Sounds great Koji, thank you for looking into that. I'm trying some tests with changes in GzipHandler included methods, will update if i have any useful info from that.
patw On Fri, Feb 15, 2019 at 3:39 AM Koji Kawamura <[email protected]> wrote: > Hi Pat, > > Thanks for sharing your insights. > I will try benchmarking before and after "gzip.setExcludedPath()" that > Mark has suggested if it helps improving S2S HTTP throughput. > > Koji > > On Fri, Feb 15, 2019 at 9:31 AM Pat White <[email protected]> > wrote: > > > > Hi Andy, > > > > My requirement is to use https with minimum tls v1.2, https being an > approved protocol. > > I haven't looked at websockets though, i need to do that, thank you for > the suggestion. > > > > patw > > > > > > > > On Thu, Feb 14, 2019 at 12:24 PM Andy LoPresto <[email protected]> > wrote: > >> > >> Pat, > >> > >> Just to clarify, your connection must be HTTPS or it just must be > secure? What about Websockets over TLS (wss://)? > >> > >> Andy LoPresto > >> [email protected] > >> [email protected] > >> PGP Fingerprint: 70EC B3E5 98A6 5A3F D3C4 BACE 3C6E F65B 2F7D EF69 > >> > >> On Feb 14, 2019, at 9:56 AM, 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 > >>>>>> > >>>>>> > >>> > >> >
