OAK-9850: ConcurrentPrefetchAndUpdateIT.cacheConsistency fails occasi…

[mreutegg]

…onally

Keys collection passed to registerTracker must not be modified later.
NodeDocumentCache.putNonConflictingDocs() must always notify other trackers.
Enable test again.

[stefan-egli]

Besides fixing the test, and besides it being related to a new feature being involved (prefetch), would there also have been a possibility for outdated cache state previously?

[mreutegg]

Besides fixing the test

The test was not broken and this PR does not make any changes to the test except enable it again.

would there also have been a possibility for outdated cache state previously?

I think that's possible. The added else clause in NodeDocumentCache.putNonConflictingDocs() is not specific to prefetching.

[stefan-egli]

Wsa wondering if it would make sense to try and come up with a test case that would have failed without this fix. Maybe it could help finding or explaining other issues (like some stray OakMerge0004 conflicts..).

[mreutegg]

Was wondering if it would make sense to try and come up with a test case that would have failed without this fix.

Running ConcurrentPrefetchAndUpdateIT in a loop eventually failed on my machine without the fix. But I agree, it would be better if there was a deterministic test that fails immediately without the fix.

[mreutegg]

Test failure in oak-blob-plugins seems unrelated. I can reproduce the failure locally and created https://issues.apache.org/jira/browse/OAK-9879

[mreutegg]

a test case that would have failed without this fix

@stefan-egli there is now a simple test that illustrates why changes to NodeDocumentCache are necessary. NodeDocumentCacheTest fails deterministically without changes to NodeDocumentCache.putNonConflictingDocs().

[stefan-egli]

+1, nice test