Details
-
Bug
-
Status: Resolved
-
Normal
-
Resolution: Duplicate
-
None
-
None
-
None
-
Normal
Description
While doing some tests on 2.2.1 I found a problem with a null value in the compaction_history table that results in a nullpointer exception when doing 'nodetool compactionhistory'.
$ ccm node1 nodetool compactionhistory
Compaction History:
error: null
– StackTrace –
java.lang.NullPointerException
at com.google.common.base.Joiner$MapJoiner.join(Joiner.java:330)
at org.apache.cassandra.utils.FBUtilities.toString(FBUtilities.java:477)
at org.apache.cassandra.db.compaction.CompactionHistoryTabularData.from(CompactionHistoryTabularData.java:78)
at org.apache.cassandra.db.SystemKeyspace.getCompactionHistory(SystemKeyspace.java:425)
at org.apache.cassandra.db.compaction.CompactionManager.getCompactionHistory(CompactionManager.java:1492)
The problem is the null value in the rows_merged column.
cqlsh> SELECT * FROM system.compaction_history ;
id | bytes_in | bytes_out | columnfamily_name | compacted_at | keyspace_name | rows_merged
-------------------------------------------------------------------------------------------------------------------------------------
86548100-5adf-11e5-a4f4-29d398af617a | 366 | 64 | compactions_in_progress | 2015-09-14 12:52:58+0000 | system | {2: 2}
9ff24610-5adf-11e5-a4f4-29d398af617a | 30 | 0 | bar | 2015-09-14 12:53:41+0000 | foo | null
e4ffb130-5ade-11e5-a4f4-29d398af617a | 366 | 64 | compactions_in_progress | 2015-09-14 12:48:27+0000 | system | {2: 2}
After some more testing and digging in the code I realised that this is the result of a compaction where all sstables are dropped (i.e. no new sstable is created).
I tried to recreate this using 2.1.9 and found that there where nothing inserted into the compaction_history table if the same thing happens. That is due to a check in CompactionTask.runMayThrow() line 181.
if (!iter.hasNext()) { // don't mark compacted in the finally block, since if there _is_ nondeleted data, // we need to sync it (via closeAndOpen) first, so there is no period during which // a crash could cause data loss. cfs.markObsolete(sstables, compactionType); return; }
The result of the return here is that we end the compaction without inserting anything in the compaction_history table, 2.2.1 does not have this logic and will insert the null value.
Not sure what the best solution would be:
1. Implement the same behaviour as in 2.1.9 and skip inserting anything into compaction_history.
2. Make sure we insert a more sensible value then null, but I'm not sure what that value would be.
3. Have nodetool handle the null value without a nullpointer exception.
Attachments
Issue Links
- is duplicated by
-
CASSANDRA-9758 nodetool compactionhistory NPE
- Resolved