Mitigation for remote snapshot filecache overflow

[finnegancarroll]

Description

Remote snapshot file_cache does not strictly limit it's disk usage to its configured capacity. This change causes subsequent requests to the file cache to fail when the cache overflows. #11676 for more details.

Related Issues

Mitigation for #11676

Check List

• Functionality includes testing.
• API changes companion pull request created, if applicable.
• Public documentation issue/PR created, if applicable.

By submitting this pull request, I confirm that my contribution is made under the terms of the Apache 2.0 license.
For more information on following Developer Certificate of Origin and signing off your commits, please check here.

[github-actions]

❌ Gradle check result for 9b494b6: FAILURE

Please examine the workflow log, locate, and copy-paste the failure(s) below, then iterate to green. Is the failure a flaky test unrelated to your change?

[sandervandegeijn]

Thanks for the effort Finn!

[sandervandegeijn]

Looking at the code:

if (fileCache.usage().usage() > fileCache.capacity()) {
                System.gc();

Would it be possible to do this at 70-90% of capacity and at > 100%? At >100% you are already too late right?

[github-actions]

❌ Gradle check result for 6f86ed4: FAILURE

Please examine the workflow log, locate, and copy-paste the failure(s) below, then iterate to green. Is the failure a flaky test unrelated to your change?

[finnegancarroll]

> Task :server:forbiddenApisMain FAILED
Forbidden method invocation: java.lang.System#gc() [Please do not try to stop the world]

@andrross tests which exceed the cache capacity have drastically fewer failed requests with 'System.gc()' and a small sleep. Is this appropriate for a mitigation or is it preferable to just fail and make no attempt to prune the cache here?

[andrross]

> Task :server:forbiddenApisMain FAILED
Forbidden method invocation: java.lang.System#gc() [Please do not try to stop the world]

@andrross tests which exceed the cache capacity have drastically fewer failed requests with 'System.gc()' and a small sleep. Is this appropriate for a mitigation or is it preferable to just fail and make no attempt to prune the cache here?

@finnegancarroll Yeah we shouldn't be trying to invoke GC manually. The better way to do this, if we need to do this at all, is to do the tracking of the unclosed cloned objects in the parent via weak references, and then deterministically close everything when the parent is closed. I think that would have the same effect, because anything that is eligible for garbage collection must have already had the originating input stream closed.

[github-actions]

❌ Gradle check result for 38b3b95: null

Please examine the workflow log, locate, and copy-paste the failure(s) below, then iterate to green. Is the failure a flaky test unrelated to your change?

[github-actions]

❌ Gradle check result for f72d436: FAILURE

Please examine the workflow log, locate, and copy-paste the failure(s) below, then iterate to green. Is the failure a flaky test unrelated to your change?

[finnegancarroll]

Reverted to just failing requests on cache overflow.

[sandervandegeijn]

Maybe I'm not looking in the right place, but did this make it into the 2.17.0 release? Can't find it in the release notes?

[dblock]

The backport was merged looks like, #15761, maybe a release notes wasn't updated? Where did you look?

[sandervandegeijn]

Ah I do see it now: https://github.com/opensearch-project/OpenSearch/releases/tag/2.17.0

Might have made a typo while searching, sorry!

[sandeshkr419]

Looks like this was backported in 2.17, but not in 2.x

#15760

@finnegancarroll Can you check once - if required please raise a manual packport and close the auto-backport PR.