Is that normal to have some percent of reads/writes time out?

[Ken Sandney <bl...@gmail.com>]

Hi

I am doing some load test with 4 nodes cluster. My client is PHP. I found
some reads/writes were time out no matter how I tuned the parameters. These
time-outs could be caught by client code. My question is: are these
time-outs normal even in production environment? Should they be treated as
normal cases or kind of error?

Re: Is that normal to have some percent of reads/writes time out?

[Miguel Verde <mi...@gmail.com>]

I see that you are aware of https://issues.apache.org/jira/browse/THRIFT-347
Have you applied the patch there?  It worked for the Digg guys (probably the
largest PHP user of Cassandra) and others in that JIRA issue.
Timeouts are typical with unusually heavy load, node failure, and/or
un-tuned parameters, but that doesn't sound like your situation.
On Thu, Apr 22, 2010 at 8:22 AM, Ken Sandney <bl...@gmail.com> wrote:

> Hi
>
> I am doing some load test with 4 nodes cluster. My client is PHP. I found
> some reads/writes were time out no matter how I tuned the parameters. These
> time-outs could be caught by client code. My question is: are these
> time-outs normal even in production environment? Should they be treated as
> normal cases or kind of error?
>

Re: Is that normal to have some percent of reads/writes time out?

[Ken Sandney <bl...@gmail.com>]

By the way, my testing cluster are 4 normal PCs with 2GB RAM assigned to
JVM, Intel(R) Celeron(R) CPU E3200 2.40GHz. How many concurrent reads/writes
should be reasonable? Or how much memory/CPU usage would be healthy for this
kind of test cluster?

Re: Is that normal to have some percent of reads/writes time out?

[Ken Sandney <bl...@gmail.com>]

yes, I've tried the patch on
https://issues.apache.org/jira/browse/THRIFT-347, but seems not work for me.
I doubt I am involving another issue with Thrift. If my column value size is
more than 8KB(with thrift php extension enabled), my client has more chances
to get "timed out error". I am still working on this issue to figure out
what happend there.

I am new to this and any clues/advices are welcome.

Re: Is that normal to have some percent of reads/writes time out?

[Jonathan Ellis <jb...@gmail.com>]

timeouts are usually a signal that you need to add capacity to handle
the load you are giving the cluster.

On Thu, Apr 22, 2010 at 8:22 AM, Ken Sandney <bl...@gmail.com> wrote:
> Hi
>
> I am doing some load test with 4 nodes cluster. My client is PHP. I found
> some reads/writes were time out no matter how I tuned the parameters. These
> time-outs could be caught by client code. My question is: are these
> time-outs normal even in production environment? Should they be treated as
> normal cases or kind of error?
>