[zeromq-dev] Problem with fileserver / fileclient

Tue Nov 20 02:20:49 CET 2012

We already have the ZMQ_ROUTER_MANDATORY, which returns EHOSTUNREACH.
We should get this error code if the message can't be delivered; then
the application can at least retry.

On Tue, Nov 20, 2012 at 10:12 AM, Min <miniway at gmail.com> wrote:
> And what if we add an option, blocking ROUTER ?
>
> For some application, it can be very annoying that some messages are dropped
> silently.
>
> If the option is set, the ROUTER will block instead of dropping. Sure this
> can make whole system slow, but its there's is a choose at least.
>
> Thanks
> Min
>
>
> On Mon, Nov 19, 2012 at 10:03 PM, Pieter Hintjens <ph at imatix.com> wrote:
>>
>> It's not deterministic. In fact the proper value is "infinite",
>> because once credit-based flow control works, the HWM is never reached
>> (the fact we reach it at 10 is a kind of fake). The reason for setting
>> it to 10 was to prove that the CBFC worked. We do know that there are
>> never more than 10 messages in flight. 0MQ just counts them
>> asynchronously in a way that sometimes 10 looks like 29...
>>
>> In other words, if you remove the HWM setting so that it is at the
>> default of 1,000, it will work properly and accurately.
>>
>> Hope that helps.
>>
>> -Pieter
>>
>> On Mon, Nov 19, 2012 at 8:09 PM, Stathis Gkotsis
>> <stathis.gkotsis at commsquare.com> wrote:
>> > Hi Pieter,
>> >
>> > Thanks a lot for your help. I have done some further tests and it seems
>> > that setting the HWM on the server to some higher value generally helps, but
>> > setting it to 20 is not always enough. Setting it to 30 seems to work, but I
>> > am not sure if that is deterministic.
>> >
>> > What would you say a value that always guarantees proper functioning
>> > would be, until the issue is fixed?
>> > Would this value only depend on the size of the PIPELINE, rather than on
>> > other random factors as well?
>> >
>> > Thanks,
>> > Stathis
>> >
>> > -----Original Message-----
>> > From: zeromq-dev-bounces at lists.zeromq.org
>> > [mailto:zeromq-dev-bounces at lists.zeromq.org] On Behalf Of Pieter Hintjens
>> > Sent: Monday, November 19, 2012 9:26 AM
>> > To: ZeroMQ development list
>> > Subject: Re: [zeromq-dev] Problem with fileserver / fileclient
>> >
>> > It's definitely timing dependent but the limit isn't two times, it still
>> > failed with a HWM 3x higher than the calculated amount.
>> >
>> > Sending more or fewer frames had no impact, as you say.
>> >
>> > So let's treat this as a documentation issue for now, and we can think
>> > about how to process LWM commands faster in the future.
>> >
>> > -Pieter
>> >
>> > On Mon, Nov 19, 2012 at 2:36 PM, Min <miniway at gmail.com> wrote:
>> >> At the pipe.cpp
>> >>
>> >> the variables, msgs_written and peers_msgs_read, which are used at the
>> >> HWM full calculation are updated only at the last frame.
>> >>
>> >> My guess is that the peers_msg_reads can be late updated because it is
>> >> updated asynchronously.
>> >>
>> >> For example, the dealer got chunks and sent subsequent requests, but
>> >> the router get the new requests before it get 'activate_write'.
>> >> The outgoing pipe can grow to twice of PIPELINE at maximum in an
>> >> extreme case.
>> >>
>> >>
>> >> Thanks
>> >> Min
>> >>
>> >>
>> >> On Mon, Nov 19, 2012 at 10:23 AM, Pieter Hintjens <ph at imatix.com>
>> >> wrote:
>> >>>
>> >>> I've updated the doc to make this clear, and will change the examples
>> >>> in the Guide accordingly.
>> >>>
>> >>> Thanks for spotting this problem!
>> >>>
>> >>>
