26 Aug
2009
26 Aug
'09
9:40 p.m.
On 26.08.2009 20:18 Uhr, Daniel-Constantin Mierla wrote:
hmm, right, I overlooked the fact the perl module uses
mk_action() at
runtime since there is a dedicated C function for append_branch().
Your patch should fix the leak, but is better to update to use the
dedicated function. I will do a patch for that.
I committed on SVN the new version
using the C appen_brnach() -- get the
branch 1.5 and let me know if works ok.
Thanks,
Daniel
Thanks,
Daniel
On 26.08.2009 17:57 Uhr, James Puckett wrote:
--
Daniel-Constantin Mierla
* http://www.asipto.com/
This is the entire routing portion of the
configuration. So the
~2000 mk_actions that a dumped on exit are significantly higher than
I would have expected. Also, in previous dumps I was able to track a
good number of these frags back to where they were allocated and they
all appear to be coming from the append_branch in the perl module.
It looks like a pkg_free() is missing in modules/perl/openserxs.c" @
1763.
if (act) {
RETVAL = do_action(act, msg);
+++ pkg_free(act);
} else {
RETVAL = -1;
}
Thoughts?
===
# main request routing logic
route{
#initial requests
if(is_method("OPTIONS")) {
# send reply for each options request
sl_send_reply("200", "ok");
log("OPTIONS");
exit;
}
route(1);
}
route[1] {
if (is_method("INVITE")) {
resetflag(9);
perl_exec("routecall");
if (isflagset(9)) {
sl_send_reply("302", "Redirect");
} else {
sl_send_reply("404", "Not Found");
}
};
exit;
}
===
James Puckett - Engineers' Consulting Group
Phone 229-316-0012 - Fax 229-316-1490
On Aug 26, 2009, at 10:39 AM, Daniel-Constantin Mierla wrote:
From the log I see that at shutdown, only pkg
allocated at startup
is still allocated (e.g., mk_action is used to create the cfg tree
which is executed at runtime).
If you can point me exact part of log that intrigues you, I will
have a look. Other option, if you get out of memory (and not only)
from a specific pid, send a SIGUSR1 to that pid and you will get the
mem status -- make that part available somewhere for investigation.
Cheers,
Daniel