Panos Thanks - we should change the default that is used at some point. The old one has been stable/solid for a long time though so switching it will need to be a deliberate choice as well. Was thinking it made sense to do at the next major release but perhaps at one of the upcoming minors it would be good too. The new repo honors the old data structure so migration should be trivial.
Thanks On Wed, Nov 29, 2017 at 1:04 AM, Panos Geo <[email protected]> wrote: > Hello Joe, > > > A quick message to say that we switched to 'WriteAheadProvenanceRepository' > and we don't see NiFi getting stuck anymore. Perhaps this would help other > people too 😉 > > > Many thanks, > > Panos > > > ------------------------------ > *From:* Joe Witt <[email protected]> > *Sent:* Saturday, November 25, 2017 3:05 AM > *To:* [email protected] > *Subject:* Re: NiFi stops processing further flowfiles > > Panos > > In looking at the previously attached stack dump it looks as though the > provenance repository might be in a live lock state with one of its > threads. The getSize call it is sitting on has been seen before and fixed > for some cases but perhaps not all. There is a whole new provenance > repository you could switch to though which offers far faster and possibly > more stable behavior called the 'WriteAheadProvenanceRepository'. You > can switch to it and it will honor the old data I believe. To switch you > just edit your nifi.properties file and where it said ' > PersistentProvenanceRepository' you change that classname to ' > WriteAheadProvenanceRepository'. > > Thanks > Joe > > On Fri, Nov 24, 2017 at 1:59 PM, Juan Sequeiros <[email protected]> > wrote: > > Hi Panos, > > The scheduling strategy defaults to timer driven on processors and I've > always avoided using event driven so if you are not purposely using it I > would just leave that setting alone. It even warns you that it is > experimental. > > Your thread count is shown on the UI or can be called through RestApi > /nifi-api/flow/status > > Thanks, > > Juan > > On Fri, Nov 24, 2017 at 1:24 PM Panos Geo <[email protected]> > wrote: > > Thanks for your reply Juan. > > > I have been using NiFi for almost 2 years now and didn't know about this > setting. I 'll try to increase it and see what happens 😊 > > > Does it also make sense to increase the "Maximum Event Driven Thread > Count"? > > > Finally, is there a way to monitor thread usage within NiFi? > > > Many thanks, > > Panos > > > > ------------------------------ > *From:* Juan Sequeiros <[email protected]> > *Sent:* Friday, November 24, 2017 4:50 PM > *To:* [email protected] > *Subject:* Re: NiFi stops processing further flowfiles > > Good afternoon, > > My quick guess is you are running out of threads. > If your system can handle it increase the default "max thread" count found: > > Left side controller settings > Maximum timer driven thread count > > On Fri, Nov 24, 2017 at 10:47 AM Panos Geo <[email protected]> > wrote: > > Hello all, > > We are running NiFi in a docker container that has roughly 500 processors > running. These include consuming messages (MQTT), interaction with a > database and some data processing with execute scripts. Our NiFi version is > v.1.4.0 and this instance has 18GB of RAM with 12GB allocated to Java heap. > > Quite recently we increased the load of the incoming messages > (1000messages/minute) that lead to more files being queued and some > backpressure. We then have experienced what appears to be some kind of > deadlock. NiFi stops processing any flowfiles and the processors running > get stuck (there is a thread running constantly; we see 1 at the top right > corner of the processor). If we try to stop these processors the icon gets > changed to stop, but we now have 2 threads stuck at the top right corner of > the processor. We cannot start these processors again and no other data > flow is being processed unless we restart NiFi. We have experienced this > multiple times over the past week. > > I am attaching a nifi dump in case it helps. Our nifi.properties is fairly > standard (we just have smaller retention periods for content and > provenance). > > Any thoughts? What can we do to debug this further? > > Many thanks, > Panos > > >
