Hi.

On 12.01.2018 11:06, Toom Andreas wrote:
Hi André,

Thanks for your reply!

The Tomcat Connector connection_pool_timeout  is specified in seconds and the 
Tomcat ajp connector connectionTimeout is specified in milliseconds so both 
sides have the same value. This is the recommended setting. See quote from the 
link that your provided: 
https://tomcat.apache.org/connectors-doc/common_howto/timeouts.html

"We generally recommend values around 10 minutes, so setting connection_pool_timeout 
to 600 (seconds). If you use this attribute, please also set the attribute 
keepAliveTimeout (if it is set explicitly) or connectionTimeout in the AJP Connector 
element of your Tomcat server.xml configuration file to an analogous value. Caution: 
keepAliveTimeout and connectionTimeout must be given in milliseconds. So if you set JK 
connection_pool_timeout to 600, you should set Tomcat keepAliveTimeout or 
connectionTimeout to 600000."


Yes, I understand this, and I did also read this in the documentation.
But
1) it is always possible that a documentation would not be perfect, or maybe somewhat outdated (considering for example various kinds of Connector implementations)
and
2) I was just wondering what happens if the two sides (mod_jk on one side, and the Connector on the other side), because of a long period of inactivity, decided per chance to close each their own side of the connection, at (almost) exactly the same time. Maybe there is some edge case there, where one side "misses a signal" from the other side, and starts some cpu-intensive wait for a connection state which will never happen on the socket side it is waiting on. I admit that this is a wild guess; but maybe just changing the timeout of one of them, by a couple of seconds, may show something interesting (at little cost in terms of re-configuration or testing).





Best regards,
Andreas

-----Original Message-----
From: André Warnier (tomcat) [mailto:a...@ice-sa.com]
Sent: den 12 januari 2018 09:57
To: users@tomcat.apache.org
Subject: Re: Ajp Nio-thread stuck in loop and consuming a lot of cpu

On 12.01.2018 07:58, Toom Andreas wrote:
Hi,

We are having an issue with an application running Apache Tomcat 8.0.47 using 
Oracle Jvm 1.8.0.151 on Linux (RHEL 7). The Tomcat process is consuming  cpu at 
a constant high level even though there is a low amount of incoming traffic. 
When inspecting the process health using JMX /JVisualVM CPU Sampler we see that 
there are 4 ajp-nio-exec threads together with a 
NioBlockingSelector.BlockPoller thread that consume most of the cpu.

A stack trace of one of the ajp-io-exec threads looks like this:

"ajp-nio-48317-exec-14233" - Thread t@201195
     java.lang.Thread.State: TIMED_WAITING
                               at sun.misc.Unsafe.park(Native Method)
                               - parking to wait for <342fab60> (a 
java.util.concurrent.CountDownLatch$Sync)
                               at 
java.util.concurrent.locks.LockSupport.parkNanos(LockSupport.java:215)
                               at 
java.util.concurrent.locks.AbstractQueuedSynchronizer.doAcquireSharedNanos(AbstractQueuedSynchronizer.java:1037)
                               at 
java.util.concurrent.locks.AbstractQueuedSynchronizer.tryAcquireSharedNanos(AbstractQueuedSynchronizer.java:1328)
                               at 
java.util.concurrent.CountDownLatch.await(CountDownLatch.java:277)
                               at 
org.apache.tomcat.util.net.NioEndpoint$KeyAttachment.awaitLatch(NioEndpoint.java:1400)
                               at 
org.apache.tomcat.util.net.NioEndpoint$KeyAttachment.awaitReadLatch(NioEndpoint.java:1402)
                               at 
org.apache.tomcat.util.net.NioBlockingSelector.read(NioBlockingSelector.java:185)
                               at 
org.apache.tomcat.util.net.NioSelectorPool.read(NioSelectorPool.java:250)
                               at 
org.apache.tomcat.util.net.NioSelectorPool.read(NioSelectorPool.java:231)
                               at 
org.apache.coyote.ajp.AjpNioProcessor.readSocket(AjpNioProcessor.java:194)
                               at 
org.apache.coyote.ajp.AjpNioProcessor.read(AjpNioProcessor.java:160)
                               at 
org.apache.coyote.ajp.AbstractAjpProcessor.readMessage(AbstractAjpProcessor.java:1091)
                               at 
org.apache.coyote.ajp.AbstractAjpProcessor.process(AbstractAjpProcessor.java:804)
                               at 
org.apache.coyote.AbstractProtocol$AbstractConnectionHandler.process(AbstractProtocol.java:684)
                               at 
org.apache.tomcat.util.net.NioEndpoint$SocketProcessor.doRun(NioEndpoint.java:1539)
                               at 
org.apache.tomcat.util.net.NioEndpoint$SocketProcessor.run(NioEndpoint.java:1495)
                               - locked <279cc7f7> (a 
org.apache.tomcat.util.net.NioChannel)
                               at 
java.util.concurrent.ThreadPoolExecutor.runWorker(ThreadPoolExecutor.java:1149)
                               at 
java.util.concurrent.ThreadPoolExecutor$Worker.run(ThreadPoolExecutor.java:624)
                               at 
org.apache.tomcat.util.threads.TaskThread$WrappingRunnable.run(TaskThread.java:61)
                               at java.lang.Thread.run(Thread.java:748)

