cassandra-user mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Wayne <wav...@gmail.com>
Subject Re: Read Latency
Date Sat, 16 Oct 2010 11:50:34 GMT
While doing a read I get a TApplicationException: Internal Error processing
get_slice.

On Fri, Oct 15, 2010 at 5:49 PM, Jonathan Ellis <jbellis@gmail.com> wrote:

> On Fri, Oct 15, 2010 at 2:21 PM, Wayne <wav100@gmail.com> wrote:
> > The optimization definitely shaved off some time. Now it is running about
> 3x
> > CFSTATS reported time. Below are the logs.
> >
> > There is a ~300ms time frame after the last ResponseVerbHandler prior to
> the
> > resolver starting. Based on a quorum read the response resolver should
> kick
> > after 2 reads come in correct? That would mean it waited 500ms before
> > starting. Where is this time going?
>
> It's going to deserialize the replies to see if it has both the data
> and enough digests to call resolve().  Then resolve deserializes them
> a 2nd time, so there's an easy win there caching the first
> deserialize, in the patch attached.  (Applies on top of the previous
> one, which has been committed to the 0.6 svn branch at
> https://svn.apache.org/repos/asf/cassandra/branches/cassandra-0.6.)
>
> > There is still the 3+ second delay between the last 2 entries. Is this
> the
> > thrift server?
>
> It's converting the reply from Cassandra's internal representation to
> thrift, and sending it to the client.  I suspect most of the time is
> the actual sending/waiting for the client to read part.  Patch 2 also
> includes a debug statement after the convert-to-thrift stage to
> verify.
>
> --
> Jonathan Ellis
> Project Chair, Apache Cassandra
> co-founder of Riptano, the source for professional Cassandra support
> http://riptano.com
>

Mime
View raw message