You are viewing a plain text version of this content. The canonical link for it is here.
Posted to user@cassandra.apache.org by Daniel Kleviansky <da...@kleviansky.com> on 2016/10/27 00:25:09 UTC
High Usage of Survivor Space
Hi everyone,
Organisation is running Cassandra for Windows v2.2.5
One of our development (non-load testing) clusters has a total of 6 nodes
across two DCs, with an RF = 3:3.
Each node has a total of 16GB of memory, and no JVM options have been
modified.
We're seeing what I believe to be unusually high usage of survivor space
across all nodes in the cluster. Some sit at a consistently 100% maxed out
state, while others are found at about 90%.
Initial thoughts is this is linked to high number of keyspaces/column
families running on this particular cluster. There is a total of 28
keyspaces and 412 column families. I have no empirical evidence to support
this theory, hence me reaching out to the mailing list.
Is there any way to verify this theory, and/or discover potential root
causes?
Please let me know what other information I can provide, and I'll be sure
to get it ASAP.
Kindest regards,
Daniel Kleviansky
Re: High Usage of Survivor Space
Posted by Vladimir Yudovin <vl...@winguzone.com>.
Hi,
did this high memory usage caused the problems? OOM crashes, GC pauses?
Best regards, Vladimir Yudovin,
Winguzone - Hosted Cloud Cassandra
Launch your cluster in minutes.
---- On Wed, 26 Oct 2016 21:51:26 -0400Daniel Kleviansky <daniel@kleviansky.com> wrote ----
Before anyone wastes any time replying, I found out that most of these were not being used, and manage to drop certain keyspaces and reduce the count of column families to 158.
Thought still a fair number, this seems to have relieved the issue significantly!
Hope this helps someone out at some point, otherwise, sorry for the unnecessary emails. ;)
Daniel
On Thu, Oct 27, 2016 at 11:25 AM, Daniel Kleviansky <daniel@kleviansky.com> wrote:
--
Daniel Kleviansky
System Engineer & CX Consultant
M: +61 (0) 499 103 043 | E: daniel@kleviansky.com | W: http://danielkleviansky.com
Hi everyone,
Organisation is running Cassandra for Windows v2.2.5
One of our development (non-load testing) clusters has a total of 6 nodes across two DCs, with an RF = 3:3.
Each node has a total of 16GB of memory, and no JVM options have been modified.
We're seeing what I believe to be unusually high usage of survivor space across all nodes in the cluster. Some sit at a consistently 100% maxed out state, while others are found at about 90%.
Initial thoughts is this is linked to high number of keyspaces/column families running on this particular cluster. There is a total of 28 keyspaces and 412 column families. I have no empirical evidence to support this theory, hence me reaching out to the mailing list.
Is there any way to verify this theory, and/or discover potential root causes?
Please let me know what other information I can provide, and I'll be sure to get it ASAP.
Kindest regards,
Daniel Kleviansky
Re: High Usage of Survivor Space
Posted by Daniel Kleviansky <da...@kleviansky.com>.
Before anyone wastes any time replying, I found out that most of these were
not being used, and manage to drop certain keyspaces and reduce the count
of column families to 158.
Thought still a fair number, this seems to have relieved the issue
significantly!
Hope this helps someone out at some point, otherwise, sorry for the
unnecessary emails. ;)
Daniel
On Thu, Oct 27, 2016 at 11:25 AM, Daniel Kleviansky <da...@kleviansky.com>
wrote:
> Hi everyone,
>
> Organisation is running Cassandra for Windows v2.2.5
> One of our development (non-load testing) clusters has a total of 6 nodes
> across two DCs, with an RF = 3:3.
> Each node has a total of 16GB of memory, and no JVM options have been
> modified.
>
> We're seeing what I believe to be unusually high usage of survivor space
> across all nodes in the cluster. Some sit at a consistently 100% maxed out
> state, while others are found at about 90%.
>
> Initial thoughts is this is linked to high number of keyspaces/column
> families running on this particular cluster. There is a total of 28
> keyspaces and 412 column families. I have no empirical evidence to support
> this theory, hence me reaching out to the mailing list.
>
> Is there any way to verify this theory, and/or discover potential root
> causes?
>
> Please let me know what other information I can provide, and I'll be sure
> to get it ASAP.
>
> Kindest regards,
> Daniel Kleviansky
>
--
Daniel Kleviansky
System Engineer & CX Consultant
M: +61 (0) 499 103 043 | E: daniel@kleviansky.com | W:
http://danielkleviansky.com