Re: When are incompatible changes acceptable (HDFS-12990)

Previous Topic Next Topic
 
classic Classic list List threaded Threaded
6 messages Options
Reply | Threaded
Open this post in threaded view
|

Re: When are incompatible changes acceptable (HDFS-12990)

larry mccay-2
Question...

Can this be addressed in some way during or before upgrade that allows it
to only affect new installs?
Even a config based workaround prior to upgrade might make this a change
less disruptive.

If part of the upgrade process includes a step (maybe even a script) to set
the NN RPC port explicitly beforehand then it would allow existing
deployments and related clients to remain whole - otherwise it will uptake
the new default port.

Meta note: we shouldn't be so pedantic about policy that we can't back out
something that is considered a bug or even mistake.

On Mon, Jan 8, 2018 at 9:17 PM, Aaron T. Myers <[hidden email]> wrote:

> Hello all,
>
> Over in HDFS-12990 <https://issues.apache.org/jira/browse/HDFS-12990>[1],
> we're having some discussion about whether or not it's ever acceptable to
> make an incompatible change in a minor or dot release. In general this is
> of course undesirable and should be avoided in almost all cases. However, I
> believe that each instance of someone desiring to make an incompatible
> change should be evaluated on a case-by-case basis to consider the costs
> and benefits of making that change. For example, I believe that we've
> historically made incompatible changes in minor or dot releases which would
> break older clients for security reasons.
>
> In this particular case linked above,  I believe that given that Hadoop
> 3.0.0 was just released, and thus very few folks are likely to have
> deployed it, it would benefit a large number of existing deployments and
> downstream applications to change the default NN RPC port number back to
> what it was in all previously-released versions of Apache Hadoop. I'd like
> to make this change in 3.0.1, and there is no question that doing so would
> should be considered an incompatible change between 3.0.0 and 3.0.1.
> However, I believe this incompatible change is warranted given the
> circumstances.
>
> Would like to hear others' thoughts on this.
>
> Thanks,
> Aaron
>
> [1] For some background, it used to be the case that many of Hadoop's
> default service ports were in the ephemeral range. This could potentially
> cause a service to fail to start up on a given host if some other process
> had happened to have already bound to said port. As part of that effort, we
> also changed the default NN RPC port from 8020 to 9820. Even though 8020
> wasn't in the ephemeral range, we moved it to 9820 to be close to the new
> range of the rest of the ports. At the time this change was made, though, I
> and others didn't realize the substantial downsides that doing so would
> introduce, for example the Hive metastore will put full HDFS paths
> including the port into its database, which can be a substantial upgrade
> headache.
>
Reply | Threaded
Open this post in threaded view
|

Re: When are incompatible changes acceptable (HDFS-12990)

Daniel Templeton
The intent of the compat guidelines is to prevent developers from making
incompatible "improvements" at inconvenient times.  The guidelines offer
some wiggle room for the cases where something truly broken is being
fixed, especially for the sake of compatibility.  I would say that in
this case, making the change is the right thing to do, but there needs
to be a plan for how to deal with the fact that the 3.0.0 release will
forever be an odd outlier with respect to NN ports.

Daniel

On 1/8/18 8:28 PM, Aaron T. Myers wrote:

> Thanks a lot for the response, Larry. Comments inline.
>
> On Mon, Jan 8, 2018 at 6:44 PM, larry mccay <[hidden email]> wrote:
>
>> Question...
>>
>> Can this be addressed in some way during or before upgrade that allows it
>> to only affect new installs?
>> Even a config based workaround prior to upgrade might make this a change
>> less disruptive.
>>
>> If part of the upgrade process includes a step (maybe even a script) to
>> set the NN RPC port explicitly beforehand then it would allow existing
>> deployments and related clients to remain whole - otherwise it will uptake
>> the new default port.
>>
> Perhaps something like this could be done, but I think there are downsides
> to anything like this. For example, I'm sure there are plenty of
> applications written on top of Hadoop that have tests which hard-code the
> port number. Nothing we do in a setup script will help here. If we don't
> change the default port back to what it was, these tests will likely all
> have to be updated.
>
>
>> Meta note: we shouldn't be so pedantic about policy that we can't back out
>> something that is considered a bug or even mistake.
>>
> This is my bigger point. Rigidly adhering to the compat guidelines in this
> instance helps almost no one, while hurting many folks.
>
> We basically made a mistake when we decided to change the default NN port
> with little upside, even between major versions. We discovered this very
> quickly, and we have an opportunity to fix it now and in so doing likely
> disrupt very, very few users and downstream applications. If we don't
> change it, we'll be causing difficulty for our users, downstream
> developers, and ourselves, potentially for years.
>
> Best,
> Aaron
>


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

Reply | Threaded
Open this post in threaded view
|

Re: When are incompatible changes acceptable (HDFS-12990)

Eric Yang-4
In reply to this post by larry mccay-2
See comments inline.

Regards,
Eric

From: <[hidden email]> on behalf of "Aaron T. Myers" <[hidden email]>
Date: Wednesday, January 10, 2018 at 9:21 AM
To: Eric Yang <[hidden email]>
Cc: Chris Douglas <[hidden email]>, larry mccay <[hidden email]>, Hadoop Common <[hidden email]>
Subject: Re: When are incompatible changes acceptable (HDFS-12990)

Hey Eric,

Comments inline.

