classloader problems) after awhile.
>>>>>>>>
>>>>>>>> It is pretty convenient to pass HTTP request information with a
>>>>>>>> ThreadLocal in a servlet(but you should cleanup the variable
>>>>>>>> before leaving the servlet) but I'm not sure how safe it is in
>>>>>>>> this case.
>>>>>>>>
>>>>>>>> robert engels wrote:
>>>>>>>>> Using synchronization is a poor/invalid substitute for thread
>>>>>>>>> locals in many cases.
>>>>>>>>>
>>>>>>>>> The point of the thread local in these referenced cases is too
>>>>>>>>> allow streaming reads on a file descriptor. if you use a  
>>>>>>>>> shared
>>>>>>>>> file descriptor/buffer you are going to continually invalidate
>>>>>>>>> the buffer.
>>>>>>>>>
>>>>>>>>> On Jul 8, 2008, at 5:12 AM, Michael McCandless wrote:
>>>>>>>>>
>>>>>>>>>>
>>>>>>>>>> Well ... SegmentReader uses ThreadLocal to hold a thread-
>>>>>>>>>> private instance of TermVectorsReader, to avoid synchronizing
>>>>>>>>>> per-document when loading term vectors.
>>>>>>>>>>
>>>>>>>>>> Clearing this ThreadLocal value per call to SegmentReader's
>>>>>>>>>> methods that load term vectors would defeat its purpose.
>>>>>>>>>>
>>>>>>>>>> Though, of course, we then synchronize on the underlying file
>>>>>>>>>> (when using FSDirectory), so perhaps we are really not saving
>>>>>>>>>> much by using ThreadLocal here.  But we are looking to relax
>>>>>>>>>> that low level synchronization with LUCENE-753.
>>>>>>>>>>
>>>>>>>>>> Maybe we could make our own ThreadLocal that just uses a
>>>>>>>>>> HashMap, which we'd have to synchronize on when getting the
>>>>>>>>>> per-
>>>>>>>>>> thread instances.  Or, go back to sharing a single
>>>>>>>>>> TermVectorsReader and synchronize per-document.
>>>>>>>>>>
>>>>>>>>>> Jason has suggested moving to a model where you ask the
>>>>>>>>>> IndexReader for an object that can return term vectors /  
>>>>>>>>>> stored
>>>>>>>>>> fields / etc, and then you interact with that many times to
>>>>>>>>>> retrieve each doc.  We could then synchronize only on
>>>>>>>>>> retrieving that object, and provide a thread-private  
>>>>>>>>>> instance.
>>>>>>>>>>
>>>>>>>>>> It seems like we should move away from using ThreadLocal in
>>>>>>>>>> Lucene and do "normal" synchronization instead.
>>>>>>>>>>
>>>>>>>>>> Mike
>>>>>>>>>>
>>>>>>>>>> Adrian Tarau wrote:
>>>>>>>>>>
>>>>>>>>>>> Usually ThreadLocal.remove() should be called at the end
>>>>>>>>>>> (in a
>>>>>>>>>>> finally block), before the current call leaves your code.
>>>>>>>>>>>
>>>>>>>>>>> Ex : if during searching ThreadLocal is used, every search
>>>>>>>>>>> (..)
>>>>>>>>>>> method should cleanup any ThreadLocal variables, or even
>>>>>>>>>>> deeper in the implementation. When the call leaves Lucene  
>>>>>>>>>>> any
>>>>>>>>>>> used ThreadLocal should be cleaned up.
>>>>>>>>>>>
>>>>>>>>>>> Michael McCandless wrote:
>>>>>>>>>>>>
>>>>>>>>>>>> ThreadLocal, which we use in several places in Lucene,  
>>>>>>>>>>>> causes
>>>>>>>>>>>> a leak in app servers because the classloader never fully
>>>>>>>>>>>> deallocates Lucene's classes because the ThreadLocal is
>>>>>>>>>>>> holding strong references.
>>>>>>>>>>>>
>>>>>>>>>>>> Yet, ThreadLocal is very convenient for avoiding
>>>>>>>>>>>> synchronization.
>>>>>>>>>>>>
>>>>>>>>>>>> Does anyone have any ideas on how to solve this w/o falling
>>>>>>>>>>>> back to "normal" synchronization?
>>>>>>>>>>>>
>>>>>>>>>>>> Mike
>>>>>>>>>>>>
>>>>>>>>>>>> Begin forwarded message:
>>>>>>>>>>>>
>>>>>>>>>>>>> From: "Yonik Seeley" <[hidden email]>
>>>>>>>>>>>>> Date: July 7, 2008 3:30:28 PM EDT
>>>>>>>>>>>>> To: [hidden email]
>>>>>>>>>>>>> Subject: Re: ThreadLocal in SegmentReader
>>>>>>>>>>>>> Reply-To: [hidden email]
>>>>>>>>>>>>>
>>>>>>>>>>>>> On Mon, Jul 7, 2008 at 2:43 PM, Michael McCandless
