Andrew Hust created CASSANDRA-10360:
---------------------------------------
Summary: UnsupportedOperationException when compacting
system.size_estimates after 2.1 -> 2.2 -> 3.0 upgrade
Key: CASSANDRA-10360
URL: https://issues.apache.org/jira/browse/CASSANDRA-10360
Project: Cassandra
Issue Type: Bug
Reporter: Andrew Hust
When upgrading from 2.1 -> 2.2 -> 3.0 the system.size_estimates table will get
stuck in a compaction loop throwing:
{code}
java.lang.UnsupportedOperationException
at
org.apache.cassandra.db.UnfilteredDeserializer$OldFormatDeserializer.readNext(UnfilteredDeserializer.java:382)
at
org.apache.cassandra.io.sstable.SSTableSimpleIterator$OldFormatIterator.computeNext(SSTableSimpleIterator.java:147)
at
org.apache.cassandra.io.sstable.SSTableSimpleIterator$OldFormatIterator.computeNext(SSTableSimpleIterator.java:96)
at
org.apache.cassandra.utils.AbstractIterator.hasNext(AbstractIterator.java:47)
at
org.apache.cassandra.io.sstable.SSTableIdentityIterator.computeNext(SSTableIdentityIterator.java:100)
at
org.apache.cassandra.io.sstable.SSTableIdentityIterator.computeNext(SSTableIdentityIterator.java:30)
at
org.apache.cassandra.utils.AbstractIterator.hasNext(AbstractIterator.java:47)
at
org.apache.cassandra.db.rows.LazilyInitializedUnfilteredRowIterator.computeNext(LazilyInitializedUnfilteredRowIterator.java:95)
at
org.apache.cassandra.db.rows.LazilyInitializedUnfilteredRowIterator.computeNext(LazilyInitializedUnfilteredRowIterator.java:32)
at
org.apache.cassandra.utils.AbstractIterator.hasNext(AbstractIterator.java:47)
at
org.apache.cassandra.utils.MergeIterator$TrivialOneToOne.computeNext(MergeIterator.java:460)
at
org.apache.cassandra.utils.AbstractIterator.hasNext(AbstractIterator.java:47)
at
org.apache.cassandra.db.rows.UnfilteredRowIterators$UnfilteredRowMergeIterator.computeNext(UnfilteredRowIterators.java:503)
at
org.apache.cassandra.db.rows.UnfilteredRowIterators$UnfilteredRowMergeIterator.computeNext(UnfilteredRowIterators.java:363)
at
org.apache.cassandra.utils.AbstractIterator.hasNext(AbstractIterator.java:47)
at
org.apache.cassandra.db.rows.WrappingUnfilteredRowIterator.hasNext(WrappingUnfilteredRowIterator.java:80)
at
org.apache.cassandra.db.rows.AlteringUnfilteredRowIterator.hasNext(AlteringUnfilteredRowIterator.java:72)
at
org.apache.cassandra.db.rows.UnfilteredRowIterator.isEmpty(UnfilteredRowIterator.java:100)
at
org.apache.cassandra.db.partitions.PurgingPartitionIterator.hasNext(PurgingPartitionIterator.java:72)
at
org.apache.cassandra.db.compaction.CompactionIterator.hasNext(CompactionIterator.java:223)
at
org.apache.cassandra.db.compaction.CompactionTask.runMayThrow(CompactionTask.java:177)
at org.apache.cassandra.utils.WrappedRunnable.run(WrappedRunnable.java:28)
at
org.apache.cassandra.db.compaction.CompactionTask.executeInternal(CompactionTask.java:78)
at
org.apache.cassandra.db.compaction.AbstractCompactionTask.execute(AbstractCompactionTask.java:60)
at
org.apache.cassandra.db.compaction.CompactionManager$8.runMayThrow(CompactionManager.java:539)
at org.apache.cassandra.utils.WrappedRunnable.run(WrappedRunnable.java:28)
at java.util.concurrent.Executors$RunnableAdapter.call(Executors.java:511)
at java.util.concurrent.FutureTask.run(FutureTask.java:266)
at
java.util.concurrent.ThreadPoolExecutor.runWorker(ThreadPoolExecutor.java:1142)
at
java.util.concurrent.ThreadPoolExecutor$Worker.run(ThreadPoolExecutor.java:617)
at java.lang.Thread.run(Thread.java:745)
{code}
It will only occur when upgrading thru 2.2. Going from 2.1 -> 3.0 will not
surface the issue. It can be reproduced with the following (note -- changing
jdks will need to be altered if you aren't on OSX)
{code}
#!/bin/sh
echo "using java7 for cassandra-2.1 instance"
export JAVA_HOME=$(/usr/libexec/java_home -v 1.7)
ccm stop
ccm remove upgrade_nodes
ccm create -n 1 -v git:cassandra-2.1 upgrade_nodes
ccm start
ccm node1 stress write n=5K -rate threads=4 -mode native cql3
sleep 10
for cver in 2.2 3.0
do
echo "Draining all nodes"
ccm node1 nodetool drain
ccm stop
echo "switching to java 8"
export JAVA_HOME=$(/usr/libexec/java_home -v 1.8)
echo "Upgrading to git:cassandra-$cver"
ccm setdir -v git:cassandra-$cver
ccm start
echo "Sleeping to all version migrations"
sleep 30
echo "Upgrading sstables"
ccm node1 nodetool upgradesstables
ccm node1 stress write n=5K -rate threads=4 -mode native cql3
sleep 10
done
echo "***********"
echo "instead of creating churn to cause compaction naturally forcing
compaction of system keyspace"
echo "***********"
ccm node1 nodetool compact system
ccm stop
{code}
The example uses {{nodetool compact system}} but it will also occur with
{{nodetool upgradesstables system}}. I'm puzzled by that since the script runs
{{upgradesstables}} on each iteration. Is the system keyspace not effected by
the command without arguments?
Ran against:
2.1: {{e889ee408bec5330c312ff6b72a81a0012fdf2a5}}
2.2: {{e63dacf793fedc8a9eed9c7fc635cde5f9fd68f3}}
3.0: {{9218d7456b36d20ebe78bab23594e67d2f0c4a20}}
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
