...interesting. I dont have an answer but will initiate some research. Hopefully someone else replies if they know off-hand.
Thanks On Thu, Feb 14, 2019 at 11:43 AM Pat White <[email protected]> wrote: > Hi Folks, > > Could someone point me at the correct way to modify Nifi's embedded jetty > configuration settings? Specifically i'd like to turn off jetty's automatic > compression of payload. > > Reason for asking, think i've found my performance issue, uncompressed > input to jetty is getting automatically compressed, by jetty, causing very > small and fragmented packets to be sent, which pegs the cpu receive thread, > recombining and uncompressing the incoming packets. I'd like to verify by > turning off auto compress. > > This is what i'm seeing, app layer compressed data (nifi output port > compression=on) is accepted by jetty as-is and sent over as large, complete > tcp packets, which the receiver is able to keep up with (do not see rcv net > buffers fill up). With app layer uncompressed data (nifi output port > compression=off), jetty automatically wants to compress and sends payload > as many small fragmented packets, this causes high cpu load on the receiver > and fills up the net buffers, causing a great deal of throttling and > backoff to the sender. This is consistent in wireshark traces, good case > shows no throttling, bad case shows constant throttling with backoff. > > I've checked the User and Admin guides, as well as looking at JettyServer > and web/webdefault.xml for such controls but i'm clearly missing something, > changes have no effect on the server behavior. Appreciate any help on how > to set jetty configs properly, thank you. > > patw > > > > > On Tue, Feb 5, 2019 at 9:07 AM Pat White <[email protected]> > wrote: > >> Hi Mark, thank you very much for the feedback, and the JettyServer >> reference, will take a look at that code. >> >> I'll update the thread if i get any more info. Very strange issue, and >> hard to see what's going on in the stream due to https encryption. >> Our usecase is fairly basic, get/put flows using https over s2s, i'd >> expect folks would have hit this if it is indeed an issue, so i tend to >> suspect my install or config, however the behavior is very consistent, >> across multiple clean installs, with small files as well as larger files >> (10s of MB vs GB sized files). >> >> Thanks again. >> >> patw >> >> >> On Mon, Feb 4, 2019 at 5:18 PM Mark Payne <[email protected]> wrote: >> >>> Hey Pat, >>> >>> I saw this thread but have not yet had a chance to look into it. So >>> thanks for following up! >>> >>> The embedded server is handled in the JettyServer class [1]. I can >>> imagine that it may automatically turn on >>> GZIP. When pushing data, though, the client would be the one supplying >>> the stream of data, so the client is not >>> GZIP'ing the data. But when requesting from Jetty, it may well be that >>> Jetty is compressing the data. If that is the >>> case, I would imagine that we could easily update the Site-to-Site >>> client to add an Accept-Encoding header of None. >>> I can't say for sure, off the top of my head, though, that it will be as >>> simple of a fix as I'm hoping :) >>> >>> Thanks >>> -Mark >>> >>> [1] >>> https://github.com/apache/nifi/blob/master/nifi-nar-bundles/nifi-framework-bundle/nifi-framework/nifi-web/nifi-jetty/src/main/java/org/apache/nifi/web/server/JettyServer.java >>> >>> >>> On Feb 4, 2019, at 5:58 PM, Pat White <[email protected]> wrote: >>> >>> This looks like a thrashing behavior in compress/decompress, found that >>> if i enable compression in the output port of the receiver's RPG, the issue >>> goes away, throughput becomes just as good as for the sender's flow. Again >>> though, i believe i have compression off for all flows and components. Only >>> thing i can think of is if jetty's enforcing compression, and with an >>> uncompressed stream has an issue, but not sure why only in one direction. >>> >>> Could someone point me to where Nifi's embedded jetty configuration code >>> is, or equiv controls? >>> >>> patw >>> >>> >>> On Fri, Feb 1, 2019 at 4:13 PM Pat White <[email protected]> >>> wrote: >>> >>>> Hi Folks, >>>> >>>> I'm trying to track a very odd performance issue, this is on 1.6.0 >>>> using S2S, would like to ask if there are any known issues like this or if >>>> my flow configuration is broken. From point of view of the RPG, receiving >>>> takes ~15x longer to xsfr the same 1.5gb file as a send from that RPG. I've >>>> setup two simple flows and see this behavior consistently, also duplicated >>>> the flows between two single node instances to verify