make repository invalidation easier on the system

[mamh2021]

Describe the bug
we have five big android code for index.
after restart tomcat the web page load take too long time
the ps command show that: there are many git process:

root     12983     0  0 04:12 ?        00:00:00 [git]
root     13029     1  0 04:12 ?        00:00:00 [git] <defunct>
root     13030     1  0 04:12 ?        00:00:00 [git] <defunct>
root     13037     1  0 04:12 ?        00:00:00 [git]
root     13038     1  0 04:12 ?        00:00:00 [git] <defunct>
root     13039     1  0 04:12 ?        00:00:00 [git] <defunct>
root     13095     1  0 04:12 ?        00:00:00 /usr/bin/git log --format=commit:%H%nDate:%at -n 1 tag_MP_20170522021208 --

tomcat run under the docker, the version is tomcat:9.0
the start docker comamnd is:

 docker run --restart always -d -e JAVA_OPTS='-Xmx60g' -p 80:8080 --name tomcat -v /work/buildsrv-ci:/work/buildsrv-ci -v /work/mirror:/work/mirror -v /opt/git:/opt/git -v /work/buildsrv-ci/web_root:/usr/local/tomcat/webapps   tomcat:9.0

the host server is ubuntu 18.04
the opengrok is 1.3.11 (* 77b6669 - 1.3.11 - Vladimir Kotal - (9 weeks ago))

Screenshots

the index command in indexer.sh like below:

#!/bin/bash -x
export OPENGROK_HOME=/home/buildsrv-ci/opengrok
export PATH=$PATH:$OPENGROK_HOME/opengrok-tools/bin
export WORK=/work/buildsrv-ci
export SITE=$1

opengrok-indexer \
  -J=-Xmx100g -J=-Djava.util.logging.config.file=$OPENGROK_HOME/etc/logging.properties \
  -a $OPENGROK_HOME/dist/opengrok-1.3.11/lib/opengrok.jar -- \
  -s $WORK/src_root/${SITE} -d $WORK/data_root/${SITE} -H -P -S -G \
  -W $WORK/data_root/${SITE}/configuration.xml -U http://androidxref.example.com/${SITE}

[vladak]

All these git commands are spawned via HistoryGuru#invalidateRepositories() which goes through all repositories and checks their status. This sort of work is legitimate. This happens whenever new configuration is set. In this case the new configuration is set at the end of indexing.

It's a question whether the slow load of the index page is caused by some internal delays (e.g. caused by locking or because Git repository handling in OpenGrok spawns superfluos git commands) or simply because the machine has high CPU load - the 2nd screenshot hints towards the latter I think.

One way to work around this would be to avoid setting the configuration at the end of the indexing - assuming the projects/repositories do not change. Same thing for repository scanning (-S). This might lead to switching to different workflow - see https://github.com/oracle/opengrok/wiki/Per-project-management-and-workflow

[vladak]

For the Git repository handling inefficiency, this is actually a bug - see #2986

[mamh2021]

thanks your reply very much.

[vladak]

Could you retry with 1.3.16 to see if the problem persists ?

[mamh2021]

I see see #2986, thanks your reply.

I found that, when redeploy the war to tomcat, this method "buildTagList" in file GItRepository.java will be called.
this method will block the web thread?? and this method not run in a background thread??

in my git repo, threre are many tags, so the git tag command takes a lot of time .

[vladak]

When webapp starts, invalidateRepositories() is called that in turn calls buildTagList() via getRepository(). All the handling is sequential so yes, the webapp fill finish deploying only after this task finishes.

[mamh2021]

I found the buildTagList() run in another thread, (final ExecutorService executor = Executors.newFixedThreadPool)
but this line "latch.await(); " will block the web thread. am I correct?

[mamh2021]

I don't understand , why webappListerner call this method "buildTagList()",
We have used the offline indexer (python opengrok-indexer command) to create the data why we need buildTagList() when war load ?
Can I remove this method call ?

thanks your reply
Thank you very much. I've learned a lot from your reply.

[vladak]

That's a good question. Quite possibly the tag list is not needed when the webapp is starting. AFAIK it is only used in the history view of particular directory/file.

[mamh2021]

maybe we can move all git time-consuming operations to the offline indexer, and save to local file?
then in web reload the file. loading a file should be quick.

[vladak]

The information about the repositories is already saved to the configuration by the indexer (when running with -W that is). The invalidation is done for the webapp to have fresh view of the repositories I think.

[ChristopheBordieu]

I have exactly same trouble with OG 1.5.11 and 1.9k projects and 30k Git repos. When restarting Tomcat webapp, it takes ages before to have access to web service.
Is this issue supposed to be fixed with #2986?

[vladak]

It might help a bit. #1113 might have more profound effect for your use case.

[mamh2021]

I have exactly same trouble with OG 1.5.11 and 1.9k projects and 30k Git repos. When restarting Tomcat webapp, it takes ages before to have access to web service.
Is this issue supposed to be fixed with #2986?

disable tags."-G", "--assignTags"Maybe it will be better