[+storm user list again, please remember to always CC the user list on your replies to threads on the list so that everyone can see the discussion]
Sounds like you've hit some scaling bottleneck with Storm. I've never tried running nearly that number of topologies. It might be an inefficiency with the number of API calls, or with the interactions between the Nimbus and ZooKeeper. If I were you I'd use a browser-based debugging system that will allow you to get an idea of which calls are being slow. e.g., Chrome Developer Tools. Thanks, - Erik On Tue, Jun 20, 2017 at 9:52 PM, daxwang(王文军) <[email protected]> wrote: > each topology has its own streamming data and its own compute logic for > diffrerent users. but the streamming data is not big. we run a topology > using one slot, so there many small topologies > in a cluster. > > ------------------------------ > > > *From:* Erik Weathers <[email protected]> > *Date:* 2017-06-21 12:37 > *To:* daxwang(王文军) <[email protected]>; [email protected] > *Subject:* Re: Re: open storm ui very slow(Internet mail) > [+user list back into the email thread] > > Interesting. Can you please answer my earlier questions, if you're > allowed to? i.e., why do you have so many topologies with only 1 worker > per topology? And how many tuples are you processing in total? > > - Erik > > On Tue, Jun 20, 2017 at 9:33 PM daxwang(王文军) <[email protected]> wrote: > >> >> it is a product cluster. not playing around with it >> ------------------------------ >> >> >> *From:* Erik Weathers <[email protected]> >> *Date:* 2017-06-21 12:27 >> *To:* daxwang(王文军) <[email protected]> >> *Subject:* Re: Re: open storm ui very slow(Internet mail) >> >> Yes, all of that info was available in the previous screenshot. >> >> So it sounds like you are just playing around with storm right now, not >> using it for a production system? >> >> - Erik >> >> On Tue, Jun 20, 2017 at 9:23 PM, daxwang(王文军) <[email protected]> >> wrote: >> >>> in my cluster, 18 supervisor, 50 slots per supervisor, so the total >>> slot is 50*18=900 and one topology run using one slot, so 799 topologies >>> used 799 slots. >>> but open storm ui is really very slow. >>> ------------------------------ >>> >>> >>> *From:* Erik Weathers <[email protected]> >>> *Date:* 2017-06-21 11:07 >>> *To:* [email protected] >>> *Subject:* Re: open storm ui very slow(Internet mail) >>> Ah, the screen capture had the number of executors: 1753. >>> We are still on storm-0.9.6, but we have over 10000 executors and the >>> base UI page loads quickly. Only have ~30 topologies right now though. >>> >>> - Erik >>> >>> On Tue, Jun 20, 2017 at 8:03 PM, Erik Weathers <[email protected]> >>> wrote: >>> >>>> > when i open the storm ui, i found it's very slow, about 60 seconds. >>>> why? is cluster too large? >>>> >>>> Seems likely that's the reason. >>>> >>>> I'm curious about your setup. Are you just playing around with Storm, >>>> or is cluster actively used for "real work"? >>>> If the latter, then I'd be curious to know what your use-case is for >>>> having 799 topologies with 1 worker per topology? >>>> How many executors are there in your cluster? Also, how many total >>>> tuples are you processing in this cluster? >>>> >>>> Thanks! >>>> >>>> - Erik >>>> >>>> >>>> On Tue, Jun 20, 2017 at 7:32 PM, daxwang(王文军) <[email protected]> >>>> wrote: >>>> >>>>> in storm cluster >>>>> version: 1.0.3 >>>>> 2 nimbus >>>>> 18 supervisor >>>>> running 799 topologies >>>>> totoal 900 slots >>>>> when i open the storm ui, i found it's very slow, about 60 seconds. >>>>> why? is cluster too large? >>>>> >>>>> >>>>> >>>>> ------------------------------ >>>>> >>>>> >>>> >>> >>
