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