(re-adding the ML since I accidentally dropped it when I replied to your message)
On Fri, Sep 11, 2015 at 5:42 PM, Iñigo Ortiz de Urbina < [email protected]> wrote: > Hi Simon > > Thank you for trying to help :) > > On Fri, Sep 11, 2015 at 3:49 PM, Simon Pasquier <[email protected]> > wrote: > > Hi, > > You're probably hitting the issue #1627 [1] that has been fixed and will > be > > included in the official 0.10.0 release (when?). You can check by > rebuilding > > the Heka package from the versions/0.10 branch. > > Yea, I saw versions/0.10 merged 2 days ago and was wondering if some > work regarding these issues had made it to that branch, but I've not > looked at the commits themselves yet. > > For the weekend, I've disabled buffering for the TCPOutput in our > producers and also for the ESOutput in our consumers. > IMO this is the right thing to do if you can't update your Heka with the versions/0.10 code. > > If 0.10 is not anywhere near release, I'll build it myself and test it > next week. > I'll keep you all posted. > It should solve your problem. > > At any rate, I'd still like to better understand the interaction > between a buffered output with action=shutdown and full queues. > > I want the producers to be "off the hook" if anything goes wrong with > the pipeline, and that's why I instinctively set it to "shutdown". > If, however, this results in the queue not being able to be consumed > due to more logs being delivered to the message router as logs are > appended to the files monitored by Logstreamer, I'd have to set it to > "block". > > Does anyone in the list have experience with these buffered outputs in > production deployments? > What do folk at mozilla run? > > Have a great weekend! > > -- > "If you want to go fast, go alone. If you want to go far, go together." >
_______________________________________________ Heka mailing list [email protected] https://mail.mozilla.org/listinfo/heka