On Wed, Jan 10, 2018 at 9:06 AM, Eric Yang <[hidden email]<mailto:[hidden email]>> wrote:
Hi Aaron,

Correct me if I am wrong, the port change is only required when making a new cluster due to the default value.  Existing cluster does not need to make the switch, if Hadoop configuration contains user defined number.

Certainly true that a port change isn't required, and if it's already properly being set everywhere throughout a deployment (i.e. all clients, client applications, scripts, etc.) it won't be an issue. I'm most worried about *client* configs throughout a large deployment, which would be difficult (impossible?) to coordinate an update to. Entirely possible, if not likely, that many clients are inadvertently relying on the default port, so when they start using the updated software they'll break because of the default port change.

Ambari, and Cloudera Manager are already handling user defined ports correctly.  Some QA tools may need to change, but it is a good exercise to run on non-standard port.

Sites which are using Ambari or Cloudera Manager are more likely to work, but again, I worry about client configs and other places that might have hard-coded the port number, e.g. in Hive or in scripts.

I will also say that Hadoop users which are *not* using Ambari or CM should be considered as well. Sites like this are perhaps the most likely to break because of this change.

Agree.
I gave my vote to keep the setting, and fully respect the community’s decision in this matter.

Thanks, Eric. I understand your argument to be that changing this default port might not be so bad, but it also sounds like you wouldn't object if others conclude that it's best to change it back. Is that right?

The decision is in the hands of Apache Hadoop community.  This is not a decision that can be made by one individual, one company or another.  Let’s start a voting thread to make sure that the decision was made by Hadoop community correctly.

Best,
Aaron



Regards,
Eric

From: <[hidden email]<mailto:[hidden email]>> on behalf of "Aaron T. Myers" <[hidden email]<mailto:[hidden email]>>
Date: Tuesday, January 9, 2018 at 9:22 PM
To: Eric Yang <[hidden email]<mailto:[hidden email]>>
Cc: Chris Douglas <[hidden email]<mailto:[hidden email]>>, larry mccay <[hidden email]<mailto:[hidden email]>>, Hadoop Common <[hidden email]<mailto:[hidden email]>>
Subject: Re: When are incompatible changes acceptable (HDFS-12990)

On Tue, Jan 9, 2018 at 3:15 PM, Eric Yang <[hidden email]<mailto:[hidden email]>> wrote:
While I agree the original port change was unnecessary, I don’t think Hadoop NN port change is a bad thing.

I worked for a Hadoop distro that NN RPC port was default to port 9000.  When we migrate from BigInsights to IOP and now to HDP, we have to move customer Hive metadata to new NN RPC port.  It only took one developer (myself) to write the tool for the migration.  The incurring workload is not as bad as most people anticipated because Hadoop depends on configuration file for referencing namenode.  Most of the code can work transparently.  It helped to harden the downstream testing tools to be more robust.

While there are of course ways to deal with this, the question really should be whether or not it's a desirable thing to do to our users.


We will never know how many people are actively working on Hadoop 3.0.0.  Perhaps, couple hundred developers or thousands.

You're right that we can't know for sure, but I strongly suspect that this is a substantial overestimate. Given how conservative Hadoop operators tend to be, I view it as exceptionally unlikely that many deployments have been created on or upgraded to Hadoop 3.0.0 since it was released less than a month ago.

Further, I hope you'll agree that the number of users/developers/deployments/applications which are currently on Hadoop 2.x is *vastly* greater than anyone who might have jumped on Hadoop 3.0.0 so quickly. When all of those users upgrade to any 3.x version, they will encounter this needless incompatible change and be forced to work around it.

I think the switch back may have saved few developers work, but there could be more people getting impacted at unexpected minor release change in the future.  I recommend keeping current values to avoid rule bending and future frustrations.

That we allow this incompatible change now does not mean that we are categorically allowing more incompatible changes in the future. My point is that we should in all instances evaluate the merit of any incompatible change on a case-by-case basis. This is not an exceptional circumstance - we've made incompatible changes in the past when appropriate, e.g. breaking some clients to address a security issue. I and others believe that in this case the benefits greatly outweigh the downsides of changing this back to what it has always been.

Best,
Aaron


Regards,
Eric

On 1/9/18, 11:21 AM, "Chris Douglas" <[hidden email]<mailto:[hidden email]>> wrote:

    Particularly since 9820 isn't in the contiguous range of ports in
    HDFS-9427, is there any value in this change?