>>>>>>>>>>>>> <[hidden email]> wrote:
>>>>>>>>>>>>>> So now I'm confused: the SegmentReader itself should no
>>>>>>>>>>>>>> longer be reachable,
>>>>>>>>>>>>>> assuming you are not holding any references to your
>>>>>>>>>>>>>> IndexReader.
>>>>>>>>>>>>>>
>>>>>>>>>>>>>> Which means the ThreadLocal instance should no longer be
>>>>>>>>>>>>>> reachable.
>>>>>>>>>>>>>
>>>>>>>>>>>>> It will still be referenced from the Thread(s)
>>>>>>>>>>>>> ThreadLocalMap
>>>>>>>>>>>>> The key (the ThreadLocal) will be weakly referenced,  
>>>>>>>>>>>>> but the
>>>>>>>>>>>>> values
>>>>>>>>>>>>> (now stale) are strongly referenced and won't be actually
>>>>>>>>>>>>> removed
>>>>>>>>>>>>> until the table is resized (under the Java6 impl at  
>>>>>>>>>>>>> least).
>>>>>>>>>>>>> Nice huh?
>>>>>>>>>>>>>
>>>>>>>>>>>>> -Yonik
>>>>>>>>>>>>>
>>>>>>>>>>>>> ----------------------------------------------------------
>>>>>>>>>>>>> --
>>>>>>>>>>>>> ---------
>>>>>>>>>>>>> To unsubscribe, e-mail: java-user-
>>>>>>>>>>>>> [hidden email]
>>>>>>>>>>>>> For additional commands, e-mail: java-user-
>>>>>>>>>>>>> [hidden email]
>>>>>>>>>>>>>
>>>>>>>>>>>>
>>>>>>>>>>>>
>>>>>>>>>>>> -----------------------------------------------------------
>>>>>>>>>>>> --
>>>>>>>>>>>> --------
>>>>>>>>>>>> To unsubscribe, e-mail: java-dev-
>>>>>>>>>>>> [hidden email]
>>>>>>>>>>>> For additional commands, e-mail: java-dev-
>>>>>>>>>>>> [hidden email]
>>>>>>>>>>>>
>>>>>>>>>>>
>>>>>>>>>>>
>>>>>>>>>>> ------------------------------------------------------------
>>>>>>>>>>> --
>>>>>>>>>>> -------
>>>>>>>>>>> To unsubscribe, e-mail: java-dev-
>>>>>>>>>>> [hidden email]
>>>>>>>>>>> For additional commands, e-mail: java-dev-
>>>>>>>>>>> [hidden email]
>>>>>>>>>>>
>>>>>>>>>>
>>>>>>>>>>
>>>>>>>>>> -------------------------------------------------------------
>>>>>>>>>> --
>>>>>>>>>> ------
>>>>>>>>>> To unsubscribe, e-mail: java-dev-
>>>>>>>>>> [hidden email]
>>>>>>>>>> For additional commands, e-mail: java-dev-
>>>>>>>>>> [hidden email]
>>>>>>>>>>
>>>>>>>>>
>>>>>>>>>
>>>>>>>>> --------------------------------------------------------------
>>>>>>>>> --
>>>>>>>>> -----
>>>>>>>>> To unsubscribe, e-mail: [hidden email]
>>>>>>>>> For additional commands, e-mail: java-dev-
>>>>>>>>> [hidden email]
>>>>>>>>>
>>>>>>>>
>>>>>>>>
>>>>>>>> ---------------------------------------------------------------
>>>>>>>> --
>>>>>>>> ----
>>>>>>>> To unsubscribe, e-mail: [hidden email]
>>>>>>>> For additional commands, e-mail: java-dev-
>>>>>>>> [hidden email]
>>>>>>>>
>>>>>>>
>>>>>>
>>>>>
>>>>
>>>>
>>>>
>>>> -------------------------------------------------------------------
>>>> --
>>>> To unsubscribe, e-mail: [hidden email]
>>>> For additional commands, e-mail: [hidden email]
>>>>
>>>
>>> --
>>> View this message in context: http://www.nabble.com/Fwd%3A-
>>> ThreadLocal-in-SegmentReader-tp18326720p18416022.html
>>> Sent from the Lucene - Java Developer mailing list archive at
>>> Nabble.com.
>>>
>>>
>>> --------------------------------------------------------------------
>>> -
>>> To unsubscribe, e-mail: [hidden email]
>>> For additional commands, e-mail: [hidden email]
>>>
>>
>>
>>
>> ---------------------------------------------------------------------
>> To unsubscribe, e-mail: [hidden email]
>> For additional commands, e-mail: [hidden email]
>>
>>
>>
>
> --
> View this message in context: http://www.nabble.com/Fwd%3A- 
> ThreadLocal-in-SegmentReader-tp18326720p18416841.html
> Sent from the Lucene - Java Developer mailing list archive at  
> Nabble.com.
>
>
> ---------------------------------------------------------------------
> To unsubscribe, e-mail: [hidden email]
> For additional commands, e-mail: [hidden email]
>


