Jeremiah Jordan created CASSANDRA-7640:
------------------------------------------
Summary: Seeing commit log grow unbounded
Key: CASSANDRA-7640
URL: https://issues.apache.org/jira/browse/CASSANDRA-7640
Project: Cassandra
Issue Type: Bug
Components: Core
Reporter: Jeremiah Jordan
Priority: Minor
While running 2.0.6 saw the commit log size grow up to 30 GB before restarting
the node, with the max set to 1 GB.
With DEBUG logging on CommitLog* we see segments are not being removed because
they are still dirty. But for some reason the force flush when the active size
is too large isn't happening (The active size gets printed out, and is 20+GB).
The logs show normal "high volume" flushes happening, and force flushes of
system.* stuff happening. But there are no flushes occurring for the memtables
that are keeping the segments dirty.
Upon restart the commit log replays fine with no errors. There are no errors
from flushing to the post flusher. tpstats for both show 0 active 0 pending at
times during.
The only config change around this is that the segment size is larger than
normal, it is set to 128 MB, as there are some very large columns being
inserted (30+MB).
--
This message was sent by Atlassian JIRA
(v6.2#6252)
