[
https://issues.apache.org/jira/browse/CASSANDRA-1155?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Jonathan Ellis updated CASSANDRA-1155:
--------------------------------------
Attachment: 1155-v4.txt
v4 moves the write to stats CF into postFlushExecutor to avoid deadlock.
the deadlock was, you have two CFs flushing, A and B.
flush of A goes to update stats CF
meanwhile flush of B has acquired flusherlock in preparation to flush, so A
can't aquire the lock to do the stats write
but B can't release the lock, because we only allow N flushes at a time, so it
blocks on the executor submission (to flush-writer-pool)
so neither A nor B can proceed
> keep persistent row statistics
> ------------------------------
>
> Key: CASSANDRA-1155
> URL: https://issues.apache.org/jira/browse/CASSANDRA-1155
> Project: Cassandra
> Issue Type: Sub-task
> Components: Core
> Reporter: Jonathan Ellis
> Assignee: Brandon Williams
> Fix For: 0.7 beta 1
>
> Attachments: 1155-v2.txt, 1155-v3.txt, 1155-v4.txt, 1155.txt
>
>
> during flush and compaction we should keep row size statistics using
> EstimatedHistogram (column count, and row size), replacing min/max/total
> sizes in CFS.
> having this detail will let us estimate, given an index CF, how many nodes we
> need to query to get the number of matching rows requested by the client.
--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.