    Let's change it back to prevent the disruption to users, but
    downstream projects should treat this as a bug in their tests. Please
    open JIRAs in affected projects. -C


    On Tue, Jan 9, 2018 at 5:18 AM, larry mccay <[hidden email]<mailto:[hidden email]>> wrote:
    > On Mon, Jan 8, 2018 at 11:28 PM, Aaron T. Myers <[hidden email]<mailto:[hidden email]>> wrote:
    >
    >> Thanks a lot for the response, Larry. Comments inline.
    >>
    >> On Mon, Jan 8, 2018 at 6:44 PM, larry mccay <[hidden email]<mailto:[hidden email]>> wrote:
    >>
    >>> Question...
    >>>
    >>> Can this be addressed in some way during or before upgrade that allows it
    >>> to only affect new installs?
    >>> Even a config based workaround prior to upgrade might make this a change
    >>> less disruptive.
    >>>
    >>> If part of the upgrade process includes a step (maybe even a script) to
    >>> set the NN RPC port explicitly beforehand then it would allow existing
    >>> deployments and related clients to remain whole - otherwise it will uptake
    >>> the new default port.
    >>>
    >>
    >> Perhaps something like this could be done, but I think there are downsides
    >> to anything like this. For example, I'm sure there are plenty of
    >> applications written on top of Hadoop that have tests which hard-code the
    >> port number. Nothing we do in a setup script will help here. If we don't
    >> change the default port back to what it was, these tests will likely all
    >> have to be updated.
    >>
    >>
    >
    > I may not have made my point clear enough.
    > What I meant to say is to fix the default port but direct folks to
    > explicitly set the port they are using in a deployment (the current
    > default) so that it doesn't change out from under them - unless they are
    > fine with it changing.
    >
    >
    >>
    >>> Meta note: we shouldn't be so pedantic about policy that we can't back
    >>> out something that is considered a bug or even mistake.
    >>>
    >>
    >> This is my bigger point. Rigidly adhering to the compat guidelines in this
    >> instance helps almost no one, while hurting many folks.
    >>
    >> We basically made a mistake when we decided to change the default NN port
    >> with little upside, even between major versions. We discovered this very
    >> quickly, and we have an opportunity to fix it now and in so doing likely
    >> disrupt very, very few users and downstream applications. If we don't
    >> change it, we'll be causing difficulty for our users, downstream
    >> developers, and ourselves, potentially for years.
    >>
    >
    > Agreed.
    >
    >
    >>
    >> Best,
    >> Aaron
    >>

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



Reply | Threaded
Open this post in threaded view
|

Re: When are incompatible changes acceptable (HDFS-12990)

larry mccay-2
In reply to this post by larry mccay-2
On Wed, Jan 10, 2018 at 1:34 PM, Daryn Sharp <[hidden email]> wrote:

> I fully agree the port changes should be reverted.  Although
> "incompatible", the potential impact to existing 2.x deploys is huge.  I'd
> rather inconvenience 3.0 deploys that compromise <1% customers.  An
> incompatible change to revert an incompatible change is called
> compatibility.
>

