OS: Solaris 5.8, 5.9
File: "tarball/gmond/machines/solaris.c"
Function: get_kstat_val( )
I've been testing the latest ganglia 2.6.0 posted on March 22nd.
I noticed that gmond segfaults & core dumps when the first online cpu is NOT in
slot #0.
When I do a "kstat cpu_stat", my cpus are instance #'s 1 & 3.
The crash occurs during the ks = kstat_lookup( ). It's hardcoded to look for
cpu instance #0.
Anyways, I wrote the following to check the next cpu instance if the default
(zero) returns NULL.
So since most machines will likely have a cpu at slot #0, I left the initial
check as is.
However, if it fails, then start checking instance #1, #2,... etc until one is
found.
It's not pretty, but it appears to work.
Here's a diff of my changes for solaris.c:
4c4
< #include "debug_msg.h"
---
> #include "lib/debug_msg.h"
229a230,250
>
> /* CPU_INFO module & instance check
> * If the first online CPU is not in slot #0, gmond will segfault and core
> dump.
> * If ks == NULL, then traverse the cpu instance #'s until the first
> online cpu is found.
> * After the traversal, if ks == NULL, then something is very wrong with
> the cpu configuration.
> */
> if ((km_name == "cpu_info") && (ks == NULL)) {
> int instance_num = 1;
> int num_tries = 0;
> while ((num_tries < (int)metriclist.cpu_num.uint32) && (ks == NULL))
> {
> char *new_ks_name = (char *)malloc(20*sizeof(char));
> sprintf(new_ks_name, "cpu_info%d", instance_num);
> debug_msg( "Lookup up kstat: km (unix?)='%s', ks
> (system_misc?)='%s',kn (resulting metric?)='%s'", km_name, new_ks_name, name);
> ks = kstat_lookup(kc, km_name, instance_num, new_ks_name);
> debug_msg("%s: Just did kstat_lookup().\n",new_ks_name);
> ++instance_num;
> ++num_tries;
> free(new_ks_name);
> }
> }
>
--
This e-mail may contain confidential and/or privileged information. If you are
not the intended recipient (or have received this e-mail in error) please
notify the sender immediately and destroy this e-mail. Any unauthorized
copying, disclosure or distribution of the material in this e-mail is strictly
forbidden.
