Lock search index while reindexing...

[luke-]

Is your feature request related to a problem? Please describe.

While reindexing the search index via console, new entries may added by queue processing.

See: #3847 (comment)

Describe the solution you'd like

Set some sort of lock.

Describe alternatives you've considered

[marc-farre]

It also can happen with the UpdateDocument jobs executed by queue/run when there are very numerous in the queue: File '/home/web/www/humhub/protected/runtime/searchdb/_oat.fdx' is not readable.:

2023-03-15 11:48:01 [361965] humhub\modules\search\jobs\UpdateDocument (attempt: 1, pid: 1756612) - Done (0.119 s, 14.06 MiB)
PHP Fatal error:  Uncaught ZendSearch\Lucene\Exception\InvalidArgumentException: File '/home/web/www/humhub/protected/runtime/searchdb/_oat.fdx' is not readable. in /home/web/www/humhub/protected/vendor/matthewbdaly/zendsearch/src/ZendSearch/Lucene/Storage/File/Filesystem.php:46
Stack trace:
#0 /home/web/www/humhub/protected/vendor/matthewbdaly/zendsearch/src/ZendSearch/Lucene/Storage/Directory/Filesystem.php(354): ZendSearch\Lucene\Storage\File\Filesystem->__construct()
#1 /home/web/www/humhub/protected/vendor/matthewbdaly/zendsearch/src/ZendSearch/Lucene/Index/SegmentWriter/AbstractSegmentWriter.php(611): ZendSearch\Lucene\Storage\Directory\Filesystem->getFileObject()
#2 /home/web/www/humhub/protected/vendor/matthewbdaly/zendsearch/src/ZendSearch/Lucene/Index/SegmentWriter/DocumentWriter.php(211): ZendSearch\Lucene\Index\SegmentWriter\AbstractSegmentWriter->_generateCFS()
#3 /home/web/www/humhub/protected/vendor/matthewbdaly/zendsearch/src/ZendSearch/Lucene/Index/Writer.php(772): ZendSearch\Lucene\Index\SegmentWriter\DocumentWriter->close()
#4 /home/web/www/humhub/protected/vendor/matthewbdaly/zendsearch/src/ZendSearch/Lucene/Index.php(1199): ZendSearch\Lucene\Index\Writer->commit()
#5 /home/web/www/humhub/protected/vendor/matthewbdaly/zendsearch/src/ZendSearch/Lucene/Index.php(443): ZendSearch\Lucene\Index->commit()
#6 [internal function]: ZendSearch\Lucene\Index->__destruct()
#7 {main}
  thrown in /home/web/www/humhub/protected/vendor/matthewbdaly/zendsearch/src/ZendSearch/Lucene/Storage/File/Filesystem.php on line 46

PS (as a reminder): error logs about search re-indexing are described here: #6168

[marc-farre]

https://community.humhub.com/content/perma?id=269660 might be caused by this issue.

[luke-]

@yurabakhtin Any idea?

[yurabakhtin]

@luke- I have read the linked issues and as I have understood the main problem is we should stop cron before run the search rebuild process.

I think we could implement an event triggering for CronController::EVENT_BEFORE_ACTION in order to stop any action(cron/daily, cron/hourly, cron/run) of the controller when the search rebuilding is run.

We should detect when the folowing commands are executed:

• content-search/queue-rebuild
• content-search/rebuild
• content-search/optimize - please note we should either remove it or fix it because the method optimize() was deleted for new Search Drivers.

We could set a setting in DB with name like contentSearchIsRebuilding as flag and delete it after the process is finished.

Do you agree?

[luke-]

@yurabakhtin

Hmm, I don't know if it's a bit too radical to stop all queue processing during the search rebuild. But that could be a solution. Possibly also for updates/maintenance.

I would prefer to realize this via a mutex: https://www.yiiframework.com/doc/api/2.0/yii-mutex-mutex But here we need a fallback in case the search rebuild crashes...

---

I originally expected a solution that the ActiveJobs SearchUpdateDocument and SearchDeleteDocument are delayed/skipped during Rebuild with a delayed retry. https://www.yiiframework.com/extension/yiisoft/yii2-queue/doc/guide/2.0/en/retryable#errors-and-retryable-jobs

[yurabakhtin]

@luke- Thanks for the advices, I have implemented it in PR #6934.