---------------------------------------------------------------------
To unsubscribe, e-mail: [hidden email]
For additional commands, e-mail: [hidden email]

Reply | Threaded
Open this post in threaded view
|

Re: ThreadLocal in SegmentReader

Roman Puchkovskiy-2
Looks fine, and there's the only problem: all this does not work. Possibly, this is JVM issue, possibly, there's another reason, but when I reproduce the described situation with the web-app, its classloader cannot be unloaded, and the profiler shows that the cause is that ThreadLocal. When I apply the workaround (run code which uses Lucene in a Thread which dies shortly after executing the code), the problem disappears.

I'm not holding references to this ThreadLocal anywhere in my code, it's package-local field in package-local class of Lucene.

Robert Engels wrote
You are mistaken - Yonik's comment in that thread is correct  
(although it is not just at table resize - any time a ThreadLocal is  
added, and any time the ThreadLocal is not found in its direct hash  
it might clear entries).

The ThreadLocals map only has a WeakReference to the ThreadLocal, so  
if that is the only reference, it will be GC'd - eventually, and then  
it will be cleared as new ThreadLocals are created.

With a static reference, the thread can reference the ThreadLocal at  
any time, and thus the WeakReference will not be cleared.

If the object is VERY large, and new ThreadLocals are not created it  
could cause a problem, but I don't think that is the case with  
Lucene, as  the objects stored in ThreadLocals are designed to live  
for the life of the SegmentReader/IndexReader and thread.


On Jul 12, 2008, at 2:12 AM, Roman Puchkovskiy wrote:

