#2890 hrSWRunEntry, don't list kernel threads (new configuration knob ?)

[Ben RUBSON]

Hi,

As explained and first reported here :
https://bugs.freebsd.org/bugzilla/show_bug.cgi?id=231732

When browsing hrSWRunEntry (1.3.6.1.2.1.25.4.2.1) on FreeBSD servers, many "kernel" processes are returned.
I have servers with around 100 user processes, and 1000 kernel threads, making hrSWRunEntry SNMP requests very slow.

KVM_GETPROCS(3) helped me to find the following solution :
I replaced KERN_PROC_ALL by KERN_PROC_PROC in kvm_getprocs() calls.

hrSWRunEntry requests now take around 0.5 second instead of sometimes more than 20 !

Could we then think about disabling kernel threads in hrSWRunEntry ?
I see no reason to keep them ?
Perhaps adding a new configuration (or compilation ?) option ?

Thank you very much !

Ben

[Bart Van Assche]

Can you share your changes, e.g. by posting here the output of git diff?

[Ben RUBSON]

Sure, here it is :

--- net-snmp-5.7.3/agent/mibgroup/host/data_access/swrun_kinfo.c.orig   2014-12-08 21:23:22.000000000 +0100
+++ net-snmp-5.7.3/agent/mibgroup/host/data_access/swrun_kinfo.c    2018-10-01 13:13:25.218415000 +0200
@@ -189,7 +189,7 @@
 #elif defined(HAVE_KVM_GETPROC2)
     proc_table = kvm_getproc2(kd, KERN_PROC_ALL, 0, sizeof(struct kinfo_proc2), &nprocs );
 #else

-    proc_table = kvm_getprocs(kd, KERN_PROC_ALL, 0, &nprocs );
+    proc_table = kvm_getprocs(kd, KERN_PROC_PROC, 0, &nprocs );
 #endif
     for ( i=0 ; i<nprocs; i++ ) {
         if ( 0 == proc_table[i].SWRUN_K_STAT )

There are some other kvm_getproc[2s](.*, KERN_PROC_ALL, .*) calls in the code, but the above one seems to be the one triggered by hrSWRunEntry SNMP requests.

[Ben RUBSON]

May I ask you some news regarding this Bart ?
Many thanks !

[Bart Van Assche]

This patch has been applied on the v5.8 and master branches. Sorry for the delay.

[Ben RUBSON]

Don't worry, perfect, thank you very much Bart !