[
https://issues.apache.org/jira/browse/CASSANDRA-13696?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16092658#comment-16092658
]
Jay Zhuang commented on CASSANDRA-13696:
----------------------------------------
Thanks [~jjirsa].
I did more investigation today. Seems it's more serious than I thought. Even
there's no down node, "drop table" while there's write traffic, it will trigger
the problem.
Here is reproduce steps:
1. Create a 3 nodes cluster:
{{$ ccm create test13696 -v 3.0.14 && ccm populate -n 3 && ccm start}}
2. Send some traffics with cassandra-stress (blogpost.yaml is only in trunk, if
you use another yaml file, change the RF=3)
{{$ tools/bin/cassandra-stress user profile=test/resources/blogpost.yaml
cl=QUORUM truncate=never ops\(insert=1\) duration=30m -rate threads=2 -mode
native cql3 -node 127.0.0.1}}
3. While the traffic is running, drop table
{{$ cqlsh -e "drop table stresscql.blogposts"}}
*All 3 nodes go down because of "Digest mismatch Exception".*
The CRC calculation problem has been there for a long time, but only got
exposed after CASSANDRA-13004 because of the MessagingService version bump. In
the normal case when the versions are the same, HintsDispatcher uses
{{[page.buffersIterator()|https://github.com/apache/cassandra/blob/cassandra-3.0/src/java/org/apache/cassandra/hints/HintsDispatcher.java#L138]}}
instead of
{{[page.hintsIterator()|https://github.com/apache/cassandra/blob/cassandra-3.0/src/java/org/apache/cassandra/hints/HintsDispatcher.java#L139]}}.
{{buffersIterator()}} doesn't need to decode hints, so it won't have the
problem.
I think the messagingVersion for the hints file should be updated:
https://github.com/apache/cassandra/compare/cassandra-3.0...cooldoger:13696.2-3.0?expand=1
so it could dispatch hints in an optimized way. Not sure if we need to
check/bump other {{MessagingService.VERSION_30}}s in the 3.0 branch.
cc [~ifesdjeen]
> Digest mismatch Exception if hints file has UnknownColumnFamily
> ---------------------------------------------------------------
>
> Key: CASSANDRA-13696
> URL: https://issues.apache.org/jira/browse/CASSANDRA-13696
> Project: Cassandra
> Issue Type: Bug
> Components: Core
> Reporter: Jay Zhuang
> Assignee: Jay Zhuang
> Priority: Critical
>
> {noformat}
> WARN [HintsDispatcher:2] 2017-07-16 22:00:32,579 HintsReader.java:235 -
> Failed to read a hint for /127.0.0.2: a2b7daf1-a6a4-4dfc-89de-32d12d2d48b0 -
> table with id 3882bbb0-6a71-11e7-9bca-2759083e3964 is unknown in file
> a2b7daf1-a6a4-4dfc-89de-32d12d2d48b0-1500242103097-1.hints
> ERROR [HintsDispatcher:2] 2017-07-16 22:00:32,580
> HintsDispatchExecutor.java:234 - Failed to dispatch hints file
> a2b7daf1-a6a4-4dfc-89de-32d12d2d48b0-1500242103097-1.hints: file is corrupted
> ({})
> org.apache.cassandra.io.FSReadError: java.io.IOException: Digest mismatch
> exception
> at
> org.apache.cassandra.hints.HintsReader$HintsIterator.computeNext(HintsReader.java:199)
> ~[main/:na]
> at
> org.apache.cassandra.hints.HintsReader$HintsIterator.computeNext(HintsReader.java:164)
> ~[main/:na]
> at
> org.apache.cassandra.utils.AbstractIterator.hasNext(AbstractIterator.java:47)
> ~[main/:na]
> at
> org.apache.cassandra.hints.HintsDispatcher.sendHints(HintsDispatcher.java:157)
> ~[main/:na]
> at
> org.apache.cassandra.hints.HintsDispatcher.sendHintsAndAwait(HintsDispatcher.java:139)
> ~[main/:na]
> at
> org.apache.cassandra.hints.HintsDispatcher.dispatch(HintsDispatcher.java:123)
> ~[main/:na]
> at
> org.apache.cassandra.hints.HintsDispatcher.dispatch(HintsDispatcher.java:95)
> ~[main/:na]
> at
> org.apache.cassandra.hints.HintsDispatchExecutor$DispatchHintsTask.deliver(HintsDispatchExecutor.java:268)
> [main/:na]
> at
> org.apache.cassandra.hints.HintsDispatchExecutor$DispatchHintsTask.dispatch(HintsDispatchExecutor.java:251)
> [main/:na]
> at
> org.apache.cassandra.hints.HintsDispatchExecutor$DispatchHintsTask.dispatch(HintsDispatchExecutor.java:229)
> [main/:na]
> at
> org.apache.cassandra.hints.HintsDispatchExecutor$DispatchHintsTask.run(HintsDispatchExecutor.java:208)
> [main/:na]
> at
> java.util.concurrent.Executors$RunnableAdapter.call(Executors.java:511)
> [na:1.8.0_111]
> at java.util.concurrent.FutureTask.run(FutureTask.java:266) [na:1.8.0_111]
> at
> java.util.concurrent.ThreadPoolExecutor.runWorker(ThreadPoolExecutor.java:1142)
> [na:1.8.0_111]
> at
> java.util.concurrent.ThreadPoolExecutor$Worker.run(ThreadPoolExecutor.java:617)
> [na:1.8.0_111]
> at
> org.apache.cassandra.concurrent.NamedThreadFactory.lambda$threadLocalDeallocator$0(NamedThreadFactory.java:79)
> [main/:na]
> at java.lang.Thread.run(Thread.java:745) ~[na:1.8.0_111]
> Caused by: java.io.IOException: Digest mismatch exception
> at
> org.apache.cassandra.hints.HintsReader$HintsIterator.computeNextInternal(HintsReader.java:216)
> ~[main/:na]
> at
> org.apache.cassandra.hints.HintsReader$HintsIterator.computeNext(HintsReader.java:190)
> ~[main/:na]
> ... 16 common frames omitted
> {noformat}
> It causes multiple cassandra nodes stop [by
> default|https://github.com/apache/cassandra/blob/cassandra-3.0/conf/cassandra.yaml#L188].
> Here is the reproduce steps on a 3 nodes cluster, RF=3:
> 1. stop node1
> 2. send some data with quorum (or one), it will generate hints file on
> node2/node3
> 3. drop the table
> 4. start node1
> node2/node3 will report "corrupted hints file" and stop. The impact is very
> bad for a large cluster, when it happens, almost all the nodes are down at
> the same time and we have to remove all the hints files (which contain the
> dropped table) to bring the node back.
--
This message was sent by Atlassian JIRA
(v6.4.14#64029)
---------------------------------------------------------------------
To unsubscribe, e-mail: commits-unsubscr...@cassandra.apache.org
For additional commands, e-mail: commits-h...@cassandra.apache.org