>
> Well, possibly I'm mistaken, but it seems that this affects non-
> static fields
> too. Please see
> http://www.nabble.com/ThreadLocal-in-SegmentReader-to18306230.html 
> where the
> use case is described in the details.
> In short: it seems that the scope of ThreadLocals does not matter.  
> What
> really matters is that they are referenced by ThreadLocals map in  
> the thread
> which is still alive.
>
>
> Robert Engels wrote:
>>
>> This is only an issue for static ThreadLocals ...
>>
>> On Jul 11, 2008, at 11:32 PM, Roman Puchkovskiy wrote:
>>
>>>
>>> The problem here is not because ThreadLocal instances are not GC'd
>>> (they are
>>> GC'd, and your test shows this clearly).
>>> But even one instance which is not removed from its Thread is
>>> enough to
>>> prevent the classloader from being unloaded, and that's the problem.
>>>
>>>
>>> Michael McCandless-2 wrote:
>>>>
>>>> OK, I created a simple test to test this (attached).  The test just
>>>> runs 10 threads, each one creating a 100 KB byte array which is
>>>> stored
>>>> into a ThreadLocal, and then periodically the ThreadLocal is  
>>>> replaced
>>>> with a new one.  This is to test whether GC of a ThreadLocal, even
>>>> though the thread is still alive, in fact leads to GC of the  
>>>> objects
>>>> held in the ThreadLocal.
>>>>
>>>> Indeed on Sun JRE 1.4, 1.5 and 1.6 it appears that the objects  
>>>> are in
>>>> fact properly collected.
>>>>
>>>> So this is not a leak but rather a "delayed collection" issue.
>>>> Java's
>>>> GC is never guaranteed to be immediate, and apparently when using
>>>> ThreadLocals it's even less immediate than "normal".  In the  
>>>> original
>>>> issue, if other things create ThreadLocals, then eventually  
>>>> Lucene's
>>>> unreferenced ThreadLocals would be properly collected.
>>>>
>>>> So I think we continue to use non-static ThreadLocals in Lucene...
>>>>
>>>> Mike
>>>>
>>>>
>>>>
>>>>
>>>>
>>>>
>>>> robert engels wrote:
>>>>
>>>>> Once again, these are "static" thread locals. A completely  
>>>>> different
>>>>> issue. Since the object is available statically, the weak  
>>>>> reference
>>>>> cannot be cleared so stale entries will never be cleared as  
>>>>> long as
>>>>> the thread is alive.
>>>>>
>>>>> On Jul 9, 2008, at 4:46 PM, Adrian Tarau wrote:
>>>>>
>>>>>> Just a few examples of "problems" using ThreadLocals.
>>>>>>
>>>>>> http://opensource.atlassian.com/projects/hibernate/browse/ 
>>>>>> HHH-2481
>>>>>> http://www.theserverside.com/news/thread.tss?thread_id=41473
>>>>>>
>>>>>> Once again, I'm not pointing to Lucene SegmentReader as a "bad"
>>>>>> implementation, and maybe the current "problems" of ThreadLocals
>>>>>> are not a problem for SegmentReader but it seems safer to use
>>>>>> ThreadLocals to pass context information which is cleared when  
>>>>>> the
>>>>>> call exits instead of storing long-lived objects.
>>>>>>
>>>>>>
>>>>>> robert engels wrote:
>>>>>>>
>>>>>>> Aside from the pre-1.5 thread local "perceived leak", there  
>>>>>>> are no
>>>>>>> issues with ThreadLocals if used properly.
>>>>>>>
>>>>>>> There is no need for try/finally blocks, unless you MUST release
>>>>>>> resources immediately, usually this is not the case, which is  
>>>>>>> why
>>>>>>> a ThreadLocal is used in the first place.
>>>>>>>
>>>>>>> From the ThreadLocalMap javadoc...
>>>>>>>
>>>>>>>  /**
>>>>>>>      * ThreadLocalMap is a customized hash map suitable only for
>>>>>>>      * maintaining thread local values. No operations are  
>>>>>>> exported
>>>>>>>      * outside of the ThreadLocal class. The class is package
>>>>>>> private to
>>>>>>>      * allow declaration of fields in class Thread.  To help  
>>>>>>> deal
>>>>>>> with
>>>>>>>      * very large and long-lived usages, the hash table entries
>>>>>>> use
>>>>>>>      * WeakReferences for keys. However, since reference queues
>>>>>>> are not
>>>>>>>      * used, stale entries are guaranteed to be removed only  
>>>>>>> when
>>>>>>>      * the table starts running out of space.
>>>>>>>      */
>>>>>>>
>>>>>>> /**
>>>>>>>          * Heuristically scan some cells looking for stale
>>>>>>> entries.
>>>>>>>          * This is invoked when either a new element is  
>>>>>>> added, or
>>>>>>>          * another stale one has been expunged. It performs a
>>>>>>>          * logarithmic number of scans, as a balance between no
>>>>>>>          * scanning (fast but retains garbage) and a number of
>>>>>>> scans
>>>>>>>          * proportional to number of elements, that would  
>>>>>>> find all
>>>>>>>          * garbage but would cause some insertions to take O(n)
>>>>>>> time.
>>>>>>>          *
>>>>>>>          * @param i a position known NOT to hold a stale entry.
>>>>>>> The
>>>>>>>          * scan starts at the element after i.
>>>>>>>          *
