Sergey Soldatov created HDDS-16081:
--------------------------------------
Summary: Flaky
TestOMRatisSnapshots#testInstallSnapshotWithClientWrite: follower may not have
all keys when the leader's last applied index is reached
Key: HDDS-16081
URL: https://issues.apache.org/jira/browse/HDDS-16081
Project: Apache Ozone
Issue Type: Test
Components: Snapshot, test
Affects Versions: 2.3.0
Reporter: Sergey Soldatov
Assignee: Sergey Soldatov
testInstallSnapshotWithClientWrite fails intermittently with "expected: not
<null>" when it verifies that the follower OM's DB contains the keys written
after the follower was started. The test waits for the follower's last applied
index to reach the leader's and reads the keys directly from the follower's DB.
An OM acks a write as soon as the response is added to the double buffer, but
the last applied index only advances *after* the double buffer commits the
batch to RocksDB. When the client writes return, the leader's applied index can
still be behind the last key written, and the follower.
This appears after HDDS-10310, which replaced the Thread.sleep(5000) with the
index-based wait, and removed the per-key sleep from the write loop, making
the writes fast enough to widen the window.
[~chihsuan] Please let me know if my RCA was wrong.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