+1


>
> Most importantly, consider that there is no good upgrade path existing
> deploys, esp. large and/or multi-cluster environments.  It’s only feasible
> for first-time deploys or simple single-cluster upgrades willing to take
> downtime.  Let's consider a few reasons why:
>
>
> 1. RU is completely broken.  Running jobs will fail.  If MR on hdfs
> bundles the configs, there's no way to transparently coordinate the switch
> to the new bundle with the port changed.  Job submissions will fail.
>
>
> 2. Users generally do not add the rpc port number to uris so unless their
> configs are updated they will contact the wrong port.  Seamlessly
> coordinating the conf change without massive failures is impossible.
>
>
> 3. Even if client confs are updated, they will break in a multi-cluster
> env with NNs using different ports.  Users/services will be forced to add
> the port.  The cited hive "issue" is not a bug since it's the only way to
> work in a multi-port env.
>
>
> 4. Coordinating the port add/change of uris is systems everywhere (you
> know something will be missed), updating of confs, restarting all services,
> requiring customers to redeploy their workflows in sync with the NN
> upgrade, will cause mass disruption and downtime that will be unacceptable
> for production environments.
>
>
> This is a solution to a non-existent problem.  Ports can be bound by
> multiple processes but only 1 can listen.  Maybe multiple listeners is an
> issue for compute nodes but not responsibly managed service nodes.  Ie. Who
> runs arbitrary services on the NNs that bind to random ports?  Besides, the
> default port is and was ephemeral so it solved nothing.
>
>
> This either standardizes ports to a particular customer's ports or is a
> poorly thought out whim.  In either case, the needs of the many outweigh
> the needs of the few/none (3.0 users).  The only logical conclusion is
> revert.  If a particular site wants to change default ports and deal with
> the massive fallout, they can explicitly change the ports themselves.
>
>
> Daryn
>
> On Tue, Jan 9, 2018 at 11:22 PM, Aaron T. Myers <[hidden email]> wrote:
>
>> On Tue, Jan 9, 2018 at 3:15 PM, Eric Yang <[hidden email]> wrote:
>>
>> > While I agree the original port change was unnecessary, I don’t think
>> > Hadoop NN port change is a bad thing.
>> >
>> > I worked for a Hadoop distro that NN RPC port was default to port 9000.
>> > When we migrate from BigInsights to IOP and now to HDP, we have to move
>> > customer Hive metadata to new NN RPC port.  It only took one developer
>> > (myself) to write the tool for the migration.  The incurring workload is
>> > not as bad as most people anticipated because Hadoop depends on
>> > configuration file for referencing namenode.  Most of the code can work
>> > transparently.  It helped to harden the downstream testing tools to be
>> more
>> > robust.
>> >
>>
>> While there are of course ways to deal with this, the question really
>> should be whether or not it's a desirable thing to do to our users.
>>
>>
>> >
>> > We will never know how many people are actively working on Hadoop 3.0.0.
>> > Perhaps, couple hundred developers or thousands.
>>
>>
>> You're right that we can't know for sure, but I strongly suspect that this
>> is a substantial overestimate. Given how conservative Hadoop operators
>> tend
>> to be, I view it as exceptionally unlikely that many deployments have been
>> created on or upgraded to Hadoop 3.0.0 since it was released less than a
>> month ago.
>>
>> Further, I hope you'll agree that the number of
>> users/developers/deployments/applications which are currently on Hadoop
>> 2.x
>> is *vastly* greater than anyone who might have jumped on Hadoop 3.0.0 so
>> quickly. When all of those users upgrade to any 3.x version, they will
>> encounter this needless incompatible change and be forced to work around
>> it.
>>
>>
>> > I think the switch back may have saved few developers work, but there
>> > could be more people getting impacted at unexpected minor release
>> change in
>> > the future.  I recommend keeping current values to avoid rule bending
>> and
>> > future frustrations.
>> >
>>
>> That we allow this incompatible change now does not mean that we are
>> categorically allowing more incompatible changes in the future. My point
>> is
>> that we should in all instances evaluate the merit of any incompatible
>> change on a case-by-case basis. This is not an exceptional circumstance -
>> we've made incompatible changes in the past when appropriate, e.g.
>> breaking
>> some clients to address a security issue. I and others believe that in
>> this
>> case the benefits greatly outweigh the downsides of changing this back to
>> what it has always been.
>>
>> Best,
>> Aaron
>>
>>
>> >
>> > Regards,
>> > Eric
>> >
>> > On 1/9/18, 11:21 AM, "Chris Douglas" <[hidden email]> wrote:
>> >
>> >     Particularly since 9820 isn't in the contiguous range of ports in
>> >     HDFS-9427, is there any value in this change?
>> >
>> >     Let's change it back to prevent the disruption to users, but
>> >     downstream projects should treat this as a bug in their tests.
>> Please
>> >     open JIRAs in affected projects. -C
>> >
>> >
>> >     On Tue, Jan 9, 2018 at 5:18 AM, larry mccay <[hidden email]>
>> wrote:
>> >     > On Mon, Jan 8, 2018 at 11:28 PM, Aaron T. Myers <[hidden email]>
>> > wrote:
>> >     >
>> >     >> Thanks a lot for the response, Larry. Comments inline.
>> >     >>
>> >     >> On Mon, Jan 8, 2018 at 6:44 PM, larry mccay <[hidden email]>
>> > wrote:
>> >     >>
>> >     >>> Question...
>> >     >>>
>> >     >>> Can this be addressed in some way during or before upgrade that
>> > allows it
>> >     >>> to only affect new installs?
>> >     >>> Even a config based workaround prior to upgrade might make this
>> a
>> > change
>> >     >>> less disruptive.
>> >     >>>
>> >     >>> If part of the upgrade process includes a step (maybe even a
>> > script) to
>> >     >>> set the NN RPC port explicitly beforehand then it would allow
>> > existing
>> >     >>> deployments and related clients to remain whole - otherwise it
>> > will uptake
>> >     >>> the new default port.
>> >     >>>
>> >     >>
>> >     >> Perhaps something like this could be done, but I think there are
>> > downsides
>> >     >> to anything like this. For example, I'm sure there are plenty of
>> >     >> applications written on top of Hadoop that have tests which
>> > hard-code the
>> >     >> port number. Nothing we do in a setup script will help here. If
>> we
>> > don't
>> >     >> change the default port back to what it was, these tests will
>> > likely all
>> >     >> have to be updated.
>> >     >>
>> >     >>
>> >     >
>> >     > I may not have made my point clear enough.
>> >     > What I meant to say is to fix the default port but direct folks to
>> >     > explicitly set the port they are using in a deployment (the
>> current
>> >     > default) so that it doesn't change out from under them - unless
>> they
>> > are
>> >     > fine with it changing.
>> >     >
>> >     >
>> >     >>
>> >     >>> Meta note: we shouldn't be so pedantic about policy that we
>> can't
>> > back
>> >     >>> out something that is considered a bug or even mistake.
>> >     >>>
>> >     >>
>> >     >> This is my bigger point. Rigidly adhering to the compat
>> guidelines
>> > in this
>> >     >> instance helps almost no one, while hurting many folks.
>> >     >>
>> >     >> We basically made a mistake when we decided to change the default
>> > NN port
>> >     >> with little upside, even between major versions. We discovered
>> this
>> > very
>> >     >> quickly, and we have an opportunity to fix it now and in so doing
>> > likely
>> >     >> disrupt very, very few users and downstream applications. If we
>> > don't
>> >     >> change it, we'll be causing difficulty for our users, downstream
>> >     >> developers, and ourselves, potentially for years.
>> >     >>
>> >     >
>> >     > Agreed.
>> >     >
>> >     >
>> >     >>
>> >     >> Best,
>> >     >> Aaron
>> >     >>
>> >
>> >     -----------------------------------------------------------
>> ----------
>> >     To unsubscribe, e-mail: [hidden email]
>> >     For additional commands, e-mail: [hidden email]
>> >
>> >
>> >
>> >
>>
>
>
Reply | Threaded
Open this post in threaded view
|

