commit 31ca2fe7afdcd7d49ea6ede395dd2e8c30a171f0
author Edwin Kempin <ekempin@google.com> Tue Feb 28 14:00:28 2023 +0100
committer Edwin Kempin <ekempin@google.com> Tue Feb 28 14:22:34 2023 +0100
tree 4da0a848e31f99522353b82a83d00eea938cfbaa
parent 3a20790135e14f222b935e57ebc4de8857549c8b

Fix flaky StreamEventsIT.batchRefsUpdatedShowSeparatelyInStreamEvents()

This test creates a change, which updates 2 refs that start with
'refs/changes/<sharded-change-id>': 'refs/changes/<sharded-change-id>/1'
(for the patch set) and 'refs/changes/<sharded-change-id>/meta' (for the
change meta data in NoteDb). Then the test polls stream events to verify
that a stream event for each of these ref updates is received.

This test was flaky due to a number of reasons:

1. Change 357298 [1] wrongly changed the number of expected ref updates
   from 2 to 1.

2. The test polls for stream events until it either received the
   expected number of events or a timeout was reached. However when
   checking the number of received events it only considered the events
   that were received by the latest polling, but not the total number of
   received events so far across all pollings. E.g. if 2 events are
   expected, the test failed if the first polling got one event, and the
   second polling got the second event (that's also how the test could
   still pass after changing the expected number of events to 1, see 1.)

3. When reading the received events into a buffer the code ignored the
   number of read bytes. Due to this it was possible that parts of the
   received bytes were appended multiple times to the eventsOutput
   StringBuilder, so that sometimes events were duplictated or broken.

An example failure of the test can be seen at [2].

[1] https://gerrit-review.git.corp.google.com/c/gerrit/+/357298
[2] https://gerrit-ci.gerritforge.com/job/Gerrit-verifier-chrome-latest/27756/console

Release-Notes: skip
Signed-off-by: Edwin Kempin <ekempin@google.com>
Change-Id: I49c18a07bc0b8a543f01c040f301b181112a4f52