>>>>>>>          * @param n scan control: <tt>log2(n)</tt> cells are
>>>>>>> scanned,
>>>>>>>          * unless a stale entry one is found, in which case
>>>>>>>          * <tt>log2(table.length)-1</tt> additional cells are
>>>>>>> scanned.
>>>>>>>          * When called from insertions, this parameter is the
>>>>>>> number
>>>>>>>          * of elements, but when from replaceStaleEntry, it  
>>>>>>> is the
>>>>>>>          * table length. (Note: all this could be changed to be
>>>>>>> either
>>>>>>>          * more or less aggressive by weighting n instead of  
>>>>>>> just
>>>>>>>          * using straight log n. But this version is simple,  
>>>>>>> fast,
>>>>>>> and
>>>>>>>          * seems to work well.)
>>>>>>>          *
>>>>>>>          * @return true if any stale entries have been removed.
>>>>>>>          */
>>>>>>>
>>>>>>>
>>>>>>> The instance ThreadLocals (and what the refer to) will be GC'd
>>>>>>> when the containing Object is GC'd.
>>>>>>>
>>>>>>> There IS NO MEMORY LEAK in ThreadLocal. If the ThreadLocal  
>>>>>>> refers
>>>>>>> to an object that has native resources (e.g. file handles),  
>>>>>>> it may
>>>>>>> not be released until other thread locals are created by the
>>>>>>> thread (or the thread terminates).
>>>>>>>
>>>>>>> You can avoid this "delay" by calling remove(), but in most
>>>>>>> applications it should never be necessary - unless a very  
>>>>>>> strange
>>>>>>> usage...
>>>>>>>
>>>>>>> On Jul 9, 2008, at 2:37 PM, Adrian Tarau wrote:
>>>>>>>
>>>>>>>> From what I know, storing objects in ThreadLocal is safe as  
>>>>>>>> long
>>>>>>>> as you release the object within a try {} finall {} block or
>>>>>>>> store objects which are independent of the rest of the code(no
>>>>>>>> dependencies).Otherwise it can get pretty tricky(memory leaks,
>>>>>>>> classloader problems) after awhile.
>>>>>>>>
>>>>>>>> It is pretty convenient to pass HTTP request information with a
>>>>>>>> ThreadLocal in a servlet(but you should cleanup the variable
>>>>>>>> before leaving the servlet) but I'm not sure how safe it is in
>>>>>>>> this case.
>>>>>>>>
>>>>>>>> robert engels wrote:
>>>>>>>>> Using synchronization is a poor/invalid substitute for thread
>>>>>>>>> locals in many cases.
>>>>>>>>>
>>>>>>>>> The point of the thread local in these referenced cases is too
>>>>>>>>> allow streaming reads on a file descriptor. if you use a  
>>>>>>>>> shared
>>>>>>>>> file descriptor/buffer you are going to continually invalidate
>>>>>>>>> the buffer.
>>>>>>>>>
>>>>>>>>> On Jul 8, 2008, at 5:12 AM, Michael McCandless wrote:
>>>>>>>>>
>>>>>>>>>>
>>>>>>>>>> Well ... SegmentReader uses ThreadLocal to hold a thread-
>>>>>>>>>> private instance of TermVectorsReader, to avoid synchronizing
>>>>>>>>>> per-document when loading term vectors.
>>>>>>>>>>
>>>>>>>>>> Clearing this ThreadLocal value per call to SegmentReader's
>>>>>>>>>> methods that load term vectors would defeat its purpose.
>>>>>>>>>>
>>>>>>>>>> Though, of course, we then synchronize on the underlying file
>>>>>>>>>> (when using FSDirectory), so perhaps we are really not saving
>>>>>>>>>> much by using ThreadLocal here.  But we are looking to relax
>>>>>>>>>> that low level synchronization with LUCENE-753.
>>>>>>>>>>
>>>>>>>>>> Maybe we could make our own ThreadLocal that just uses a
>>>>>>>>>> HashMap, which we'd have to synchronize on when getting the
>>>>>>>>>> per-
>>>>>>>>>> thread instances.  Or, go back to sharing a single
>>>>>>>>>> TermVectorsReader and synchronize per-document.
>>>>>>>>>>
>>>>>>>>>> Jason has suggested moving to a model where you ask the
>>>>>>>>>> IndexReader for an object that can return term vectors /  
>>>>>>>>>> stored
>>>>>>>>>> fields / etc, and then you interact with that many times to
>>>>>>>>>> retrieve each doc.  We could then synchronize only on
>>>>>>>>>> retrieving that object, and provide a thread-private  
>>>>>>>>>> instance.
>>>>>>>>>>
>>>>>>>>>> It seems like we should move away from using ThreadLocal in
>>>>>>>>>> Lucene and do "normal" synchronization instead.
>>>>>>>>>>
>>>>>>>>>> Mike
>>>>>>>>>>
>>>>>>>>>> Adrian Tarau wrote:
>>>>>>>>>>
>>>>>>>>>>> Usually ThreadLocal.remove() should be called at the end
>>>>>>>>>>> (in a
>>>>>>>>>>> finally block), before the current call leaves your code.
>>>>>>>>>>>
>>>>>>>>>>> Ex : if during searching ThreadLocal is used, every search
>>>>>>>>>>> (..)
>>>>>>>>>>> method should cleanup any ThreadLocal variables, or even
>>>>>>>>>>> deeper in the implementation. When the call leaves Lucene  
>>>>>>>>>>> any
>>>>>>>>>>> used ThreadLocal should be cleaned up.
>>>>>>>>>>>
>>>>>>>>>>> Michael McCandless wrote:
>>>>>>>>>>>>
>>>>>>>>>>>> ThreadLocal, which we use in several places in Lucene,  
>>>>>>>>>>>> causes
>>>>>>>>>>>> a leak in app servers because the classloader never fully