Re: When are incompatible changes acceptable (HDFS-12990)

Eric Yang-4
The fix in HDFS-9427 can potentially bring in new customers because less chance for new comer to encountering “port already in use” problem.  If we make change according to HDFS-12990, then this incompatible change does not make incompatible change compatible.  Other ports are not reverted according to HDFS-12990.  User will encounter the bad taste in the mouth that HDFS-9427 attempt to solve.  Please do consider both negative side effects of reverting as well as incompatible minor release change.  Thanks

Regards,
Eric

From: larry mccay <[hidden email]>
Date: Wednesday, January 10, 2018 at 10:53 AM
To: Daryn Sharp <[hidden email]>
Cc: "Aaron T. Myers" <[hidden email]>, Eric Yang <[hidden email]>, Chris Douglas <[hidden email]>, Hadoop Common <[hidden email]>
Subject: Re: When are incompatible changes acceptable (HDFS-12990)

On Wed, Jan 10, 2018 at 1:34 PM, Daryn Sharp <[hidden email]<mailto:[hidden email]>> wrote:

I fully agree the port changes should be reverted.  Although "incompatible", the potential impact to existing 2.x deploys is huge.  I'd rather inconvenience 3.0 deploys that compromise <1% customers.  An incompatible change to revert an incompatible change is called compatibility.

+1




Most importantly, consider that there is no good upgrade path existing deploys, esp. large and/or multi-cluster environments.  It’s only feasible for first-time deploys or simple single-cluster upgrades willing to take downtime.  Let's consider a few reasons why:



1. RU is completely broken.  Running jobs will fail.  If MR on hdfs bundles the configs, there's no way to transparently coordinate the switch to the new bundle with the port changed.  Job submissions will fail.



2. Users generally do not add the rpc port number to uris so unless their configs are updated they will contact the wrong port.  Seamlessly coordinating the conf change without massive failures is impossible.



3. Even if client confs are updated, they will break in a multi-cluster env with NNs using different ports.  Users/services will be forced to add the port.  The cited hive "issue" is not a bug since it's the only way to work in a multi-port env.



4. Coordinating the port add/change of uris is systems everywhere (you know something will be missed), updating of confs, restarting all services, requiring customers to redeploy their workflows in sync with the NN upgrade, will cause mass disruption and downtime that will be unacceptable for production environments.



This is a solution to a non-existent problem.  Ports can be bound by multiple processes but only 1 can listen.  Maybe multiple listeners is an issue for compute nodes but not responsibly managed service nodes.  Ie. Who runs arbitrary services on the NNs that bind to random ports?  Besides, the default port is and was ephemeral so it solved nothing.



This either standardizes ports to a particular customer's ports or is a poorly thought out whim.  In either case, the needs of the many outweigh the needs of the few/none (3.0 users).  The only logical conclusion is revert.  If a particular site wants to change default ports and deal with the massive fallout, they can explicitly change the ports themselves.



Daryn

On Tue, Jan 9, 2018 at 11:22 PM, Aaron T. Myers <[hidden email]<mailto:[hidden email]>> wrote:
On Tue, Jan 9, 2018 at 3:15 PM, Eric Yang <[hidden email]<mailto:[hidden email]>> wrote:

> While I agree the original port change was unnecessary, I don’t think
> Hadoop NN port change is a bad thing.
>
> I worked for a Hadoop distro that NN RPC port was default to port 9000.
> When we migrate from BigInsights to IOP and now to HDP, we have to move
> customer Hive metadata to new NN RPC port.  It only took one developer
> (myself) to write the tool for the migration.  The incurring workload is
> not as bad as most people anticipated because Hadoop depends on
> configuration file for referencing namenode.  Most of the code can work
> transparently.  It helped to harden the downstream testing tools to be more
> robust.
>

While there are of course ways to deal with this, the question really
should be whether or not it's a desirable thing to do to our users.


>
> We will never know how many people are actively working on Hadoop 3.0.0.
> Perhaps, couple hundred developers or thousands.


You're right that we can't know for sure, but I strongly suspect that this
is a substantial overestimate. Given how conservative Hadoop operators tend
to be, I view it as exceptionally unlikely that many deployments have been
created on or upgraded to Hadoop 3.0.0 since it was released less than a
month ago.

Further, I hope you'll agree that the number of
users/developers/deployments/applications which are currently on Hadoop 2.x
is *vastly* greater than anyone who might have jumped on Hadoop 3.0.0 so
quickly. When all of those users upgrade to any 3.x version, they will
encounter this needless incompatible change and be forced to work around it.


> I think the switch back may have saved few developers work, but there
> could be more people getting impacted at unexpected minor release change in
> the future.  I recommend keeping current values to avoid rule bending and
> future frustrations.
>

That we allow this incompatible change now does not mean that we are
categorically allowing more incompatible changes in the future. My point is
that we should in all instances evaluate the merit of any incompatible
change on a case-by-case basis. This is not an exceptional circumstance -
we've made incompatible changes in the past when appropriate, e.g. breaking
some clients to address a security issue. I and others believe that in this
case the benefits greatly outweigh the downsides of changing this back to
what it has always been.

