A Markov created CASSANDRA-10138:
------------------------------------
Summary: Millions of compaction tasks on empty DB
Key: CASSANDRA-10138
URL: https://issues.apache.org/jira/browse/CASSANDRA-10138
Project: Cassandra
Issue Type: Bug
Environment: CentOS 6.5 and Cassandra 2.1.8
Reporter: A Markov
Fresh installation of 2.1.8 Cassandra with no data in the database except
systems tables becomes unresponsive after about 5-10 minutes from the start.
Initially problem was discovered on empty cluster of 12 nodes because of the
creation schema error - script was exiting by timeout giving an error.
Analysis of log files showed that nodes were constantly reported as DOWN and
then after some period of time UP. That was reported for multiple nodes.
Verification of the system.log file showed that nodes constantly perform GC and
while doing that all cores of the system were 100% busy which caused node
disconnect after some time.
Further analysis with nodetool (tpstats option) showed us that just after 10
minutes since clean node restart node completed more then 47M compaction tasks
and had more then 12M pending. Here is example of the output:
nodetool tpstats
Pool Name Active Pending Completed Blocked All
time blocked
CounterMutationStage 0 0 0 0
0
ReadStage 0 0 0 0
0
RequestResponseStage 0 0 0 0
0
MutationStage 0 0 257 0
0
ReadRepairStage 0 0 0 0
0
GossipStage 0 0 0 0
0
CacheCleanupExecutor 0 0 0 0
0
MigrationStage 0 0 0 0
0
ValidationExecutor 0 0 0 0
0
Sampler 0 0 0 0
0
MemtableReclaimMemory 0 0 8 0
0
InternalResponseStage 0 0 0 0
0
AntiEntropyStage 0 0 0 0
0
MiscStage 0 0 0 0
0
CommitLogArchiver 0 0 0 0
0
MemtableFlushWriter 0 0 8 0
0
PendingRangeCalculator 0 0 1 0
0
MemtablePostFlush 0 0 44 0
0
CompactionExecutor 0 12996398 47578625 0
0
AntiEntropySessions 0 0 0 0
0
HintedHandoff 0 1 2 0
0
I am repeating myself but that was on TOTALLY EMPTY DB after 10 minutes since
cassandra was started.
I was able to repeateadly reproduce same issue and behaviour with single
cassandra instance. Issue was persistent after I did full cassandra wipe out
and reinstall from repository.
I discovered that issue dissipaters if I execute
nodetool disableautocompaction
in that case system quickly (in a matter of 20-30 seconds) goes though all
pending tasks and becomes idle. If I enable autocompaction again in about 1
minute it jumps to millions of pending tasks again.
I verified it on the save server with version of Cassandra 2.1.6 and issue was
not present.
logs file do not show any ERROR messages. There were only warnings about GC
events that were taking too long.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
