I'm working on integrating a fairly large legacy system to an ejb
server. I've been using Inline::Java 0.50. Lately I've been seeing a
lot of jni failures under load.
I've been investigating some mysterious malloc failures that occured
in the jvm durring hostname lookups. I found this bug report:
http://bugs.sun.com/bugdatabase/view_bug.do?bug_id=6369541
And a libc related bug filed at redhat (we're using RH enterprise 4),
http://sources.redhat.com/bugzilla/show_bug.cgi?id=156
My symptoms are pretty similar. I have a perl process with a jvm
running in JNI mode. Under load, I see random SEGV's. The hotspot
error log shows failures in the libc/name resolution libraries. So
far I've tried the latest jvm from sun. I've yet to test this with
jrocket. Has anyone else experienced this problem? Is the only path
forward really making name resolution synchronised (as discussed in
some forums linked to off the bug report on sun.com)?
On a possibly related note the snippet below is happening on the new
build box. What exactly is the test for shared_fork trying to do?
From taking a quick look at the code, its instantiating an object,
forking off a bunch of processes and modifying the integer in the
instantiated object. Its not calling wait or waitpid, just doing a
sleep(1). Is this just a simple race condition? It used to pass, but
that was on a single cpu machine (the build box has 4). I'm still
digging to see what's going on in the test classes, but I'm not sure
what the point of the test is. I really think the processes are
completing out of order.
t/08_study.............ok
t/09_usages............ok
t/10_1_shared_alone....ok
t/10_2_shared_start....ok
t/10_3_shared_use......ok
t/10_4_shared_stop.....ok
t/10_5_shared_fork.....ok 5/8# Test 6 got: "14" (t/10_5_shared_fork.t at
line 54)
t/10_5_shared_fork.....NOK 6
# t/10_5_shared_fork.t line 54 is: ok($t10::t10::i, $sum) ;
t/10_5_shared_fork.....FAILED test 6
Failed 1/8 tests, 87.50% okay
t/10_6_shared_sim......ok
t/11_exceptions........ok
t/12_1_callbacks.......ok
t/13_end...............ok
Failed Test Stat Wstat Total Fail Failed List of Failed
------------------------------------------------------------------------
--------------------------------------------------------
t/10_5_shared_fork.t 8 1 12.50% 6
Failed 1/18 test scripts, 94.44% okay. 1/314 subtests failed, 99.68%
okay.
any thoughts/feedback would be appreciated.
--
J.
