...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