[CASSANDRA-14069] Node stopped serving write requests when a table has a lot of sstables

• Type: Bug
• Status: Resolved
• Priority: Normal
• Resolution: Fixed
• Component/s: Local/Compaction
• Severity: Normal
• Assignee: Sergey Lapukhov
• Reporter: Sergey Lapukhov

Authors

Sergey Lapukhov

Reviewers

Marcus Eriksson

• Created: 23/Nov/17 16:52
• Updated: 08/Sep/20 11:23
• Resolved: 08/Sep/20 11:23

Description

Cluster was flooded with SSTables. A table had ~20000 sstables. Write requests started failing.

Steps to reproduce:

• Create cluster with 3 nodes
• Specify

  memtable_heap_space_in_mb: 10
  

  in cassandra.yaml

• Create table standard1 in keyspace1 (for the cassandra-stress tool) with the script create.cql. Please note

   compaction = {'class': 'SizeTieredCompactionStrategy', 'enabled': 'false'} 

  i.e. compaction will be turned off for now.

• Populate node with data:

   cassandra-stress write n=1000000000 -node 127.0.0.1 

• After node was populated, put both read and write pressure on it:

    cassandra-stress read n=1000000000 -node 127.0.0.1 
    cassandra-stress write n=1000000000 -node 127.0.0.1 

• While still under pressure, enable LeveledCompactionStrategy

    echo "ALTER TABLE keyspace1.standard1 WITH compaction = { 'class' : 'LeveledCompactionStrategy', 'sstable_size_in_mb' : 1 }; DESC keyspace1.standard1; exit" | bin/cqlsh; 

Results:
Write requests failing.
'bin/nodetool cfstats' and 'bin/nodetool compactionstats' commands hanging, if issued from the node running cassandra-stress tool.

If issued from another node:

 $ bin/nodetool cfstats
...
Table: standard1
                SSTable count: 22637
                SSTables in each level: [22651/4, 0, 0, 0, 0, 0, 0, 0, 0] 
...

$ bin/nodetool compactionstats

pending tasks: 12656
                                     id   compaction type    keyspace       table   completed       total    unit   progress
   935bbc00-d03b-11e7-a47d-2b44293495b8        Compaction   keyspace1   standard1    59556014    59557860   bytes    100.00%
   a29ee660-d03b-11e7-a47d-2b44293495b8        Compaction   keyspace1   standard1    80432114   742151655   bytes     10.84%
   9766e400-d03b-11e7-a47d-2b44293495b8        Compaction   keyspace1   standard1    58891604    58893215   bytes    100.00%
   9cdc9880-d03b-11e7-a47d-2b44293495b8        Compaction   keyspace1   standard1    20289449    20290800   bytes     99.99%
   90f98910-d03b-11e7-a47d-2b44293495b8        Compaction   keyspace1   standard1    59689824    59695545   bytes     99.99%
   986ede20-d03b-11e7-a47d-2b44293495b8        Compaction   keyspace1   standard1    40598594    40598820   bytes    100.00%
   9cd322a0-d03b-11e7-a47d-2b44293495b8        Compaction   keyspace1   standard1    60756739    60766660   bytes     99.98% 

Special note about 'bin/nodetool compactionstats' - picture above is quite typical for this issue. I.e. compaction tasks manage to make it through, but hinder near the full completion (around 99.9 %).

Maybe the root of the problem is in this thread (see stack.txt):

