Gary Tully created AMQ-6644:
-------------------------------
Summary: Incorrect logging from KahaDB cleanup task when
enableAckCompaction=true
Key: AMQ-6644
URL: https://issues.apache.org/jira/browse/AMQ-6644
Project: ActiveMQ
Issue Type: Bug
Components: KahaDB
Affects Versions: 5.14.0
Reporter: Gary Tully
Assignee: Gary Tully
Priority: Minor
Fix For: 5.15.0
When KahaDB is configured for enableAckCompaction=true, it moves acks into a
new journal file. Such journal file will only contains the compacted acks, it
won't be used to hold messages.
If the actual journal (to which new messages are written to) has a lower number
than the journal files that were created during ack compaction, the periodic
cleanup task will not delete any journals that are higher than the actual
journal file. So multiple journal files may remain active on disk although
there is no single unconsumed message on the broker.
This in itself is okay, however when trace logging for the cleanup task is
enabled, it reports differently, namely that it is going to delete these
journals, where in fact it is not deleting them.
E.g. lets take the following example.
The KahaDB folder on disk consists of
{code}
[kahadb]$ ls -alh
total 54M
drwxr-xr-x. 2 fuse fuse 128K Feb 1 15:50 .
drwxr-xr-x. 13 fuse fuse 4.0K Nov 4 13:14 ..
-rw-r--r--. 1 fuse fuse 32M Feb 1 16:26 db-65.log
-rw-r--r--. 1 fuse fuse 4.6M Feb 1 15:24 db-66.log
-rw-r--r--. 1 fuse fuse 4.5M Feb 1 15:29 db-67.log
-rw-r--r--. 1 fuse fuse 4.6M Feb 1 15:34 db-68.log
-rw-r--r--. 1 fuse fuse 4.5M Feb 1 15:39 db-69.log
-rw-r--r--. 1 fuse fuse 2.5M Feb 1 16:26 db.data
-rw-r--r--. 1 fuse fuse 32M Feb 1 14:51 db-log.template
-rw-r--r--. 1 fuse fuse 1002K Feb 1 16:26 db.redo
-rw-r--r--. 1 fuse fuse 8 Feb 1 14:51 lock
{code}
and the logging says:
{code}
Last update: 65:26636520, full gc candidates set: [65, 66, 67, 68, 69]
gc candidates after producerSequenceIdTrackerLocation:65, [66, 67, 68, 69]
gc candidates after ackMessageFileMapLocation:65, [66, 67, 68, 69]
...
gc candidates: [66, 67, 68, 69]
ackMessageFileMap: {65=[65]}
Cleanup removing the data files: [66, 67, 68, 69]
{code}
In this example the actual journal file to which msgs are written is 65. The
journal files 66-69 were created during ack compaction and have a higher number
than 65.
So KahaDB won't delete the journals 66-69 until the actual journal file is
moved to journal 70, despite that there are no unconsumed messages on the
broker.
However the last log line suggests that it will remove the journals 66-69, but
they will not get removed due to rule above.
We should align the logging output with the logic used to determine which
journals to delete.
--
This message was sent by Atlassian JIRA
(v6.3.15#6346)