Best,
Aaron


>
> Regards,
> Eric
>
> On 1/9/18, 11:21 AM, "Chris Douglas" <[hidden email]<mailto:[hidden email]>> wrote:
>
>     Particularly since 9820 isn't in the contiguous range of ports in
>     HDFS-9427, is there any value in this change?
>
>     Let's change it back to prevent the disruption to users, but
>     downstream projects should treat this as a bug in their tests. Please
>     open JIRAs in affected projects. -C
>
>
>     On Tue, Jan 9, 2018 at 5:18 AM, larry mccay <[hidden email]<mailto:[hidden email]>> wrote:
>     > On Mon, Jan 8, 2018 at 11:28 PM, Aaron T. Myers <[hidden email]<mailto:[hidden email]>>
> wrote:
>     >
>     >> Thanks a lot for the response, Larry. Comments inline.
>     >>
>     >> On Mon, Jan 8, 2018 at 6:44 PM, larry mccay <[hidden email]<mailto:[hidden email]>>
> wrote:
>     >>
>     >>> Question...
>     >>>
>     >>> Can this be addressed in some way during or before upgrade that
> allows it
>     >>> to only affect new installs?
>     >>> Even a config based workaround prior to upgrade might make this a
> change
>     >>> less disruptive.
>     >>>
>     >>> If part of the upgrade process includes a step (maybe even a
> script) to
>     >>> set the NN RPC port explicitly beforehand then it would allow
> existing
>     >>> deployments and related clients to remain whole - otherwise it
> will uptake
>     >>> the new default port.
>     >>>
>     >>
>     >> Perhaps something like this could be done, but I think there are
> downsides
>     >> to anything like this. For example, I'm sure there are plenty of
>     >> applications written on top of Hadoop that have tests which
> hard-code the
>     >> port number. Nothing we do in a setup script will help here. If we
> don't
>     >> change the default port back to what it was, these tests will
> likely all
>     >> have to be updated.
>     >>
>     >>
>     >
>     > I may not have made my point clear enough.
>     > What I meant to say is to fix the default port but direct folks to
>     > explicitly set the port they are using in a deployment (the current
>     > default) so that it doesn't change out from under them - unless they
> are
>     > fine with it changing.
>     >
>     >
>     >>
>     >>> Meta note: we shouldn't be so pedantic about policy that we can't
> back
>     >>> out something that is considered a bug or even mistake.
>     >>>
>     >>
>     >> This is my bigger point. Rigidly adhering to the compat guidelines
> in this
>     >> instance helps almost no one, while hurting many folks.
>     >>
>     >> We basically made a mistake when we decided to change the default
> NN port
>     >> with little upside, even between major versions. We discovered this
> very
>     >> quickly, and we have an opportunity to fix it now and in so doing
> likely
>     >> disrupt very, very few users and downstream applications. If we
> don't
>     >> change it, we'll be causing difficulty for our users, downstream
>     >> developers, and ourselves, potentially for years.
>     >>
>     >
>     > Agreed.
>     >
>     >
>     >>
>     >> Best,
>     >> Aaron
>     >>
>
>     ---------------------------------------------------------------------
>     To unsubscribe, e-mail: [hidden email]<mailto:[hidden email]>
>     For additional commands, e-mail: [hidden email]<mailto:[hidden email]>
>
>
>
>


Reply | Threaded
Open this post in threaded view
|

Re: When are incompatible changes acceptable (HDFS-12990)

Aaron T. Myers-2
Yes indeed, that's the proposal being discussed on HDFS-12990 - just to
revert the default NN RPC port change, and none of the other port changes.
The other default port changes actually do have some technical benefit, and
I believe are far less likely to be embedded in databases, scripts, tests,
etc. in real deployments.

Best,
Aaron

On Thu, Jan 11, 2018 at 11:42 AM, larry mccay <[hidden email]> wrote:

> No, the proposal was to only fix the NN port change - as I understood it.
>
> On Thu, Jan 11, 2018 at 2:01 PM, Eric Yang <[hidden email]> wrote:
>
> > If I am reading this correctly, Daryn and Larry are in favor of complete
> > revert instead of namenode only.  Please charm in if I am wrong.  This is
> > the reason that I try to explore each perspective to understand the cost
> of
> > each options.  It appears that we have a fragment of opinions, and only
> one
> > choice will serve the need of majority of the community.  It would be
> good
> > for a PMC to call the vote at reasonable pace to address this issue to
> > reduce the pain point from either side of oppositions.
> >
> >
> >
> > Regards,
> >
> > Eric
> >
> >
> >
> > *From: *Chris Douglas <[hidden email]>
> > *Date: *Wednesday, January 10, 2018 at 7:36 PM
> > *To: *Eric Yang <[hidden email]>
> > *Cc: *"Aaron T. Myers" <[hidden email]>, Daryn Sharp <[hidden email]>,
> > Hadoop Common <[hidden email]>, larry mccay <
> > [hidden email]>
> >
> > *Subject: *Re: When are incompatible changes acceptable (HDFS-12990)
> >
> >
> >
> > Isn't this limited to reverting the 8020 -> 9820 change? -C
> >
> >
> >
> > On Wed, Jan 10, 2018 at 6:13 PM Eric Yang <[hidden email]> wrote:
> >
> > The fix in HDFS-9427 can potentially bring in new customers because less
> > chance for new comer to encountering “port already in use” problem.  If
> we
> > make change according to HDFS-12990, then this incompatible change does
> not
> > make incompatible change compatible.  Other ports are not reverted
> > according to HDFS-12990.  User will encounter the bad taste in the mouth
> > that HDFS-9427 attempt to solve.  Please do consider both negative side
> > effects of reverting as well as incompatible minor release change.
> Thanks
> >
> > Regards,
> > Eric
> >
> > From: larry mccay <[hidden email]>
> > Date: Wednesday, January 10, 2018 at 10:53 AM
> > To: Daryn Sharp <[hidden email]>
> > Cc: "Aaron T. Myers" <[hidden email]>, Eric Yang <[hidden email]
> >,
> > Chris Douglas <[hidden email]>, Hadoop Common <
> > [hidden email]>
> > Subject: Re: When are incompatible changes acceptable (HDFS-12990)
> >
> > On Wed, Jan 10, 2018 at 1:34 PM, Daryn Sharp <[hidden email]<mailto:
> daryn@
> > oath.com>> wrote:
> >
> > I fully agree the port changes should be reverted.  Although
> > "incompatible", the potential impact to existing 2.x deploys is huge.
> I'd
> > rather inconvenience 3.0 deploys that compromise <1% customers.  An
> > incompatible change to revert an incompatible change is called
> > compatibility.
> >
> > +1
> >
> >
> >
> >
> > Most importantly, consider that there is no good upgrade path existing
> > deploys, esp. large and/or multi-cluster environments.  It’s only
> feasible
> > for first-time deploys or simple single-cluster upgrades willing to take
> > downtime.  Let's consider a few reasons why:
> >
> >
> >
> > 1. RU is completely broken.  Running jobs will fail.  If MR on hdfs
> > bundles the configs, there's no way to transparently coordinate the
> switch
> > to the new bundle with the port changed.  Job submissions will fail.
> >
> >
> >
> > 2. Users generally do not add the rpc port number to uris so unless their
> > configs are updated they will contact the wrong port.  Seamlessly
> > coordinating the conf change without massive failures is impossible.
> >
> >
> >
> > 3. Even if client confs are updated, they will break in a multi-cluster
> > env with NNs using different ports.  Users/services will be forced to add
> > the port.  The cited hive "issue" is not a bug since it's the only way to
> > work in a multi-port env.
> >
> >
> >
> > 4. Coordinating the port add/change of uris is systems everywhere (you
> > know something will be missed), updating of confs, restarting all
> services,
> > requiring customers to redeploy their workflows in sync with the NN
> > upgrade, will cause mass disruption and downtime that will be
> unacceptable
> > for production environments.
> >
> >
> >
> > This is a solution to a non-existent problem.  Ports can be bound by
> > multiple processes but only 1 can listen.  Maybe multiple listeners is an
> > issue for compute nodes but not responsibly managed service nodes.  Ie.
> Who
> > runs arbitrary services on the NNs that bind to random ports?  Besides,
> the
> > default port is and was ephemeral so it solved nothing.
> >
> >
> >
> > This either standardizes ports to a particular customer's ports or is a
> > poorly thought out whim.  In either case, the needs of the many outweigh
> > the needs of the few/none (3.0 users).  The only logical conclusion is
> > revert.  If a particular site wants to change default ports and deal with
> > the massive fallout, they can explicitly change the ports themselves.
> >
> >
> >
> > Daryn
> >
> > On Tue, Jan 9, 2018 at 11:22 PM, Aaron T. Myers <[hidden email]<mailto:
> > [hidden email]>> wrote:
> > On Tue, Jan 9, 2018 at 3:15 PM, Eric Yang <[hidden email]<mailto:
> > [hidden email]>> wrote:
> >
> > > While I agree the original port change was unnecessary, I don’t think
> > > Hadoop NN port change is a bad thing.
> > >
> > > I worked for a Hadoop distro that NN RPC port was default to port 9000.
> > > When we migrate from BigInsights to IOP and now to HDP, we have to move
> > > customer Hive metadata to new NN RPC port.  It only took one developer
> > > (myself) to write the tool for the migration.  The incurring workload
> is
> > > not as bad as most people anticipated because Hadoop depends on
> > > configuration file for referencing namenode.  Most of the code can work
> > > transparently.  It helped to harden the downstream testing tools to be
> > more
> > > robust.
> > >
> >
