Hello,
Java 1.6 prior to 1.6.0_22 (though I think it may have been fixed in a
later version) when they switched to demand loading of their probes can cause
performance problems if they are running many Java processes of one of the
earlier 1.6 versions. The question I am trying to answer is how much of a
problem this may actually be in an individual case. Initially I thought it
should be easy just count the number of times probes are registered and when
they are cleared, not nearly as easy as I had hoped.
Using the simple test of running java -version and stopping the process with
ctrl-Z or truss -Texit and can see:
$ ./jdk1.6.0_10/bin/java -version
^Z
[1]+ Stopped ./jdk1.6.0_10/bin/java -version
# /usr/sbin/dtrace -lf ':lib*:' | wc -l
515
$ ./jre1.6.0_26/bin/java -version
^Z
# /usr/sbin/dtrace -lf ':lib*:'
ID PROVIDER MODULE FUNCTION NAME
dtrace: failed to match :lib*::: No probe matches description
I think I have managed to track when the helpers are copies or removed, but
I can't think of any easy way to record when they are registered with Dtrace
and I am confused a little as to which version of java used lazy loading because
some of my testing seem to perplex me, so I am hoping for some help from the
alias.
Here is what I have so far for my Dtrace script.
$ cat dtrace_helpers.d
#!/usr/sbin/dtrace -qs
fbt:genunix:exec_args:entry,
fbt:genunix:proc_exit:entry
{
printf("Dtrace probes %s de-listed by %s for %s ( %d )\n", \
((proc_t *)(curpsinfo->pr_addr))->p_dtrace_helpers == 0 ? "NOT" : "ARE", \
probefunc , execname , pid );
}
fbt:genunix:cfork:entry
{
printf("Dtrace probes listing %s copied by %s from %s ( %d )\n", \
((proc_t *)(curpsinfo->pr_addr))->p_dtrace_helpers == 0 ? "NOT" : "ARE", \
probefunc , execname , pid );
}
/* end */
Thanks in advance for any help,
Robert
_______________________________________________
dtrace-discuss mailing list
dtrace-discuss@opensolaris.org
