Sean -

Unfortunately we do not see any errors logged in any of the files
(console.log.x, erlang.log.x, error.log.x and crash.log.x). Is there
another level of logging we should enable? (note: I sent our log files
directly to your basho email address).

Hosting environment: AWS EC2 via VPC and dedicated instances
OS: CentOS release 6.4 (Final)
Infrastructure: No servers between the client and the Riak servers.

Jason


On Thu, Apr 17, 2014 at 11:57 AM, Sean Cribbs <[email protected]> wrote:

> Jason,
>
> It's really hard to tell what's going on without more information about
> the environment.
>
> Reviewing the manpage for recv (the C API behind socket.recv), you can see
> that the call returns 0 if the connection is closed. Python interprets this
> as an empty buffer rather than an error condition.
>
> Now, why would the server-side close the connection? The only thing I can
> think of is that there is an unhandled error path in the server code such
> that the Erlang process serving the requests exits abnormally, and the
> linked server socket shuts down as a result.  This is why I asked about the
> error logs from your Riak nodes.
>
> So, any additional information you can give would be helpful, especially
> what your deployment environment looks like. For example, are you on a
> cloud provider or your own hardware? What OS/distribution are you using?
> What sorts of other infrastructure is deployed for your app?
>
>
> On Thu, Apr 17, 2014 at 12:43 PM, Jason Wang <[email protected]> wrote:
>
>> Sean,
>>
>> RE http vs pbc port: Verified that our production environment is using
>> the pbc protocol on port 8087.
>>
>> RE crashlog: There are no entries in console.log, crash.log and error.log
>> for over a month.
>>
>> RE HAprodxy: Our clients are connecting to the Riak servers. No reverse
>> proxy systems are in between.
>>
>> Anything else I can look into?
>>
>> Jason
>>
>>
>> On Wed, Apr 16, 2014 at 1:02 PM, Sean Cribbs <[email protected]> wrote:
>>
>>> Hi Jason,
>>>
>>> We usually see these errors when someone connects to the HTTP port
>>> instead of the PB port. First, check that your client is configured
>>> correctly. Also, please check the server logs for any crashes related to
>>> riak_api_pb_server, and paste those here if present.
>>>
>>> It is also conceivable this could occur if the socket were closed before
>>> anything were sent, that is, the read buffer on it is empty, so it returns
>>> 0. I don't know of any specific reason that might happen unless you are
>>> using haproxy between the client and the server (see related PR/issue
>>> https://github.com/basho/riak_api/pull/54).
>>>
>>>
>>> On Tue, Apr 15, 2014 at 6:34 PM, Jason Wang <[email protected]> wrote:
>>>
>>>> HI all,
>>>>
>>>> In production, we are experiencing "Socket returned short packet length
>>>> 0 - expected 4" exceptions whenever we try to store an object >20K in size.
>>>> In addition, the exception typically takes over 60 seconds to manifest. The
>>>> content of each object is an bytearray.
>>>>
>>>> Any idea what could be causing this exception?
>>>>
>>>> Other details:
>>>> Library: Python
>>>> Version:  riak==2.0.1, riak-pb==1.4.1.1
>>>> Protocol: pbc
>>>> Steps to reproduce: N/A. This only happens in production, not on dev
>>>> machines.
>>>>
>>>> Thanks in advance,
>>>>
>>>> Jason
>>>>
>>>> _______________________________________________
>>>> riak-users mailing list
>>>> [email protected]
>>>> http://lists.basho.com/mailman/listinfo/riak-users_lists.basho.com
>>>>
>>>>
>>>
>>>
>>> --
>>> Sean Cribbs <[email protected]>
>>> Software Engineer
>>> Basho Technologies, Inc.
>>> http://basho.com/
>>>
>>
>>
>> _______________________________________________
>> riak-users mailing list
>> [email protected]
>> http://lists.basho.com/mailman/listinfo/riak-users_lists.basho.com
>>
>>
>
>
> --
> Sean Cribbs <[email protected]>
> Software Engineer
> Basho Technologies, Inc.
> http://basho.com/
>
_______________________________________________
riak-users mailing list
[email protected]
http://lists.basho.com/mailman/listinfo/riak-users_lists.basho.com

Reply via email to