     Locked ownable synchronizers:
                               - locked <10e1f474> (a
java.util.concurrent.ThreadPoolExecutor$Worker)

A stack trace of the NioBlockingSelector.BlockPoller thread looks like this:

"NioBlockingSelector.BlockPoller-2" - Thread t@17
     java.lang.Thread.State: RUNNABLE
                               at sun.nio.ch.EPollArrayWrapper.epollWait(Native 
Method)
                               at 
sun.nio.ch.EPollArrayWrapper.poll(EPollArrayWrapper.java:269)
                               at 
sun.nio.ch.EPollSelectorImpl.doSelect(EPollSelectorImpl.java:93)
                               at 
sun.nio.ch.SelectorImpl.lockAndDoSelect(SelectorImpl.java:86)
                               - locked <648895e4> (a sun.nio.ch.Util$3)
                               - locked <17921532> (a 
java.util.Collections$UnmodifiableSet)
                               - locked <67762111> (a 
sun.nio.ch.EPollSelectorImpl)
                               at 
sun.nio.ch.SelectorImpl.selectNow(SelectorImpl.java:105)
                               at
org.apache.tomcat.util.net.NioBlockingSelector$BlockPoller.run(NioBloc
kingSelector.java:339)

     Locked ownable synchronizers:
                               - None

It seems like the worker thread is trying to read from the underlying socket. When we 
inspect the OS tcp/ip statistics using the ss-command (ss -i src :<ajp-port>)  
the following connections are suspicious:

State      Recv-Q Send-Q Local Address:Port                 Peer Address:Port
CLOSE-WAIT 2      0      10.x.x.x:48317                10.x.x.11:53724
                                cubic wscale:7,7 rto:202 rtt:1.724/2.424 ato:40 
mss:1448 cwnd:10 ssthresh:31 bytes_acked:343869507 bytes_received:22259372 
segs_out:302687 segs_in:141943 send 67.2Mbps lastsnd:1290791171 
lastrcv:1290786778 lastack:1290786767 pacing_rate 134.4Mbps retrans:0/433 
rcv_rtt:1.875 rcv_space:28960
CLOSE-WAIT 1      0      10.x.x.x:48317                10.x.x.12:41990
                                cubic wscale:7,7 rto:202 rtt:1.132/0.899 ato:40 
mss:1448 cwnd:10 ssthresh:22 bytes_acked:37347221 bytes_received:3346200 
segs_out:35273 segs_in:20608 send 102.3Mbps lastsnd:289840681 lastrcv:289730802 
lastack:289720796 pacing_rate 204.5Mbps retrans:0/13 reordering:24 
rcv_rtt:1.875 rcv_space:28960
CLOSE-WAIT 1      0      10.x.x.x:48317                10.x.x.12:60068
                                cubic wscale:7,7 rto:209 rtt:8.757/14.317 
ato:40 mss:1448 cwnd:10 ssthresh:36 bytes_acked:180903401 
bytes_received:16401965 segs_out:170017 segs_in:100392 send 13.2Mbps 
lastsnd:395602314 lastrcv:395602267 lastack:395602257 pacing_rate 26.5Mbps 
retrans:0/106 rcv_rtt:1.875 rcv_space:28960
CLOSE-WAIT 2      0      10.x.x.x:48317                10.x.x.11:50628
                                cubic wscale:7,7 rto:203
rtt:2.992/4.252 ato:40 mss:1448 cwnd:13 ssthresh:25
bytes_acked:4185335 bytes_received:471220 segs_out:4193 segs_in:2181
send 50.3Mbps lastsnd:1036916135 lastrcv:1036871377 lastack:1036871367
pacing_rate 100.6Mbps retrans:0/1 reordering:26 rcv_rtt:6.375
rcv_space:28960

