[
https://issues.apache.org/jira/browse/CASSANDRA-16253?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Alain RODRIGUEZ updated CASSANDRA-16253:
----------------------------------------
Description:
About the documentation on commit logs:
* Commit logs default is every 10 sec - *periodic*, not batch
[https://github.com/apache/cassandra/blob/cassandra-3.11.8/conf/cassandra.yaml#L385-L386]
(same in 3.11.9 or trunk). It can lead to data losses but still makes sense to
me because fsync-ing every write is very expensive, and (even in batches)
should be reserved for the super paranoid where handling multiple power-out (or
sudden disk failures) simultaneously on multiple nodes needs to be handled (as
[~mck] explained to me.
* Nonetheless, apart from the theory behind it, in a more practical
perspective, *the doc is wrong* and says *Default Value: batch* which mislead
me and this could happen to more people:
[https://cassandra.apache.org/doc/latest/architecture/storage_engine.html#commit-log]
I think this is important enough for us to quickly fix the documentation in
trunk. As this information is fully missing in the 3.11 branch documentation, I
suggest we add it as well on that branch (based on information about commit
logs from trunk).
was:
About the documentation on commit logs:
* Commit logs default is every 10 sec (periodic), not batch
[https://github.com/apache/cassandra/blob/cassandra-3.11.8/conf/cassandra.yaml#L385-L386]
(same in 3.11.9 or trunk). It can lead to data losses but still makes sense to
me because fsync-ing every write is very expensive, and (even in batches)
should be reserved for the super paranoid where handling multiple power-out (or
sudden disk failures) simultaneously on multiple nodes needs to be handled (as
[~mck] explained to me.
* Nonetheless, apart from the theory behind it, in a more practical
perspective, *the doc is wrong* and says default is `_Default Value:_ batch`
which mislead me and this could happen to more people:
[https://cassandra.apache.org/doc/latest/architecture/storage_engine.html#commit-log]
I think this is important enough for us to quickly fix the documentation in
trunk. As this information is fully missing in the 3.11 branch documentation, I
suggest we add it as well on that branch (based on information about commit
logs from trunk).
> Fix documentation to reflect real defaults for commit logs
> ----------------------------------------------------------
>
> Key: CASSANDRA-16253
> URL: https://issues.apache.org/jira/browse/CASSANDRA-16253
> Project: Cassandra
> Issue Type: Task
> Reporter: Alain RODRIGUEZ
> Assignee: Alain RODRIGUEZ
> Priority: Normal
>
> About the documentation on commit logs:
> * Commit logs default is every 10 sec - *periodic*, not batch
> [https://github.com/apache/cassandra/blob/cassandra-3.11.8/conf/cassandra.yaml#L385-L386]
> (same in 3.11.9 or trunk). It can lead to data losses but still makes sense
> to me because fsync-ing every write is very expensive, and (even in batches)
> should be reserved for the super paranoid where handling multiple power-out
> (or sudden disk failures) simultaneously on multiple nodes needs to be
> handled (as [~mck] explained to me.
> * Nonetheless, apart from the theory behind it, in a more practical
> perspective, *the doc is wrong* and says *Default Value: batch* which mislead
> me and this could happen to more people:
> [https://cassandra.apache.org/doc/latest/architecture/storage_engine.html#commit-log]
>
> I think this is important enough for us to quickly fix the documentation in
> trunk. As this information is fully missing in the 3.11 branch documentation,
> I suggest we add it as well on that branch (based on information about commit
> logs from trunk).
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