>>>>>>>>>>>> deallocates Lucene's classes because the ThreadLocal is
>>>>>>>>>>>> holding strong references.
>>>>>>>>>>>>
>>>>>>>>>>>> Yet, ThreadLocal is very convenient for avoiding
>>>>>>>>>>>> synchronization.
>>>>>>>>>>>>
>>>>>>>>>>>> Does anyone have any ideas on how to solve this w/o falling
>>>>>>>>>>>> back to "normal" synchronization?
>>>>>>>>>>>>
>>>>>>>>>>>> Mike
>>>>>>>>>>>>
>>>>>>>>>>>> Begin forwarded message:
>>>>>>>>>>>>
>>>>>>>>>>>>> From: "Yonik Seeley" <yonik@apache.org>
>>>>>>>>>>>>> Date: July 7, 2008 3:30:28 PM EDT
>>>>>>>>>>>>> To: java-user@lucene.apache.org
>>>>>>>>>>>>> Subject: Re: ThreadLocal in SegmentReader
>>>>>>>>>>>>> Reply-To: java-user@lucene.apache.org
>>>>>>>>>>>>>
>>>>>>>>>>>>> On Mon, Jul 7, 2008 at 2:43 PM, Michael McCandless
>>>>>>>>>>>>> <lucene@mikemccandless.com> wrote:
>>>>>>>>>>>>>> So now I'm confused: the SegmentReader itself should no
>>>>>>>>>>>>>> longer be reachable,
>>>>>>>>>>>>>> assuming you are not holding any references to your
>>>>>>>>>>>>>> IndexReader.
>>>>>>>>>>>>>>
>>>>>>>>>>>>>> Which means the ThreadLocal instance should no longer be
>>>>>>>>>>>>>> reachable.
>>>>>>>>>>>>>
>>>>>>>>>>>>> It will still be referenced from the Thread(s)
>>>>>>>>>>>>> ThreadLocalMap
>>>>>>>>>>>>> The key (the ThreadLocal) will be weakly referenced,  
>>>>>>>>>>>>> but the
>>>>>>>>>>>>> values
>>>>>>>>>>>>> (now stale) are strongly referenced and won't be actually
>>>>>>>>>>>>> removed
>>>>>>>>>>>>> until the table is resized (under the Java6 impl at  
>>>>>>>>>>>>> least).
>>>>>>>>>>>>> Nice huh?
>>>>>>>>>>>>>
>>>>>>>>>>>>> -Yonik
>>>>>>>>>>>>>
>>>>>>>>>>>>> ----------------------------------------------------------
>>>>>>>>>>>>> --
>>>>>>>>>>>>> ---------
>>>>>>>>>>>>> To unsubscribe, e-mail: java-user-
>>>>>>>>>>>>> unsubscribe@lucene.apache.org
>>>>>>>>>>>>> For additional commands, e-mail: java-user-
>>>>>>>>>>>>> help@lucene.apache.org
>>>>>>>>>>>>>
>>>>>>>>>>>>
>>>>>>>>>>>>
>>>>>>>>>>>> -----------------------------------------------------------
>>>>>>>>>>>> --
>>>>>>>>>>>> --------
>>>>>>>>>>>> To unsubscribe, e-mail: java-dev-
>>>>>>>>>>>> unsubscribe@lucene.apache.org
>>>>>>>>>>>> For additional commands, e-mail: java-dev-
>>>>>>>>>>>> help@lucene.apache.org
>>>>>>>>>>>>
>>>>>>>>>>>
>>>>>>>>>>>
>>>>>>>>>>> ------------------------------------------------------------
>>>>>>>>>>> --
>>>>>>>>>>> -------
>>>>>>>>>>> To unsubscribe, e-mail: java-dev-
>>>>>>>>>>> unsubscribe@lucene.apache.org
>>>>>>>>>>> For additional commands, e-mail: java-dev-
>>>>>>>>>>> help@lucene.apache.org
>>>>>>>>>>>
>>>>>>>>>>
>>>>>>>>>>
>>>>>>>>>> -------------------------------------------------------------
>>>>>>>>>> --
>>>>>>>>>> ------
>>>>>>>>>> To unsubscribe, e-mail: java-dev-
>>>>>>>>>> unsubscribe@lucene.apache.org
>>>>>>>>>> For additional commands, e-mail: java-dev-
>>>>>>>>>> help@lucene.apache.org
>>>>>>>>>>
>>>>>>>>>
>>>>>>>>>
>>>>>>>>> --------------------------------------------------------------
>>>>>>>>> --
>>>>>>>>> -----
>>>>>>>>> To unsubscribe, e-mail: java-dev-unsubscribe@lucene.apache.org
>>>>>>>>> For additional commands, e-mail: java-dev-
>>>>>>>>> help@lucene.apache.org
>>>>>>>>>
>>>>>>>>
>>>>>>>>
>>>>>>>> ---------------------------------------------------------------
>>>>>>>> --
>>>>>>>> ----
>>>>>>>> To unsubscribe, e-mail: java-dev-unsubscribe@lucene.apache.org
>>>>>>>> For additional commands, e-mail: java-dev-
>>>>>>>> help@lucene.apache.org
>>>>>>>>
>>>>>>>
>>>>>>
>>>>>
>>>>
>>>>
>>>>
>>>> -------------------------------------------------------------------
>>>> --
>>>> To unsubscribe, e-mail: java-dev-unsubscribe@lucene.apache.org
>>>> For additional commands, e-mail: java-dev-help@lucene.apache.org
>>>>
>>>
>>> --
>>> View this message in context: http://www.nabble.com/Fwd%3A-
>>> ThreadLocal-in-SegmentReader-tp18326720p18416022.html
>>> Sent from the Lucene - Java Developer mailing list archive at
>>> Nabble.com.
>>>
>>>
>>> --------------------------------------------------------------------
>>> -
>>> To unsubscribe, e-mail: java-dev-unsubscribe@lucene.apache.org
>>> For additional commands, e-mail: java-dev-help@lucene.apache.org
>>>
>>
>>
>>
>> ---------------------------------------------------------------------
>> To unsubscribe, e-mail: java-dev-unsubscribe@lucene.apache.org
>> For additional commands, e-mail: java-dev-help@lucene.apache.org
>>
>>
>>
>
> --
> View this message in context: http://www.nabble.com/Fwd%3A- 
> ThreadLocal-in-SegmentReader-tp18326720p18416841.html
> Sent from the Lucene - Java Developer mailing list archive at  
> Nabble.com.
>
>
> ---------------------------------------------------------------------
> To unsubscribe, e-mail: java-dev-unsubscribe@lucene.apache.org
> For additional commands, e-mail: java-dev-help@lucene.apache.org
>