>> >>> https://github.com/hintjens/libzmq/commit/7533ebb33d1cc6f23061e0228bf
>> >>> 7eb22713aaa17
>> >>>
>> >>> -Pieter
>> >>>
>> >>> On Mon, Nov 19, 2012 at 10:01 AM, Pieter Hintjens <ph at imatix.com>
>> >>> wrote:
>> >>> > Stathis,
>> >>> >
>> >>> > I've tested this in detail.
>> >>> >
>> >>> > Tthere seems to be a problem with the HWM in the ROUTER, which is
>> >>> > dropping outgoing messages unexpectedly. That is, the HWM setting
>> >>> > of
>> >>> > 10 isn't actually enough for 10 outgoing messages.
>> >>> >
>> >>> > What it looks like is that the HWM is in fact per message frame,
>> >>> > not per message. So in this case you'd have to use:
>> >>> >
>> >>> >     zsocket_set_hwm (router, PIPELINE * 2);
>> >>> >
>> >>> > In the router since the identity frame is counted in the HWM
>> >>> > calculation.
>> >>> >
>> >>> > This is a bug IMO, or at least it's rather surprising.
>> >>> >
>> >>> > -Pieter
>> >>> >
>> >>> >
>> >>> >
>> >>> >
>> >>> > On Mon, Nov 19, 2012 at 3:13 AM, Stathis Gkotsis
>> >>> > <stathis.gkotsis at commsquare.com> wrote:
>> >>> >> Indeed, sorry about this, here are the two links to the code:
>> >>> >>
>> >>> >> http://pastebin.com/1DLa4e2J
>> >>> >> http://pastebin.com/cbJUxJB4
>> >>> >>
>> >>> >> Thanks,
>> >>> >> Stathis
>> >>> >> ________________________________________
>> >>> >> From: zeromq-dev-bounces at lists.zeromq.org
>> >>> >> [zeromq-dev-bounces at lists.zeromq.org] On Behalf Of Min
>> >>> >> [miniway at gmail.com]
>> >>> >> Sent: Sunday, November 18, 2012 7:36 PM
>> >>> >> To: ZeroMQ development list
>> >>> >> Subject: Re: [zeromq-dev] Problem with fileserver / fileclient
>> >>> >>
>> >>> >> The two links seem to be identical.
>> >>> >>
>> >>> >> Could you upload the client code also ?
>> >>> >>
>> >>> >> Thanks
>> >>> >> Min
>> >>> >>
>> >>> >>
>> >>> >> On Mon, Nov 19, 2012 at 1:22 AM, Stathis Gkotsis
>> >>> >> <stathis.gkotsis at commsquare.com<mailto:stathis.gkotsis at commsquare.
>> >>> >> com>>
>> >>> >> wrote:
>> >>> >> Hello,
>> >>> >>
>> >>> >>    I am using zeromq v3.2.1-RC2 on Ubuntu 10.04 / Ubuntu 12.04
>> >>> >> 64bit. I created two simple apps: one fileserver and one
>> >>> >> fileclient, based on the "File Transfer" example code in the
>> >>> >> documentation
>> >>> >> (http://zguide.zeromq.org/page:all#header-151 , model 3). The code
>> >>> >> for these two example apps can be found here:
>> >>> >> http://pastebin.com/AuGU9EDC and here:
>> >>> >> http://pastebin.com/LusbjGCa . As you can see, I use a file of
>> >>> >> 100KB.
>> >>> >>
>> >>> >> The problem is that when I start the server and then run the
>> >>> >> client multiple times, sometimes it receives less than 102400
>> >>> >> bytes, like: 98304 or
>> >>> >> 94208 or 90112 bytes. For example, when I run the client 1000
>> >>> >> times with an interval of 100ms between the runs, for 970 runs it
>> >>> >> receives 12400 bytes, while for the rest (30 runs) it receives
>> >>> >> less bytes. I am bit puzzled by this, could you help?
>> >>> >>
>> >>> >> Thanks,
>> >>> >> Stathis
>> >>> >> _______________________________________________
>> >>> >> zeromq-dev mailing list
>> >>> >> zeromq-dev at lists.zeromq.org<mailto: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
>> >>> _______________________________________________
>> >>> 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
>> >>
>> > _______________________________________________
>> > 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
>> _______________________________________________
>> 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
>