proxmox-mirrors/mirror_frr
1
0
Fork 0
mirror of https://git.proxmox.com/git/mirror_frr synced 2025-08-06 10:54:47 +00:00
Code Issues Actions Packages Projects Releases Wiki Activity

lib: Differentiate between real and cpu bound processes

Browse Source 
When generating SLOW_THREAD warnings let's differentiate between
a cpu bound process and a wall bound process.  Effectively
a slow thread can now be a process in FRR doing lots of work( cpu bound )
or wall bound ( the cpu is heavy load and a FRR process may be pre-empted
and never scheduled ).

Signed-off-by: Donald Sharp <sharpd@nvidia.com>
This commit is contained in:
Donald Sharp 2021-02-02 12:56:06 -05:00
parent 694df37daf
commit 039d547f6f
4 changed files with 38 additions and 11 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

12
lib/lib_errors.c
View File

@ -45,9 +45,15 @@ static struct log_ref ferr_lib_warn[] = {
.suggestion = "Gather log data and open an Issue. restart FRR",
},
{
.code = EC_LIB_SLOW_THREAD,
.title = "The Event subsystem has detected a slow process",
.description = "The Event subsystem has detected a slow process, this typically indicates that FRR is having trouble completing work in a timely manner. This can be either a misconfiguration, bug, or some combination therof.",
.code = EC_LIB_SLOW_THREAD_CPU,
.title = "The Event subsystem has detected a slow cpu time process",
.description = "The Event subsystem has detected a slow process, this typically indicates that FRR is having trouble completing work in a timely manner. This can be either a misconfiguration, bug, or some combination therof. In this case total CPU time was over 5 seconds. Which indicates that FRR is very busy doing some work and should be addressed",
.suggestion = "Gather log data and open an Issue",
},
{
.code = EC_LIB_SLOW_THREAD_WALL,
.title = "The Event subsystem has detected a slow wall time process",
.description = "The Event subsystem has detected a slow process, this typically indicates that FRR is having trouble completing work in a timely manner. This can be either a misconfiguration, bug or some combination therof. In this case total WALL time was over 5 seconds. Which indicates that FRR might be having trouble being scheduled or some system call is delaying",
.suggestion = "Gather log data and open an Issue",
},
{

3
lib/lib_errors.h
View File

@ -44,7 +44,8 @@ enum lib_log_refs {
EC_LIB_SNMP,
EC_LIB_STREAM,
EC_LIB_LINUX_NS,
EC_LIB_SLOW_THREAD,
EC_LIB_SLOW_THREAD_CPU,
EC_LIB_SLOW_THREAD_WALL,
EC_LIB_NO_THREAD,
EC_LIB_RMAP_RECURSION_LIMIT,
EC_LIB_BACKUP_CONFIG,

22
lib/thread.c
View File

@ -1850,15 +1850,29 @@ void thread_call(struct thread *thread)
memory_order_seq_cst);
#ifdef CONSUMED_TIME_CHECK
if (realtime > CONSUMED_TIME_CHECK) {
if (cputime > CONSUMED_TIME_CHECK) {
/*
* We have a CPU Hog on our hands.
* We have a CPU Hog on our hands. The time FRR
* has spent doing actual work ( not sleeping )
* is greater than 5 seconds.
* Whinge about it now, so we're aware this is yet another task
* to fix.
*/
flog_warn(
EC_LIB_SLOW_THREAD,
"SLOW THREAD: task %s (%lx) ran for %lums (cpu time %lums)",
EC_LIB_SLOW_THREAD_CPU,
"CPU HOG: task %s (%lx) ran for %lums (cpu time %lums)",
thread->xref->funcname, (unsigned long)thread->func,
realtime / 1000, cputime / 1000);
} else if (realtime > CONSUMED_TIME_CHECK) {
/*
* The runtime for a task is greater than 5 seconds, but
* the cpu time is under 5 seconds. Let's whine
* about this because this could imply some sort of
* scheduling issue.
*/
flog_warn(
EC_LIB_SLOW_THREAD_WALL,
"STARVATION: task %s (%lx) ran for %lums (cpu time %lums)",
thread->xref->funcname, (unsigned long)thread->func,
realtime / 1000, cputime / 1000);
}

12
lib/vty.c
View File

@ -512,13 +512,19 @@ static int vty_command(struct vty *vty, char *buf)
#ifdef CONSUMED_TIME_CHECK
GETRUSAGE(&after);
if ((realtime = thread_consumed_time(&after, &before, &cputime))
> CONSUMED_TIME_CHECK)
realtime = thread_consumed_time(&after, &before, &cputime);
if (cputime > CONSUMED_TIME_CHECK) {
/* Warn about CPU hog that must be fixed. */
flog_warn(
EC_LIB_SLOW_THREAD,
EC_LIB_SLOW_THREAD_CPU,
"SLOW COMMAND: command took %lums (cpu time %lums): %s",
realtime / 1000, cputime / 1000, buf);
} else if (realtime > CONSUMED_TIME_CHECK) {
flog_warn(
EC_LIB_SLOW_THREAD_WALL,
"SLOW COMMAND: command took %lums (cpu time %lums): %s",
realtime / 1000, cputime / 1000, buf);
}
}
#endif /* CONSUMED_TIME_CHECK */