Thank you Bryan, i have a lot to look at using raw protocol, will be sure
to keep this setting in mind.

patw

On Thu, Feb 14, 2019 at 12:05 PM Bryan Bende <[email protected]> wrote:

> You can use TLS with raw s2s by setting nifi.remote.input.secure=true
>
> On Thu, Feb 14, 2019 at 12:56 PM Pat White <[email protected]>
> wrote:
> >
> > Thanks very much folks, definitely appreciate the feedback.
> >
> > Right, required to use tls/https connections for s2s, so raw is not an
> option for me.
> >
> > Will look further at JettyServer and setIncludedMethods, thanks again.
> >
> > patw
> >
> > On Thu, Feb 14, 2019 at 11:07 AM Mark Payne <[email protected]>
> wrote:
> >>
> >> Pat,
> >>
> >> It appears to be hard-coded, in JettyServer (full path is
> >>
> nifi/nifi-nar-bundles/nifi-framework-bundle/nifi-framework/nifi-web/nifi-jetty/src/main/java/org/apache/nifi/web/server/JettyServer.java
> )
> >>
> >> Line 294 calls the gzip method, which looks like:
> >>
> >> private Handler gzip(final Handler handler) {
> >>     final GzipHandler gzip = new GzipHandler();
> >>     gzip.setIncludedMethods("GET", "POST", "PUT", "DELETE");
> >>     gzip.setHandler(handler);
> >>     return gzip;
> >> }
> >>
> >>
> >> We probably would want to add a "gzip.setExcludedPath()" call to
> exclude anything that goes to the site-to-site path.
> >>
> >> Thanks
> >> -Mark
> >>
> >>
> >> On Feb 14, 2019, at 11:46 AM, Joe Witt <[email protected]> wrote:
> >>
> >> ...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
> >>>>>
> >>>>>
> >>
>

Reply via email to