[ https://issues.apache.org/jira/browse/CASSANDRA-12763?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15560791#comment-15560791 ]
Jeff Jirsa commented on CASSANDRA-12763: ---------------------------------------- As mentioned in IRC, I *THINK* that ultimately the problem is that this loop: https://github.com/jeffjirsa/cassandra/blob/trunk/src/java/org/apache/cassandra/db/lifecycle/Helpers.java#L130 Calls {{listFiles()}} for each sstablereader in the transaction. Since {{listFiles}} is notoriously slow when there are a ton of files in the directory, and you call it N times, you end up waiting at the end of the compaction preparing to commit/marking files for deletion. It may be worth someone investigating whether or not we can avoid the full directory scan N times - due to nature of that code, it may not be safe to cache the directory listing, but it's worth looking into. > Compaction performance issues when a table has a lot of sstables > ---------------------------------------------------------------- > > Key: CASSANDRA-12763 > URL: https://issues.apache.org/jira/browse/CASSANDRA-12763 > Project: Cassandra > Issue Type: Bug > Components: Compaction > Reporter: Tom van der Woerdt > > An issue with a script flooded my cluster with sstables. There is now a table > with 100k sstables, all on the order of KBytes, and it's taking a long time > (ETA 20 days) to compact, even though the table is only ~30GB. > Stack trace : > {noformat} > "CompactionExecutor:269" #7536 daemon prio=1 os_prio=4 tid=0x00007f4acd40fc00 > nid=0x14f8 runnable [0x00007f4798436000] > java.lang.Thread.State: RUNNABLE > at java.io.UnixFileSystem.list(Native Method) > at java.io.File.list(File.java:1122) > at java.io.File.listFiles(File.java:1248) > at > org.apache.cassandra.db.lifecycle.LogRecord.getExistingFiles(LogRecord.java:268) > at org.apache.cassandra.db.lifecycle.LogRecord.make(LogRecord.java:150) > at > org.apache.cassandra.db.lifecycle.LogFile.makeRecord(LogFile.java:293) > at org.apache.cassandra.db.lifecycle.LogFile.add(LogFile.java:283) > at > org.apache.cassandra.db.lifecycle.LogTransaction.obsoleted(LogTransaction.java:158) > at > org.apache.cassandra.db.lifecycle.Helpers.prepareForObsoletion(Helpers.java:134) > at > org.apache.cassandra.db.lifecycle.LifecycleTransaction.doPrepare(LifecycleTransaction.java:193) > at > org.apache.cassandra.utils.concurrent.Transactional$AbstractTransactional.prepareToCommit(Transactional.java:173) > at > org.apache.cassandra.io.sstable.SSTableRewriter.doPrepare(SSTableRewriter.java:376) > at > org.apache.cassandra.utils.concurrent.Transactional$AbstractTransactional.prepareToCommit(Transactional.java:173) > at > org.apache.cassandra.db.compaction.writers.CompactionAwareWriter.doPrepare(CompactionAwareWriter.java:84) > at > org.apache.cassandra.utils.concurrent.Transactional$AbstractTransactional.prepareToCommit(Transactional.java:173) > at > org.apache.cassandra.utils.concurrent.Transactional$AbstractTransactional.finish(Transactional.java:184) > at > org.apache.cassandra.db.compaction.writers.CompactionAwareWriter.finish(CompactionAwareWriter.java:94) > at > org.apache.cassandra.db.compaction.CompactionTask.runMayThrow(CompactionTask.java:194) > 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:61) > at > org.apache.cassandra.db.compaction.CompactionManager$BackgroundCompactionCandidate.run(CompactionManager.java:263) > at > java.util.concurrent.Executors$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) > {noformat} > listFiles is being called over and over, apparently scaling with the number > of files in the compaction. -- This message was sent by Atlassian JIRA (v6.3.4#6332)