TestQueryOneToOneRelations flaky in CI

[fredcarle]

TestQueryOneToOneRelations often fails on the change detector on the CI. It's not clear yet what is causing this.

It looks to be #1672 - only now flaky instead of consistently failing (and on a different, but similar test).

Possibly significant is that both this and #1672 involved tests with one-one relations.

• Executing TestQueryOneToOneRelations only does not resolve the issue (other tests do not appear to be a factor)
• Removing the create statements for either Publishers or Authors appears to remove the issue
• Reducing the query to just Book {name} appears to resolve the issue - this is unlike AssertFail in badger file on db close #1672 which persisted on a simplified query
  • Reducing the query to Book {name author {name}} (removing the publisher join) also resolves the issue
  • Reducing the query to Book {name publisher {name}} (removing the author join) does not resolve the issue. This interestingly is the one-one join (same as AssertFail in badger file on db close #1672)
• Setting n.subtype.Spans() does not resolve the issue (is a difference between the primary and secondary)
• No errors in valuesSecondary appear to be generated, despite typeJoinOne silently discards errors in Next #1711
• Returning early from valuesSecondary immediately after n.subType.Init() resolves the issue, this is before n.subType.Next()
• Returning early from valuesSecondary immediately after n.subType.Next() does not resolve the issue, suggesting Next is the trigger (same as in AssertFail in badger file on db close #1672)
• Removing n.subType.Init() from typeJoinOne.Init() does not resolve the issue (although it is a duplicate, pointless call I think)
• Updating to badger v4 still has no affect
• Removing the 'extra' df.kvResultsIter.Close() call added to the fetcher in AssertFail in badger file on db close #1672 appears to resolve the issue It did for a while, and then broke following a git push

[AndrewSisley]

It looks to be #1672 - only now flaky instead of consistently failing (and on a different, but similar test).

The fact that it is now flaky is pulling me closer to believing this is something GC related within the badger code.

[AndrewSisley]

Oddly, whilst this has been problematic in the CI today, I have no trouble running the change detector on my laptop atm (3 out of 3) (unlike #1672 which was failing pretty consistently locally)

EDIT: 2 out of 3 runs failed locally this morning (same test as in desc).

[fredcarle]

It fails every third run for me locally. Like clockwork

[AndrewSisley]

It fails every third run for me locally. Like clockwork

I wonder if that is significant. How many times have you seen this set-of-three pattern?

[fredcarle]

It fails every third run for me locally. Like clockwork

I wonder if that is significant. How many times have you seen this set-of-three pattern?

Pretty consistently. So at least 20 times thus far. Sometimes I get 3 or 4 successful runs but it's usually 2 good ones and the third fails. A couple subsequent failures too.