the behavior follows >>>> the xsfr direction versus the node, behavior follows the direction of xsfr, >>>> ie a receive on both nodes is much slower than sending. >>>> >>>> Flows are: >>>> >>>> FlowA: GetFile_nodeA > OutputPort_nodeA > RPG_nodeB > PutFile_nodeB >>>> FlowB: GetFile_nodeB > RPG_nodeB > InputPort_nodeA > PutFile_nodeA >>>> >>>> For the same 1.5gb file, FlowA will consistently xsfr at ~3.5MB/s, >>>> FlowB xsfrs at ~52.0MB/s, this is leaving default values for all >>>> processors, connections and the RPG with the exception that RPG uses https >>>> (instead of raw), the nodes are running secure. Same policy values were >>>> applied on both nodes to both flows. >>>> >>>> Aside from the latency diff, the xsfrs appear to work fine with no >>>> anomalies that i can find, the file transfers correctly in both directions. >>>> The one anomaly i do see is in the slow case, the destination node will >>>> have cpu go to 100% for the majority of the 6 to 7 minutes it takes to >>>> transfer the file, from a jstack on the thread that's using 99%+ of cpu, it >>>> looks like this thread is spending a lot of time in >>>> nifi.remote.util.SiteToSiteRestApiClient.read doing >>>> LazyDecompressingInputStream/InflaterInputStream, which puzzles me quite a >>>> bit because all of the ports have compression turned off, there should be >>>> no compress/decompress activity, as far as i can tell. >>>> >>>> Example stack for that thread: >>>> "Timer-Driven Process Thread-6" #90 prio=5 os_prio=0 >>>> tid=0x00007f4c48002000 nid=0xdb38 runnable [0x00007f4c734f5000] >>>> java.lang.Thread.State: RUNNABLE >>>> at java.util.zip.Inflater.inflateBytes(Native Method) >>>> at java.util.zip.Inflater.inflate(Inflater.java:259) >>>> - locked <0x00007f55d891cf50> (a java.util.zip.ZStreamRef) >>>> at >>>> java.util.zip.InflaterInputStream.read(InflaterInputStream.java:152) >>>> at java.util.zip.GZIPInputStream.read(GZIPInputStream.java:117) >>>> at >>>> java.util.zip.InflaterInputStream.read(InflaterInputStream.java:122) >>>> at >>>> org.apache.http.client.entity.LazyDecompressingInputStream.read(LazyDecompressingInputStream.java:58) >>>> at >>>> org.apache.nifi.remote.util.SiteToSiteRestApiClient$3.read(SiteToSiteRestApiClient.java:722) >>>> at java.io.InputStream.read(InputStream.java:179) >>>> at >>>> org.apache.nifi.remote.io.InterruptableInputStream.read(InterruptableInputStream.java:57) >>>> at >>>> org.apache.nifi.stream.io.ByteCountingInputStream.read(ByteCountingInputStream.java:51) >>>> at >>>> java.util.zip.CheckedInputStream.read(CheckedInputStream.java:82) >>>> at >>>> org.apache.nifi.stream.io.LimitingInputStream.read(LimitingInputStream.java:88) >>>> at java.io.FilterInputStream.read(FilterInputStream.java:133) >>>> at >>>> org.apache.nifi.stream.io.MinimumLengthInputStream.read(MinimumLengthInputStream.java:57) >>>> at >>>> org.apache.nifi.stream.io.MinimumLengthInputStream.read(MinimumLengthInputStream.java:53) >>>> at >>>> org.apache.nifi.controller.repository.io.TaskTerminationInputStream.read(TaskTerminationInputStream.java:62) >>>> at >>>> org.apache.nifi.stream.io.StreamUtils.copy(StreamUtils.java:35) >>>> at >>>> org.apache.nifi.controller.repository.FileSystemRepository.importFrom(FileSystemRepository.java:744) >>>> at >>>> org.apache.nifi.controller.repository.StandardProcessSession.importFrom(StandardProcessSession.java:2990) >>>> at >>>> org.apache.nifi.remote.StandardRemoteGroupPort.receiveFlowFiles(StandardRemoteGroupPort.java:419) >>>> at >>>> org.apache.nifi.remote.StandardRemoteGroupPort.onTrigger(StandardRemoteGroupPort.java:286) >>>> at >>>> org.apache.nifi.controller.AbstractPort.onTrigger(AbstractPort.java:250) >>>> at >>>> org.apache.nifi.controller.tasks.ConnectableTask.invoke(ConnectableTask.java:175) >>>> at >>>> org.apache.nifi.controller.scheduling.TimerDrivenSchedulingAgent$1.run(TimerDrivenSchedulingAgent.java:117) >>>> at >>>> java.util.concurrent.Executors$RunnableAdapter.call(Executors.java:511) >>>> at >>>> java.util.concurrent.FutureTask.runAndReset(FutureTask.java:308) >>>> at >>>> java.util.concurrent.ScheduledThreadPoolExecutor$ScheduledFutureTask.access$301(ScheduledThreadPoolExecutor.java:180) >>>> at >>>> java.util.concurrent.ScheduledThreadPoolExecutor$ScheduledFutureTask.run(ScheduledThreadPoolExecutor.java:294) >>>> at >>>> java.util.concurrent.ThreadPoolExecutor.runWorker(ThreadPoolExecutor.java:1142) >>>> at >>>> java.util.concurrent.ThreadPoolExecutor$Worker.run(ThreadPoolExecutor.java:617) >>>> at java.lang.Thread.run(Thread.java:748) >>>> >>>> Has anyone seen this behavior or symptoms like this? >>>> >>>> patw >>>> >>> >>>
