You are viewing a plain text version of this content. The canonical link for it is here.
Posted to dev@cassandra.apache.org by Chris Burroughs <ch...@gmail.com> on 2011/04/24 04:41:26 UTC
General usefulness of "write only mode"
I wanted to see if this was generally useful before I just made a ticket.
The scenario is a read heavy workload. Running a major compaction or
cleanup significantly reduces throughput to the point that the
compaction drags on for a long time, performance dramatically degrades,
etc. I can disablegossip which effectively disables all reads and
writes while I run those tasks offline. But if my goal was to run
cleanup across the cluster (say after another node was out for an
extended period of time), this is a somewhat ironic proposition.
What I wish I could do is disable reads but not write to the nodes.
This would allow major compaction/repairs to proceed in a reasonable
manner without hints building up on other nodes. Does this seem useful
enough for inclusion, or is there a better way to solve the same problem?
Re: General usefulness of "write only mode"
Posted by Jonathan Ellis <jb...@gmail.com>.
It's not crazy, but the implementation would be a little involved (see
Brandon's comments to
https://issues.apache.org/jira/browse/CASSANDRA-1108).
On Sat, Apr 23, 2011 at 9:41 PM, Chris Burroughs
<ch...@gmail.com> wrote:
> I wanted to see if this was generally useful before I just made a ticket.
>
> The scenario is a read heavy workload. Running a major compaction or
> cleanup significantly reduces throughput to the point that the
> compaction drags on for a long time, performance dramatically degrades,
> etc. I can disablegossip which effectively disables all reads and
> writes while I run those tasks offline. But if my goal was to run
> cleanup across the cluster (say after another node was out for an
> extended period of time), this is a somewhat ironic proposition.
>
> What I wish I could do is disable reads but not write to the nodes.
> This would allow major compaction/repairs to proceed in a reasonable
> manner without hints building up on other nodes. Does this seem useful
> enough for inclusion, or is there a better way to solve the same problem?
>
--
Jonathan Ellis
Project Chair, Apache Cassandra
co-founder of DataStax, the source for professional Cassandra support
http://www.datastax.com