If your clocks are not synchronized you will have a hard time... Why can you not use NTP?
The only other idea I have, would be to use a custom scheduler that ensure that all spout and sink instances are running on the same machine, such that they can access the same local clock. But it is not clear if this approach is feasible for your use-case/setup. Btw: If you disable acking, you also disable at-least-once processing. In case of a failure, you will loose date. (Just in case, you are not aware of this -- if you use-case allows for incomplete results, you are of course fine) -Matthias On 09/24/2015 02:22 PM, Denis DEBARBIEUX wrote: > Hi all, > > I would like to compute the latency of my topology (time spend by a > message form the spout to the end of my topology). > > Until now, I enable the acking (1 acker per worker) and I get the result > in storm UI. But today I discover that disable the acking seep up my > topology by a factor of 10 (unit: tupes per second). > > I also try to compute this value by my self by seting a time stamp in > the tuples in the spout. Then I compute the lantency at the end of the > topology. As the clocks of my machies are not synchronized, I get very > strange results. > > So my question is how to compute the latency with a weak algorithm/tool. > > Thanks for your help. > > Regads, > > D. Debarbieux > > > ------------------------------------------------------------------------ > Avast logo <https://www.avast.com/antivirus> > > L'absence de virus dans ce courrier électronique a été vérifiée par le > logiciel antivirus Avast. > www.avast.com <https://www.avast.com/antivirus> > >
signature.asc
Description: OpenPGP digital signature
