[zeromq-dev] Delay in receive by SUB from PUB

Navneet Kumar navneet.krg at gmail.com
Mon Mar 14 14:14:21 CET 2016


Yes, I thought about that and did a profiling with "net/http/pprof".
The GC pauses are no more than 20ms at a time and usually 1-5ms.

On Mon, Mar 14, 2016 at 2:51 PM, Aaron Sokoloski <asokoloski at gmail.com>
wrote:

> Hi Navneet, since you are using Go, is it possible that it has something
> to do with garbage collections pauses?
>
> Recent versions of Go pause for much shorter times, but if you're using an
> old version it might easily stop that long to do GC.
>
> Cheers,
> Aaron
>
>
> On 14 March 2016 at 01:06, Navneet Kumar <navneet.krg at gmail.com> wrote:
>
>> Hi Jeff
>>
>> Can you please elaborate on the "over-comsume" part.
>> I am using zmq4 Go library https://github.com/pebbe/zmq4
>>
>>
>> On Sat, Mar 12, 2016 at 6:37 PM, Jeff Shanab <jshanab at jfs-tech.com>
>> wrote:
>>
>>> I have seen this when I make a mistake and over-consume from the
>>> publisher on multipart messages. the subscriber then blocks.
>>> In my case the next message eventually comes along and things resume,
>>> but out of sync.
>>>
>>> To avoid this I created a small class that recvies the entire multipart
>>> message then it either has or has not enough "frames."
>>> I separated my interpretation of the frames from their recv.
>>>
>>> On Fri, Mar 11, 2016 at 6:39 AM, Navneet Kumar <navneet.krg at gmail.com>
>>> wrote:
>>>
>>>> Hello everone
>>>>
>>>> I am building a pubsub, wherein the PUB and the SUB socket are on the
>>>> same host.
>>>> I am able to send lots of messages through it. The latency of the
>>>> system is around 2-50 ms.
>>>>
>>>> But out of blue some messages get stuck for 5-6 seconds before
>>>> receiving.
>>>> I have put the logs and added timestamps to the message itself.
>>>> The publisher publishes and sets the time stamp, the subscriber
>>>> timestamp has a
>>>> difference of 5-10 seconds. This occurs for roughly 1 in 1000 messages.
>>>>
>>>> I am using aws c4.xlarge machine. I tried ss and there were few blips
>>>> in the recv and send queue.
>>>> If i am sending lots of messages in bulk then it sometimes get stuck
>>>> and the resumes, with no
>>>> messages being lost. Sometimes a single message itself without any
>>>> other load gets stuck.
>>>>
>>>> Has anyone faced such a scenario or give any pointers to look.
>>>>
>>>>
>>>>
>>>> --
>>>> Navneet Kumar
>>>>
>>>> _______________________________________________
>>>> zeromq-dev mailing list
>>>> zeromq-dev at lists.zeromq.org
>>>> http://lists.zeromq.org/mailman/listinfo/zeromq-dev
>>>>
>>>>
>>>
>>> _______________________________________________
>>> zeromq-dev mailing list
>>> zeromq-dev at lists.zeromq.org
>>> http://lists.zeromq.org/mailman/listinfo/zeromq-dev
>>>
>>>
>>
>>
>> --
>> Navneet Kumar
>>
>> _______________________________________________
>> zeromq-dev mailing list
>> zeromq-dev at lists.zeromq.org
>> http://lists.zeromq.org/mailman/listinfo/zeromq-dev
>>
>>
>
> _______________________________________________
> zeromq-dev mailing list
> zeromq-dev at lists.zeromq.org
> http://lists.zeromq.org/mailman/listinfo/zeromq-dev
>
>


-- 
Navneet Kumar
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.zeromq.org/pipermail/zeromq-dev/attachments/20160314/bdffdbdb/attachment.htm>


More information about the zeromq-dev mailing list