Matt, Our main use, which provenance data handles well, is figuring out **what** data was handled. We drop everything but DROP out of convenience because we have no known scenarios where data will be removed before it reaches the end of the flow.
FWIW, this is what inspired the record stats processor we added about a release ago. You can use it to drill down into a record set with path operations and add attributes to a flowfile describing what the record set therein contains. It should be particularly helpful to anyone who wants to do partial or full table fetches and then say "count this batch of records using this path." Mike On Wed, Jul 25, 2018 at 9:59 PM Matt Burgess <mattyb...@apache.org> wrote: > Mike, Ryan, Boris et al, > > I'd like to wrap my head around the kinds of use cases y'all have for > provenance data in NiFi: what's good, what's bad, what we need to do > to make things better. Are there questions you want to ask of > provenance that you can't today? Do the DROP events give you what you > need for your reporting, or would you benefit from some sort of > "lifetime record" that might be generated from a NiFi subflow based on > provenance events? I've been bouncing around the following > concepts/improvements: > > 1) Let the ProcessSession keep track of the duration in a processor > (for those that don't explicitly report it) [1] > 2) Add a "durationNanos" field to provenance events, possibly > replacing "durationMillis" in NiFi 2.0, to give better precision [2] > 3) A processor to generate lineage when a DROP event is received > (likely via the SiteToSiteProvenanceReportingTask), dependent on the > persistence settings of the provenance repository > 4) A "Query Filter" property on the SiteToSiteProvenanceReportingTask > (or a separate reporting task if need be), perhaps leveraging Calcite > for SQL filters or supporting the Lucene query language (since the > prov events are indexed by Lucene) > > I still haven't come up with the New Feature Wiki page for graph tech > (from a previous discussion on the list) but #3 above lends itself to > also generating a lineage graph for a FlowFile, in some well-known > format perhaps (Kryo, GraphML, etc.) I'll try to get that Wiki (and > the discussion) going soon... > > Regards, > Matt > > [1] https://issues.apache.org/jira/browse/NIFI-5420 > [2] https://issues.apache.org/jira/browse/NIFI-5463 > On Wed, Jul 25, 2018 at 9:18 PM Mike Thomsen <mikerthom...@gmail.com> > wrote: > > > > Ryan, > > > > Understandable. We haven't found a need for Beats or Forwarders here > either because S2S gives everything you need to reliably ship the data. > > > > FWIW, if your need changes, I would recommend stripping down the > provenance data. We cut out about 66-75% of the fields and dropped the > intermediate records in favor of keeping DROP events for our simple > dashboarding needs because we figured if a record never made it there > something very bad happened. > > > > On Wed, Jul 25, 2018 at 8:54 PM Ryan H < > ryan.howell.developm...@gmail.com> wrote: > >> > >> Thanks Mike for the suggestion on it. I'm looking for a solution that > doesn't involve the additional components such as any > Beats/Forwarders/Elasticsearch/etc. > >> > >> Boris, thanks for the link for the Monitoring introduction--I've > checked it out multiple times. What I want to avoid is having the need for > anything to be set on the Canvas and have the metrics collection via the > rest api. I'm thinking that the api in the original question may be the way > to go, but unsure of it without a little more information on the data model > and how that data is collected/aggregated (such as what the data returned > actually represents). I may just dig into the source if this email goes > stale. > >> > >> -Ryan > >> > >> > >> On Wed, Jul 25, 2018 at 9:17 AM, Boris Tyukin <bo...@boristyukin.com> > wrote: > >>> > >>> Ryan, if you have not seen these posts from Pierre, I suggest starting > there. He does a good job explaining different options > >>> https://pierrevillard.com/2017/05/11/monitoring-nifi-introduction/ > >>> > >>> I do agree that 5 minute thing is super confusing and pretty useless > and you cannot change that interval. I think it is only useful to check > quickly on your real-time pipelines at the moment. > >>> > >>> I wish NiFi provided nicer out of the box logging/monitoring > capabilities but on a bright side, it seems to me that you can build your > own and customize it as you want. > >>> > >>> > >>> On Tue, Jul 24, 2018 at 10:55 PM Ryan H < > ryan.howell.developm...@gmail.com> wrote: > >>>> > >>>> Hi All, > >>>> > >>>> I am looking for a way to obtain the total amount of data that has > been processed by a running cluster for a period of time, ideally via the > rest api. > >>>> > >>>> Example of my use case: > >>>> I have say 50 different process groups, each that have a connection > to some data source. Each one is continuously pulling data in, doing > something to it, then sending it out to some other external place. I'd like > to programmatically gather some metrics about the amount of data flowing > thru the cluster as a whole (everything that is running across the cluster). > >>>> > >>>> It looks like the following api may be the solution, but I am curious > about some of the properties: > >>>> "nifi-api/flow/process-groups/root/status?recursive=true". > >>>> > >>>> Looking at the data model (as defined in the rest api documentation) > and the actual data that is returned, my questions are: > >>>> 1. Would this be the correct way to obtain this information? > >>>> 2. And if so, I'm not sure which properties to look at as it isn't > immediately clear to me the difference between some of them. Example being > "bytesSent" vs "bytesOut". > >>>> 3. How is this data updated? It looks like a lot of these metrics are > supposed to updated every 5 minutes. So would it be that the info I would > get now is what was collected from the last 5 minute interval and would > stay the same until the next 5 minute interval? And does the data aggregate > or is it only representative of a single 5 minute period? Something else? > >>>> > >>>> > >>>> > >>>> { > >>>> "processGroupStatus": { > >>>> ... > >>>> "aggregateSnapshot": { > >>>> ... > >>>> "flowFilesIn": 0, > >>>> "bytesIn": 0, > >>>> "input": "value", > >>>> "flowFilesQueued": 0, > >>>> "bytesQueued": 0, > >>>> "queued": "value", > >>>> "queuedCount": "value", > >>>> "queuedSize": "value", > >>>> "bytesRead": 0, > >>>> "read": "value", > >>>> "bytesWritten": 0, > >>>> "written": "value", > >>>> "flowFilesOut": 0, > >>>> "bytesOut": 0, > >>>> "output": "value", > >>>> "flowFilesTransferred": 0, > >>>> "bytesTransferred": 0, > >>>> "transferred": "value", > >>>> "bytesReceived": 0, // I think this is the one, but not sure > >>>> "flowFilesReceived": 0, > >>>> "received": "value", > >>>> "bytesSent": 0, // I think this is the other one, but not sure > >>>> "flowFilesSent": 0, > >>>> "sent": "value", > >>>> "activeThreadCount": 0, > >>>> "terminatedThreadCount": 0 > >>>> }, > >>>> "nodeSnapshots": [{…}] > >>>> }, > >>>> "canRead": true > >>>> } > >>>> > >>>> > >>>> > >>>> > >>>> Any help or insight is always appreciated! > >>>> > >>>> > >>>> Cheers, > >>>> > >>>> Ryan H. > >>>> > >>>> > >> >