---------------------------------------------------------------------
To unsubscribe, e-mail: java-dev-unsubscribe@lucene.apache.org
For additional commands, e-mail: java-dev-help@lucene.apache.org
Reply | Threaded
Open this post in threaded view
|

Re: ThreadLocal in SegmentReader

Roman Puchkovskiy-2
In reply to this post by Robert Engels
Oops, sent too early, excuse me :) One more to say: yes, Yonik is correct. Values remain strong-referenced until the stale values are removed from the table. Just a quote from the ThreadLocals' javadoc:
"However, since reference queues are not used, stale entries are guaranteed to be removed only when the table starts running out of space."
How do you suggest to force the removal of stale entries to the Lucene user?

Robert Engels wrote
You are mistaken - Yonik's comment in that thread is correct  
(although it is not just at table resize - any time a ThreadLocal is  
added, and any time the ThreadLocal is not found in its direct hash  
it might clear entries).

The ThreadLocals map only has a WeakReference to the ThreadLocal, so  
if that is the only reference, it will be GC'd - eventually, and then  
it will be cleared as new ThreadLocals are created.

With a static reference, the thread can reference the ThreadLocal at  
any time, and thus the WeakReference will not be cleared.

If the object is VERY large, and new ThreadLocals are not created it  
could cause a problem, but I don't think that is the case with  
Lucene, as  the objects stored in ThreadLocals are designed to live  
for the life of the SegmentReader/IndexReader and thread.