> > While there are of course ways to deal with this, the question really
> > should be whether or not it's a desirable thing to do to our users.
> >
> >
> > >
> > > We will never know how many people are actively working on Hadoop
> 3.0.0.
> > > Perhaps, couple hundred developers or thousands.
> >
> >
> > You're right that we can't know for sure, but I strongly suspect that
> this
> > is a substantial overestimate. Given how conservative Hadoop operators
> tend
> > to be, I view it as exceptionally unlikely that many deployments have
> been
> > created on or upgraded to Hadoop 3.0.0 since it was released less than a
> > month ago.
> >
> > Further, I hope you'll agree that the number of
> > users/developers/deployments/applications which are currently on Hadoop
> > 2.x
> > is *vastly* greater than anyone who might have jumped on Hadoop 3.0.0 so
> > quickly. When all of those users upgrade to any 3.x version, they will
> > encounter this needless incompatible change and be forced to work around
> > it.
> >
> >
> > > I think the switch back may have saved few developers work, but there
> > > could be more people getting impacted at unexpected minor release
> change
> > in
> > > the future.  I recommend keeping current values to avoid rule bending
> and
> > > future frustrations.
> > >
> >
> > That we allow this incompatible change now does not mean that we are
> > categorically allowing more incompatible changes in the future. My point
> is
> > that we should in all instances evaluate the merit of any incompatible
> > change on a case-by-case basis. This is not an exceptional circumstance -
> > we've made incompatible changes in the past when appropriate, e.g.
> breaking
> > some clients to address a security issue. I and others believe that in
> this
> > case the benefits greatly outweigh the downsides of changing this back to
> > what it has always been.
> >
> > Best,
> > Aaron
> >
> >
> > >
> > > Regards,
> > > Eric
> > >
> > > On 1/9/18, 11:21 AM, "Chris Douglas" <[hidden email]<mailto:cd
> > [hidden email]>> wrote:
> > >
> > >     Particularly since 9820 isn't in the contiguous range of ports in
> > >     HDFS-9427, is there any value in this change?
> > >
> > >     Let's change it back to prevent the disruption to users, but
> > >     downstream projects should treat this as a bug in their tests.
> Please
> > >     open JIRAs in affected projects. -C
> > >
> > >
> > >     On Tue, Jan 9, 2018 at 5:18 AM, larry mccay <[hidden email]
> > <mailto:[hidden email]>> wrote:
> > >     > On Mon, Jan 8, 2018 at 11:28 PM, Aaron T. Myers <[hidden email]
> > <mailto:[hidden email]>>
> > > wrote:
> > >     >
> > >     >> Thanks a lot for the response, Larry. Comments inline.
> > >     >>
> > >     >> On Mon, Jan 8, 2018 at 6:44 PM, larry mccay <[hidden email]
> > <mailto:[hidden email]>>
> > > wrote:
> > >     >>
> > >     >>> Question...
> > >     >>>
> > >     >>> Can this be addressed in some way during or before upgrade that
> > > allows it
> > >     >>> to only affect new installs?
> > >     >>> Even a config based workaround prior to upgrade might make
> this a
> > > change
> > >     >>> less disruptive.
> > >     >>>
> > >     >>> If part of the upgrade process includes a step (maybe even a
> > > script) to
> > >     >>> set the NN RPC port explicitly beforehand then it would allow
> > > existing
> > >     >>> deployments and related clients to remain whole - otherwise it
> > > will uptake
> > >     >>> the new default port.
> > >     >>>
> > >     >>
> > >     >> Perhaps something like this could be done, but I think there are
> > > downsides
> > >     >> to anything like this. For example, I'm sure there are plenty of
> > >     >> applications written on top of Hadoop that have tests which
> > > hard-code the
> > >     >> port number. Nothing we do in a setup script will help here. If
> we
> > > don't
> > >     >> change the default port back to what it was, these tests will
> > > likely all
> > >     >> have to be updated.
> > >     >>
> > >     >>
> > >     >
> > >     > I may not have made my point clear enough.
> > >     > What I meant to say is to fix the default port but direct folks
> to
> > >     > explicitly set the port they are using in a deployment (the
> current
> > >     > default) so that it doesn't change out from under them - unless
> > they
> > > are
> > >     > fine with it changing.
> > >     >
> > >     >
> > >     >>
> > >     >>> Meta note: we shouldn't be so pedantic about policy that we
> can't
> > > back
> > >     >>> out something that is considered a bug or even mistake.
> > >     >>>
> > >     >>
> > >     >> This is my bigger point. Rigidly adhering to the compat
> guidelines
> > > in this
> > >     >> instance helps almost no one, while hurting many folks.
> > >     >>
> > >     >> We basically made a mistake when we decided to change the
> default
> > > NN port
> > >     >> with little upside, even between major versions. We discovered
> > this
> > > very
> > >     >> quickly, and we have an opportunity to fix it now and in so
> doing
> > > likely
> > >     >> disrupt very, very few users and downstream applications. If we
> > > don't
> > >     >> change it, we'll be causing difficulty for our users, downstream
> > >     >> developers, and ourselves, potentially for years.
> > >     >>
> > >     >
> > >     > Agreed.
> > >     >
> > >     >
> > >     >>
> > >     >> Best,
> > >     >> Aaron
> > >     >>
> > >
> > >     -----------------------------------------------------------
> > ----------
> > >     To unsubscribe, e-mail: [hidden email]
> > <mailto:[hidden email]>
> > >     For additional commands, e-mail: [hidden email]
> > <mailto:[hidden email]>
> > >
> > >
> > >
> > >
> >
> >
>