We are running our Tomcat node behind and Apache server and use Tomcat 
Connector (mod_jk) as a load balancer. When looking at the hosts running the 
Apache node the corresponding sockets are closed. To us it seems that Tomcat 
has not properly closed the underlying socket when Apache/mod_jk initiated a 
FIN and is instead continuously trying to read from the socket even though it 
has been closed on the Apache side ?

I know that we are one version behind the latest release of Tomcat 8.0.x but I 
checked the changelog and I could not see any fixed issues matching this 
problem. Unfortunately we are only seeing this in production and we have not 
been able to reproduce in test. Has anyone experienced a similar issue like 
this before ?

For reference our connector configuration in server.xml looks like this:

<Connector port="48317"
                 protocol="AJP/1.3"
                 maxThreads="200"
                 connectionTimeout="600000"
                 xpoweredBy="false"
                 allowTrace="false"
                 URIEncoding="UTF-8"
                 requiredSecret="xxx"
                 secure="true"
                 scheme="https"
                 proxyPort="443" />

And our Tomcat Connector configuration for Apache in workers.properties looks 
like this:

worker.list = status,app

# Create workers for app
worker.app-1.type=ajp13
worker.app-1.host=host8128.corpdomain.internal
worker.app-1.port=48317
worker.app-1.ping_mode=A
worker.app-1.connection_pool_timeout=600
worker.app-1.secret=xxx
worker.app-2.type=ajp13
worker.app-2.host=host8129.corpdomain.internal
worker.app-2.port=48317
worker.app-2.ping_mode=A
worker.app-2.connection_pool_timeout=600
worker.app-2.secret=xxx

# Create a loadbalancer for app
worker.app.type=lb
worker.app.sticky_session=1
worker.app.balance_workers=app-1, app-2

Best regards,
Andreas


Andreas Toom
Systemutvecklare/IT
TEL: +46 702 03 04 98
[Volvofinans Bank AB]<http://www.volvofinans.se/>


Hi.
First, thanks for the very complete issue description.

This page may be of help :
https://tomcat.apache.org/connectors-doc/common_howto/timeouts.html
(which considering the above, you probably have read already).

I did not really analyse in detail your description, nor that page.
But I am just wondering if in this case, setting both the
worker.app-x.connection_pool_timeout=600
and the
<Connector .. connectionTimeout="600000" ..> (to exactly the same value in 
seconds) is not in the end counter-productive.

This is just a hunch, because determining the exact interaction and 
side-effects of these multiple timeout parameters is somewhat of a black art, 
but it is maybe worth changing one of them (for example back to the default 
value), and check what happens then.
(You say that traffic is low, so you probably do hit one or more of these 
timeouts at some
point)

That is not to say that there may not be some bug somewhere in the code. But 
doing the experiment above and reporting the result may provide some more clues 
for someone to analyse your issue in detail.

You may also want to try swapping the AJP Connector, using the following 
attribute :
http://tomcat.apache.org/tomcat-8.5-doc/config/ajp.html#Common_Attributes
--> protocol

Of course, doing such changes one at a time facilitates the analysis.


---------------------------------------------------------------------
To unsubscribe, e-mail: users-unsubscr...@tomcat.apache.org
For additional commands, e-mail: users-h...@tomcat.apache.org


---------------------------------------------------------------------
To unsubscribe, e-mail: users-unsubscr...@tomcat.apache.org
For additional commands, e-mail: users-h...@tomcat.apache.org



---------------------------------------------------------------------
To unsubscribe, e-mail: users-unsubscr...@tomcat.apache.org
For additional commands, e-mail: users-h...@tomcat.apache.org

Reply via email to