Replies are inlined below:

On Thu, Mar 13, 2014 at 9:40 AM, Matthew MacClary
<[email protected]> wrote:
> Interesting, so the Java client is just honoring the server's request that
> it try back in 60 seconds?

No, again, the timeout is on the Riak side. There is no timing in the
client, it's just doing a blocking read on a socket (with no timeout).
Riak sends an error message to the client when a timeout occurs and
the default behavior of the client is to retry all failed operations.

> Do you happen to know if there is a configuration
> variable to tune that default timeout value?

The operations in the client have an optional timeout to send with the
request that will override the default setting on the Riak side for
that operation. See:
http://basho.github.io/riak-java-client/1.4.4/com/basho/riak/client/operations/FetchObject.html#timeout(int)

> Our automated testing since yesterday has shown that changing pb_backlog did
> resolve this 60 second timeout issue for our system.

If this is the case then it seems what you're talking about isn't
related to operation timeouts at all but rather TCP connection
timeouts.

I do find it odd that the change you made would resolve any sort of
TCP timeout issue. AFAIK the pb_backlog is simply how many TCP
connections can be queued waiting to be accepted (the same as you
would pass directly to the listen() system call).

If you performed an operation with the client and it needed to create
a new TCP connection to Riak, you would expect the client connection
to be refused instantly if the listen queue was full. While the client
is going to retry making that connection, it certainly wouldn't take
60 seconds to do so - I'd expect all retries (3, by default) to fail
instantly and the client to throw an exception back to you saying it
couldn't connect.

Connections that were *in* the listen queue I could see causing a TCP
connection timeout if Riak never accepted them before the client
side's TCP stack gave up, but in my head your change would increase
the frequency of that happening rather than reducing it.

- Roach

_______________________________________________
riak-users mailing list
[email protected]
http://lists.basho.com/mailman/listinfo/riak-users_lists.basho.com

Reply via email to