TestHashKVWhenCompacting: hashes differ

[heyitsanthony]

via https://semaphoreci.com/coreos/etcd/branches/master/builds/2189

--- FAIL: TestHashKVWhenCompacting (1.28s)
	kvstore_test.go:561: Hashes differ (current 3655990207) != (saved 597152871)

/cc @fanminshi

[fanminshi]

taking a look.

[fanminshi]

@heyitsanthony does ci run again when pr gets merging in?

[heyitsanthony]

@fanminshi yes; it's a build off of master

[fanminshi]

observed compaction at rev < 899. those cases shouldn't happen because the test code is written to compact from rev [899, 999]. Investigating.

2017-07-26 18:25:13.466656 I | mvcc: finished scheduled compaction at 4 (took 207.206µs)
2017-07-26 18:25:13.472053 I | mvcc: store.index: compact 4
2017-07-26 18:25:13.472416 I | mvcc: finished scheduled compaction at 4 (took 179.986µs)
2017-07-26 18:25:13.584272 I | mvcc: store.index: compact 1
2017-07-26 18:25:13.584649 I | mvcc: finished scheduled compaction at 1 (took 108.484µs)
2017-07-26 18:25:13.584858 I | mvcc: store.index: compact 2
2017-07-26 18:25:13.585071 I | mvcc: finished scheduled compaction at 2 (took 98.398µs)
2017-07-26 18:25:13.585253 I | mvcc: store.index: compact 3
2017-07-26 18:25:13.585435 I | mvcc: finished scheduled compaction at 3 (took 75.431µs)
2017-07-26 18:25:13.585608 I | mvcc: store.index: compact 4
2017-07-26 18:25:13.585828 I | mvcc: finished scheduled compaction at 4 (took 117.553µs)
2017-07-26 18:25:13.589930 I | mvcc: store.index: compact 0

[heyitsanthony]

@fanminshi that's probably from TestRestoreContinueUnfinishedCompaction which runs before the hash test. The test script isn't running the unit tests with -v so the RUN/PASS lines are suppressed.

[fanminshi]

@heyitsanthony i see.

[fanminshi]

reproduced the issue with following:

rev := 1000 => rev := 100000 at https://github.com/coreos/etcd/blob/master/mvcc/kvstore_test.go#L525

--- FAIL: TestHashKVWhenCompacting (10.42s)
	kvstore_test.go:561: Hashes differ (current 2544228963) != (saved 95822769)