RE: Jira Upgrade August 19th 0500 UTC

classic Classic list List threaded Threaded
1 message Options
Reply | Threaded
Open this post in threaded view
|

RE: Jira Upgrade August 19th 0500 UTC

Uwe Schindler-2
Hi John, hi committers,

This is great news. As a Lucene committer, I was waiting for this since long time!

> Please be aware that Jira (https://issues.apache.org/jira) will be
> unavailable for approximately 4 hours beginning August 19, 2019 0500
> UTC for migration to new hardware and upgrades to the Jira Software
> suite of tools including all plugins and the major version of Jira.
> For more details on what is new in Jira 8.x you can look at the
> release notes
>
> https://confluence.atlassian.com/jirasoftware/jira-software-8-0-x-release-
> notes-957981626.html

This version of JIRA also ships with a new version of Apache Lucene instead of the stone-aged Apache Lucene 3.3.0 integration. Next to speed improvements and much smaller indexes, the new version also hopefully fixes the scoring issues when you quicksearch (top-right) for issues - often showing irrelevant results first. I worked together with Atlassian last year to revamp the Apache Lucene integration and we used the new Lucene features like index-joins to improve scoring, and numeric/point field types for date searches and usage of DocValues for sorting. Also complex permission schemes are now much faster due to denormalization 😊. FYI, Lucene in JIRA is now on 7.3.1.

Here is the talk:
https://archive.fosdem.org/2019/schedule/event/lucene_upgrade_jira_8/

@John: Please keep in mind that the upgrade to JIRA 8 requires a full reindexing of all issues. Although it's faster now, it still takes some significant time, especially if the database is on a different machine. But in the future, regular reindexing should no longer be needed, as Lucene now uses checksums and fsyncs on the directory metadata to prevent consistency issues on power outages or other crashes. Also JIRA no longer "optimizes" (aka force-merge) indexes, which is not recommended for indexes that have high update rates. Due to the use of DocValues for sorting, the heap-space requirements of JIRA lowered quite a lot! Also keep in mind to have at least 50 % of physical RAM available for memory-mapping of indexes, so reduce heap size accordingly.

You can contact me, if you need some advice on installation!

Uwe

-----
Uwe Schindler
[hidden email]
ASF Member, Apache Lucene PMC / Committer
Bremen, Germany
https://lucene.apache.org/



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