boB Gage wrote:
You're making me think ... :-) and that's a good thing.
I would have kept the same handler, just switched the codec. Creating
a new NioConnector is quite expensive.
Hmmm, the (someone else's) code I started from doesn't lend itself
well to that... The handler, codec factory, encoders, and decoders
are all tightly linked functionally with a single discoverable device
family (protocol). And it's working fairly well... ( at least
with 2.0.0-M4 ) except for the unbound thread growth on the discovery
failures.
If I run a test with twice our normal expected number of devices
attached and discoverable, everything runs through a few discovery
cycles and eventually each lands on the right port and everyone's fat
& happy for days. There's some number of threads stuck in WAIT
states, but it's been a relatively constant number since the last
device was discovered. [ IOW, the problem hides well ]
Ok, let put aside the architecture change... I guess you have no time
nor need to do that. Let's address the real problem !
So you have a thread stuck in a session.write() : seems like the other
side is not responding. Can you do something ? There is a small but
smart tool dumping all the methods calls into a text file. It's not
intrusive, you just have to add some parameters into your VM command line :
-javaagent:~/jip/profile/profile.jar
-Dprofile.properties=~/jip/profile/profile-mina.properties
JIP is a small profiler, you can download it at
http://jiprof.sourceforge.net/
Here is the properties I have set for mina :
profiler=on
remote=off
port=15599
ClassLoaderFilter.1=com.mentorgen.tools.profile.instrument.clfilter.StandardClassLoaderFilter
thread-depth=-1
thread.compact.threshold.ms=1
max-method-count=-1
method.compact.threshold.ms=1
file=~/profile-mina.txt
exclude=org.eclipse,org.junit,org.slf4j,org.apache.log4j,org.apache.commons,org.bouncycastle
track.object.alloc=on
output=text
output-method-signatures=yes
clock-resolution=ns
The most important line is the 'exclude' one : it removes all the
methods from a specific package.
Give it a try, and attach the resulting file (after a single run) in a
JIRA (it might be stripped out by the mailer).
--
--
cordialement, regards,
Emmanuel Lécharny
www.iktek.com
directory.apache.org
