[
https://issues.apache.org/jira/browse/CASSANDRA-21352?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=18080461#comment-18080461
]
Radmir Mukhambetov commented on CASSANDRA-21352:
------------------------------------------------
[https://github.com/apache/cassandra/pull/4805]
> TCM: AtomicLongBackedProcessor sort inversion
> ---------------------------------------------
>
> Key: CASSANDRA-21352
> URL: https://issues.apache.org/jira/browse/CASSANDRA-21352
> Project: Apache Cassandra
> Issue Type: Bug
> Reporter: Alex Petrov
> Priority: Normal
>
> {{needsSorting = entry.epoch.isDirectlyAfter(last)}} is inverted. It sets
> {{needsSorting=true}}¬ when entries are already in order (directly after = no
> sort needed) and leaves {{needsSorting=false}} when they arrive out of order.
> In-memory log produces non-deterministic epoch ordering in tests.
> Repro:
> {code}
> /**
> * Epochs arrive out of order: 1, 3, 2.
> * With the bug, needsSorting is false for both 3 (not directly after 1) and
> 2 (not directly after 3),
> * so no sort ever fires and the list stays [1, 3, 2].
> * After the fix (!isDirectlyAfter), both trigger a sort and the list becomes
> [1, 2, 3].
> */
> @Test
> public void outOfOrderAppendsAreSorted()
> {
> InMemoryStorage storage = new InMemoryStorage();
> storage.append(entry(1));
> storage.append(entry(3)); // out of order — gap between 1 and 3
> storage.append(entry(2)); // fills the gap, but storage already has [1,3]
> unsorted by bug
> List<Entry> entries = storage.getLogState(Epoch.EMPTY).entries;
> assertEquals(3, entries.size());
> assertEquals("epoch 1 must be first", Epoch.create(1),
> entries.get(0).epoch);
> assertEquals("epoch 2 must be second", Epoch.create(2),
> entries.get(1).epoch);
> assertEquals("epoch 3 must be third", Epoch.create(3),
> entries.get(2).epoch);
> }
> {code}
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
