EZP-30823: [6.x] Isolated cache during transactions to avoid race condition

[andrerom]

Question	Answer
JIRA issue	EZP-30823
Bug/Improvement	yes
Target version	6.7+ (see #2749 for 7.x PR)
BC breaks	no
Tests pass	yes
Doc needed	no

Similar to what was done for 2.5.3 (#2703) and followup in #2749, but adapted for Stash:

• As Stash has "fat Items" need to catch calls to store/clear on the object itself, this part is not relevant for 7.x
• Route calls to store items to just deferred delete the key instead
• Make it so that Items with keys that are marked as invalidated gets returned as a "miss" by the system
• And with that be able to get rid of the need to clear all cache on rollback, we instead just need to wipe out local changes

TODO:

• Implement feature / fix a bug.
• Implement tests.
• Fix new code according to Coding Standards ($ composer fix-cs).
• Ask for Code Review.

[andrerom]

Tests passing and it also passed scenario defined by @mateuszbieniek on JIRA.

But ideally I'd also like to somehow have better test coverage for this that would be relevant also for 7.5. But a bit unsure how, suggestions welcome.

On that there is actually two things to tests:

• the race condition, so will have to use separate processes (or some other way of having separate cache pool instance), for instance call some other process before doing a commit
• within same process/instance checks to see that handlers returns correct expected values, even before we commit all the changes Added a test that tries to verify this

[alongosz]

Looks almost good, I have some naming/CS concerns mostly.

[andrerom]

@alongosz Updated, thanks for the suggestions 👍

[lserwatka]

@micszo we can proceed with QA

[micszo]

QA Approved.
Retest successful for Content List block issue and wrong current version issue on eZ Platform EE v1.7.9 with 6.7_race_cond_testing branch. Verified with sanities.
As agreed testing on v1.13.5 will be done post merge.
Testing on v2.5 is ongoing in scope of #2749 / 7.5_race_cond_testing branch.

[micszo]

@andrerom is it ok if we merge this one as well?

[andrerom]

@micszo We can merge this once the 2.5 variant of this patch is approved (#2749), you can actually re-test that more easily by just asking for dev-EZP-30823_7.5_transaction_cache_isolation as 7.5.x-dev for kernel directly. Now that the version loading patch has been merged it has been rebased to include that and other changes in the meantime.

If you still experience login issues, then let's do a zoom session as I did not manage to reproduce that.

[andrerom]

Merging this up so we can ship this to 1.13LTS reporter on this, to not be blocked by further 2.5LTS testing.

Merge details: I'll wipe out all changes here when merging up to 2.5 (7.5 ), and move changes needed over to the 7.5 PR (tests etc)