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] 
> <mailto:[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] 
>> <mailto:[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] 
>> <mailto:[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] 
>> <mailto:[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] 
>> <mailto:[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
>>  
>> <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] 
>>> <mailto:[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] 
>>> <mailto:[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