"CompactionExecutor:1748" #4649 daemon prio=1 os_prio=4 tid=0x00007f35a0096100 nid=0x65f6 runnable [0x00007f3228bce000]
   java.lang.Thread.State: RUNNABLE
  at org.apache.cassandra.dht.AbstractBounds.<init>(AbstractBounds.java:53)
  at org.apache.cassandra.dht.Bounds.<init>(Bounds.java:42)
  at org.apache.cassandra.db.compaction.LeveledManifest.overlapping(LeveledManifest.java:562)
  at org.apache.cassandra.db.compaction.LeveledManifest.overlapping(LeveledManifest.java:549)
  at org.apache.cassandra.db.compaction.LeveledManifest.getCandidatesFor(LeveledManifest.java:624)
  at org.apache.cassandra.db.compaction.LeveledManifest.getCompactionCandidates(LeveledManifest.java:378)
  - locked <0x000000070d4c3bc8> (a org.apache.cassandra.db.compaction.LeveledManifest)
  at org.apache.cassandra.db.compaction.LeveledCompactionStrategy.getNextBackgroundTask(LeveledCompactionStrategy.java:105)
  - locked <0x000000070d6cb2c8> (a org.apache.cassandra.db.compaction.LeveledCompactionStrategy)
  at org.apache.cassandra.db.compaction.CompactionStrategyManager.getNextBackgroundTask(CompactionStrategyManager.java:102)
  - locked <0x00000006467268b8> (a org.apache.cassandra.db.compaction.CompactionStrategyManager)
  at org.apache.cassandra.db.compaction.CompactionManager$BackgroundCompactionCandidate.run(CompactionManager.java:258)
  at java.util.concurrent.Executors$RunnableAdapter.call(Executors.java:511)
  at java.util.concurrent.FutureTask.run(FutureTask.java:266)
  at java.util.concurrent.ThreadPoolExecutor.runWorker(ThreadPoolExecutor.java:1149)
  at java.util.concurrent.ThreadPoolExecutor$Worker.run(ThreadPoolExecutor.java:624)
  at org.apache.cassandra.concurrent.NamedThreadFactory.lambda$threadLocalDeallocator$0(NamedThreadFactory.java:79)
  at org.apache.cassandra.concurrent.NamedThreadFactory$$Lambda$4/1671507048.run(Unknown Source)
  at java.lang.Thread.run(Thread.java:748)

As I see it, thread is running a cycle, which has O(n^2) dependency on the number of SSTables, while still holding the lock.

Say this one:

"CompactionExecutor:1743" #4644 daemon prio=1 os_prio=4 tid=0x00007f35a01591a0 nid=0x65f1 waiting for monitor entry [0x00007f322d016000]
   java.lang.Thread.State: BLOCKED (on object monitor)
	at org.apache.cassandra.db.compaction.CompactionStrategyManager.handleNotification(CompactionStrategyManager.java:252)
	- waiting to lock <0x00000006467268b8> (a org.apache.cassandra.db.compaction.CompactionStrategyManager)
	at org.apache.cassandra.db.lifecycle.Tracker.notifyDeleting(Tracker.java:448)
	at org.apache.cassandra.db.lifecycle.LogTransaction.obsoleted(LogTransaction.java:161)
	at org.apache.cassandra.db.lifecycle.Helpers.prepareForObsoletion(Helpers.java:134)
	at org.apache.cassandra.db.lifecycle.LifecycleTransaction.doPrepare(LifecycleTransaction.java:199)
	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:206)
	at org.apache.cassandra.utils.WrappedRunnable.run(WrappedRunnable.java:28)
	at org.apache.cassandra.db.compaction.CompactionTask.executeInternal(CompactionTask.java:89)
	at org.apache.cassandra.db.compaction.AbstractCompactionTask.execute(AbstractCompactionTask.java:61)
	at org.apache.cassandra.db.compaction.CompactionManager$BackgroundCompactionCandidate.run(CompactionManager.java:264)
	at java.util.concurrent.Executors$RunnableAdapter.call(Executors.java:511)
	at java.util.concurrent.FutureTask.run(FutureTask.java:266)
	at java.util.concurrent.ThreadPoolExecutor.runWorker(ThreadPoolExecutor.java:1149)
	at java.util.concurrent.ThreadPoolExecutor$Worker.run(ThreadPoolExecutor.java:624)
	at org.apache.cassandra.concurrent.NamedThreadFactory.lambda$threadLocalDeallocator$0(NamedThreadFactory.java:79)
	at org.apache.cassandra.concurrent.NamedThreadFactory$$Lambda$4/1671507048.run(Unknown Source)
	at java.lang.Thread.run(Thread.java:748)

Issue Links

• duplicates: CASSANDRA-14826 cassandra spinning forever on 1 thread while initializing keyspace
• relates to: CASSANDRA-12763 Compaction performance issues when a table has a lot of sstables