On Jul 12, 2008, at 2:12 AM, Roman Puchkovskiy wrote:

>
> Well, possibly I'm mistaken, but it seems that this affects non-
> static fields
> too. Please see
> http://www.nabble.com/ThreadLocal-in-SegmentReader-to18306230.html 
> where the
> use case is described in the details.
> In short: it seems that the scope of ThreadLocals does not matter.  
> What
> really matters is that they are referenced by ThreadLocals map in  
> the thread
> which is still alive.
>
>
> Robert Engels wrote:
>>
>> This is only an issue for static ThreadLocals ...
>>
>> On Jul 11, 2008, at 11:32 PM, Roman Puchkovskiy wrote:
>>
>>>
>>> The problem here is not because ThreadLocal instances are not GC'd
>>> (they are
>>> GC'd, and your test shows this clearly).
>>> But even one instance which is not removed from its Thread is
>>> enough to
>>> prevent the classloader from being unloaded, and that's the problem.
>>>
>>>
>>> Michael McCandless-2 wrote:
>>>>
>>>> OK, I created a simple test to test this (attached).  The test just
>>>> runs 10 threads, each one creating a 100 KB byte array which is
>>>> stored
>>>> into a ThreadLocal, and then periodically the ThreadLocal is  
>>>> replaced
>>>> with a new one.  This is to test whether GC of a ThreadLocal, even
>>>> though the thread is still alive, in fact leads to GC of the  
>>>> objects
>>>> held in the ThreadLocal.
>>>>
>>>> Indeed on Sun JRE 1.4, 1.5 and 1.6 it appears that the objects  
>>>> are in
>>>> fact properly collected.
>>>>
>>>> So this is not a leak but rather a "delayed collection" issue.
>>>> Java's
>>>> GC is never guaranteed to be immediate, and apparently when using
>>>> ThreadLocals it's even less immediate than "normal".  In the  
>>>> original
>>>> issue, if other things create ThreadLocals, then eventually  
>>>> Lucene's
>>>> unreferenced ThreadLocals would be properly collected.
>>>>
>>>> So I think we continue to use non-static ThreadLocals in Lucene...
>>>>
>>>> Mike
>>>>
>>>>
>>>>
>>>>
>>>>
>>>>
>>>> robert engels wrote:
>>>>
>>>>> Once again, these are "static" thread locals. A completely  
>>>>> different
>>>>> issue. Since the object is available statically, the weak  
>>>>> reference
>>>>> cannot be cleared so stale entries will never be cleared as  
>>>>> long as
>>>>> the thread is alive.
>>>>>
>>>>> On Jul 9, 2008, at 4:46 PM, Adrian Tarau wrote:
>>>>>
>>>>>> Just a few examples of "problems" using ThreadLocals.
>>>>>>
>>>>>> http://opensource.atlassian.com/projects/hibernate/browse/ 
>>>>>> HHH-2481
>>>>>> http://www.theserverside.com/news/thread.tss?thread_id=41473
>>>>>>
>>>>>> Once again, I'm not pointing to Lucene SegmentReader as a "bad"
>>>>>> implementation, and maybe the current "problems" of ThreadLocals
>>>>>> are not a problem for SegmentReader but it seems safer to use
>>>>>> ThreadLocals to pass context information which is cleared when  
>>>>>> the
>>>>>> call exits instead of storing long-lived objects.
>>>>>>
>>>>>>
>>>>>> robert engels wrote:
>>>>>>>
>>>>>>> Aside from the pre-1.5 thread local "perceived leak", there  
>>>>>>> are no
>>>>>>> issues with ThreadLocals if used properly.
>>>>>>>
>>>>>>> There is no need for try/finally blocks, unless you MUST release
>>>>>>> resources immediately, usually this is not the case, which is  
>>>>>>> why
>>>>>>> a ThreadLocal is used in the first place.
>>>>>>>
>>>>>>> From the ThreadLocalMap javadoc...
>>>>>>>
>>>>>>>  /**
>>>>>>>      * ThreadLocalMap is a customized hash map suitable only for
>>>>>>>      * maintaining thread local values. No operations are  
>>>>>>> exported
>>>>>>>      * outside of the ThreadLocal class. The class is package
>>>>>>> private to
>>>>>>>      * allow declaration of fields in class Thread.  To help  
>>>>>>> deal
>>>>>>> with
>>>>>>>      * very large and long-lived usages, the hash table entries
>>>>>>> use
>>>>>>>      * WeakReferences for keys. However, since reference queues
>>>>>>> are not
>>>>>>>      * used, stale entries are guaranteed to be removed only  
>>>>>>> when
>>>>>>>      * the table starts running out of space.
>>>>>>>      */
>>>>>>>
>>>>>>> /**
>>>>>>>          * Heuristically scan some c