[
https://issues.apache.org/jira/browse/CASSANDRA-15861?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17141548#comment-17141548
]
ZhaoYang commented on CASSANDRA-15861:
--------------------------------------
bq.
https://github.com/apache/cassandra/pull/642/files#diff-de503ddc819368b078a86f6d9e3921aeR211-R228.
isn't the write async; we are just storing on a buffer to be written later?
Looking at
org.apache.cassandra.net.AsyncStreamingOutputPlus#writeFileToChannelZeroCopy it
calls channel.writeAndFlush but doesn't look at the future, so the write is
async.
{{writeFileToChannelZeroCopy}} is async, but if I remember correctly about unix
file system: once a file is opened, reader won't be affected by file deletion
or rewrite. So when the synchronized block completes, all component
file-channels are already opened and in-sync with {{ComponentManifest}}, even
if they are not flushed from netty outbound buffer to kernel.
I agree with you that the atomicity should be made more obvious rather than
relying on underlying FS. I will find a cleaner approach.
bq. have you done any longevity testing of this? My fear is that compaction
will get blocked while streaming is running which could cause slowness or
stability issues.
Not yet, ticket is still "in-progress". It'd be nice to reuse the tests did in
CASSANDRA-14556..
bq. Make STATS mutation as a proper compaction to create hard link on the
compacting sstable components with a new descriptor, except STATS files which
will be copied entirely. Then mutation will be applied on the new STATS file.
At the end, old sstable will be released. This ensures all sstable components
are immutable and shouldn't make these special compaction tasks slower.
I had a
[prototype|https://github.com/jasonstack/cassandra/blob/cb9bdaf037fd550b84fe5b7da89f9c56dc729c35/src/java/org/apache/cassandra/db/compaction/StatsMutationCompaction.java#L72]
using the hardlink approach which should avoid blocking between compaction and
streaming. The only thing I don't like is redistributing index summary every
hour will explode sstable generation.. what do you guys think?
> Mutating sstable component may race with entire-sstable-streaming(ZCS)
> causing checksum validation failure
> ----------------------------------------------------------------------------------------------------------
>
> Key: CASSANDRA-15861
> URL: https://issues.apache.org/jira/browse/CASSANDRA-15861
> Project: Cassandra
> Issue Type: Bug
> Components: Consistency/Repair, Consistency/Streaming,
> Local/Compaction
> Reporter: ZhaoYang
> Assignee: ZhaoYang
> Priority: Normal
> Fix For: 4.0-beta
>
>
> Flaky dtest: [test_dead_sync_initiator -
> repair_tests.repair_test.TestRepair|https://ci-cassandra.apache.org/view/all/job/Cassandra-devbranch-dtest/143/testReport/junit/dtest.repair_tests.repair_test/TestRepair/test_dead_sync_initiator/]
> {code:java|title=stacktrace}
> Unexpected error found in node logs (see stdout for full details). Errors:
> [ERROR [Stream-Deserializer-127.0.0.1:7000-570871f3] 2020-06-03 04:05:19,081
> CassandraEntireSSTableStreamReader.java:145 - [Stream
> 6f1c3360-a54f-11ea-a808-2f23710fdc90] Error while reading sstable from stream
> for table = keyspace1.standard1
> org.apache.cassandra.io.sstable.CorruptSSTableException: Corrupted:
> /home/cassandra/cassandra/cassandra-dtest/tmp/dtest-te4ty0r9/test/node3/data0/keyspace1/standard1-5f5ab140a54f11eaa8082f23710fdc90/na-2-big-Statistics.db
> at
> org.apache.cassandra.io.sstable.metadata.MetadataSerializer.maybeValidateChecksum(MetadataSerializer.java:219)
> at
> org.apache.cassandra.io.sstable.metadata.MetadataSerializer.deserialize(MetadataSerializer.java:198)
> at
> org.apache.cassandra.io.sstable.metadata.MetadataSerializer.deserialize(MetadataSerializer.java:129)
> at
> org.apache.cassandra.io.sstable.metadata.MetadataSerializer.mutate(MetadataSerializer.java:226)
> at
> org.apache.cassandra.db.streaming.CassandraEntireSSTableStreamReader.read(CassandraEntireSSTableStreamReader.java:140)
> at
> org.apache.cassandra.db.streaming.CassandraIncomingFile.read(CassandraIncomingFile.java:78)
> at
> org.apache.cassandra.streaming.messages.IncomingStreamMessage$1.deserialize(IncomingStreamMessage.java:49)
> at
> org.apache.cassandra.streaming.messages.IncomingStreamMessage$1.deserialize(IncomingStreamMessage.java:36)
> at
> org.apache.cassandra.streaming.messages.StreamMessage.deserialize(StreamMessage.java:49)
> at
> org.apache.cassandra.streaming.async.StreamingInboundHandler$StreamDeserializingTask.run(StreamingInboundHandler.java:181)
> at
> io.netty.util.concurrent.FastThreadLocalRunnable.run(FastThreadLocalRunnable.java:30)
> at java.lang.Thread.run(Thread.java:748)
> Caused by: java.io.IOException: Checksums do not match for
> /home/cassandra/cassandra/cassandra-dtest/tmp/dtest-te4ty0r9/test/node3/data0/keyspace1/standard1-5f5ab140a54f11eaa8082f23710fdc90/na-2-big-Statistics.db
> {code}
>
> In the above test, it executes "nodetool repair" on node1 and kills node2
> during repair. At the end, node3 reports checksum validation failure on
> sstable transferred from node1.
> {code:java|title=what happened}
> 1. When repair started on node1, it performs anti-compaction which modifies
> sstable's repairAt to 0 and pending repair id to session-id.
> 2. Then node1 creates {{ComponentManifest}} which contains file lengths to be
> transferred to node3.
> 3. Before node1 actually sends the files to node3, node2 is killed and node1
> starts to broadcast repair-failure-message to all participants in
> {{CoordinatorSession#fail}}
> 4. Node1 receives its own repair-failure-message and fails its local repair
> sessions at {{LocalSessions#failSession}} which triggers async background
> compaction.
> 5. Node1's background compaction will mutate sstable's repairAt to 0 and
> pending repair id to null via
> {{PendingRepairManager#getNextRepairFinishedTask}}, as there is no more
> in-progress repair.
> 6. Node1 actually sends the sstable to node3 where the sstable's STATS
> component size is different from the original size recorded in the manifest.
> 7. At the end, node3 reports checksum validation failure when it tries to
> mutate sstable level and "isTransient" attribute in
> {{CassandraEntireSSTableStreamReader#read}}.
> {code}
> This isn't a problem in legacy streaming as STATS file length didn't matter.
> Ideally it will be great to make sstable STATS metadata immutable, just like
> other sstable components, so we don't have to worry this special case.
> I can think of 2 ways:
> # Make STATS mutation as a proper compaction to create hard link on the
> compacting sstable components with a new descriptor, except STATS files which
> will be copied entirely. Then mutation will be applied on the new STATS file.
> At the end, old sstable will be released. This ensures all sstable components
> are immutable and shouldn't make these special compaction tasks slower.
> # Change STATS metadata format to use fixed length encoding for repair info
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
---------------------------------------------------------------------
To unsubscribe, e-mail: commits-unsubscr...@cassandra.apache.org
For additional commands, e-mail: commits-h...@cassandra.apache.org
