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 >>>>> >>>>> >>
