Attachment #24712 for ticket #13950

View | Details | Raw Unified | Return to ticket 13950
Collapse All | Expand All

Lines 126-132 typedef struct { Link Here

(-)slurm-21.08.7.orig/src/common/cgroup.h (-1 / +2 lines)
126	typedef struct {	126	typedef struct {
127	uint64_t usec;	127	uint64_t usec;
128	uint64_t ssec;	128	uint64_t ssec;
129	uint64_t total_rss;	129	uint64_t total_inactive_anon;
		130	uint64_t total_active_anon;
130	uint64_t total_pgmajfault;	131	uint64_t total_pgmajfault;
131	} cgroup_acct_t;	132	} cgroup_acct_t;
132		133




	stats = xmalloc(sizeof(*stats));
	stats->usec = NO_VAL64;
	stats->ssec = NO_VAL64;
	stats->total_inactive_anon = NO_VAL64;
	stats->total_active_anon = NO_VAL64;
	stats->total_pgmajfault = NO_VAL64;

	if (cpu_time != NULL)
		sscanf(cpu_time, "%*s %lu %*s %lu", &stats->usec, &stats->ssec);

	if ((ptr = xstrstr(memory_stat, "total_inactive_anon")))
		sscanf(ptr, "total_inactive_anon %lu", &stats->total_inactive_anon);
	if ((ptr = xstrstr(memory_stat, "total_active_anon")))
		sscanf(ptr, "total_active_anon %lu", &stats->total_active_anon);
	if ((ptr = xstrstr(memory_stat, "total_pgmajfault")))
		sscanf(ptr, "total_pgmajfault %lu", &stats->total_pgmajfault);





		prec->ssec = cgroup_acct_data->ssec;
	}

	if (cgroup_acct_data->total_inactive_anon == NO_VAL64 &&
	    cgroup_acct_data->total_active_anon == NO_VAL64 &&
	    cgroup_acct_data->total_pgmajfault == NO_VAL64) {
		debug2("failed to collect cgroup memory stats pid %d ppid %d",
		       prec->pid, prec->ppid);
	} else {
		/*
		 * use rss+tmpfs instead of just rss. this means we include /dev/shm etc. usage
		 * rss + tmpfs == total_inactive_anon + total_active_anon
		 * see 
		 *    https://access.redhat.com/documentation/en-us/red_hat_enterprise_linux/6/html/resource_management_guide/sec-memory
		 */
		prec->tres_data[TRES_ARRAY_MEM].size_read =
			cgroup_acct_data->total_inactive_anon + cgroup_acct_data->total_active_anon;

		/*
		 * total_pgmajfault is what is reported in proc, so we use

Return to ticket 13950

Lines 1591-1604 extern cgroup_acct_t *cgroup_p_task_get_ Link Here

(-)slurm-21.08.7.orig/src/plugins/cgroup/v1/cgroup_v1.c (-3 / +6 lines)
1591	stats = xmalloc(sizeof(*stats));	1591	stats = xmalloc(sizeof(*stats));
1592	stats->usec = NO_VAL64;	1592	stats->usec = NO_VAL64;
1593	stats->ssec = NO_VAL64;	1593	stats->ssec = NO_VAL64;
1594	stats->total_rss = NO_VAL64;	1594	stats->total_inactive_anon = NO_VAL64;
		1595	stats->total_active_anon = NO_VAL64;
1595	stats->total_pgmajfault = NO_VAL64;	1596	stats->total_pgmajfault = NO_VAL64;
1596		1597
1597	if (cpu_time != NULL)	1598	if (cpu_time != NULL)
1598	sscanf(cpu_time, "%s %lu %s %lu", &stats->usec, &stats->ssec);	1599	sscanf(cpu_time, "%s %lu %s %lu", &stats->usec, &stats->ssec);
1599		1600
1600	if ((ptr = xstrstr(memory_stat, "total_rss")))	1601	if ((ptr = xstrstr(memory_stat, "total_inactive_anon")))
1601	sscanf(ptr, "total_rss %lu", &stats->total_rss);	1602	sscanf(ptr, "total_inactive_anon %lu", &stats->total_inactive_anon);
		1603	if ((ptr = xstrstr(memory_stat, "total_active_anon")))
		1604	sscanf(ptr, "total_active_anon %lu", &stats->total_active_anon);
1602	if ((ptr = xstrstr(memory_stat, "total_pgmajfault")))	1605	if ((ptr = xstrstr(memory_stat, "total_pgmajfault")))
1603	sscanf(ptr, "total_pgmajfault %lu", &stats->total_pgmajfault);	1606	sscanf(ptr, "total_pgmajfault %lu", &stats->total_pgmajfault);
1604		1607

Lines 105-123 static void _prec_extra(jag_prec_t *prec Link Here

(-)slurm-21.08.7.orig/src/plugins/jobacct_gather/cgroup/jobacct_gather_cgroup.c (-6 / +7 lines)
105	prec->ssec = cgroup_acct_data->ssec;	105	prec->ssec = cgroup_acct_data->ssec;
106	}	106	}
107		107
108	if (cgroup_acct_data->total_rss == NO_VAL64 &&	108	if (cgroup_acct_data->total_inactive_anon == NO_VAL64 &&
		109	cgroup_acct_data->total_active_anon == NO_VAL64 &&
109	cgroup_acct_data->total_pgmajfault == NO_VAL64) {	110	cgroup_acct_data->total_pgmajfault == NO_VAL64) {
110	debug2("failed to collect cgroup memory stats pid %d ppid %d",	111	debug2("failed to collect cgroup memory stats pid %d ppid %d",
111	prec->pid, prec->ppid);	112	prec->pid, prec->ppid);
112	} else {	113	} else {
113	/*	114	/*
114	* This number represents the amount of "dirty" private memory	115	* use rss+tmpfs instead of just rss. this means we include /dev/shm etc. usage
115	* used by the cgroup. From our experience this is slightly	116	* rss + tmpfs == total_inactive_anon + total_active_anon
116	* different than what proc presents, but is probably more	117	* see
117	* accurate on what the user is actually using.	118	* https://access.redhat.com/documentation/en-us/red_hat_enterprise_linux/6/html/resource_management_guide/sec-memory
118	*/	119	*/
119	prec->tres_data[TRES_ARRAY_MEM].size_read =	120	prec->tres_data[TRES_ARRAY_MEM].size_read =
120	cgroup_acct_data->total_rss;	121	cgroup_acct_data->total_inactive_anon + cgroup_acct_data->total_active_anon;
121		122
122	/*	123	/*
123	* total_pgmajfault is what is reported in proc, so we use	124	* total_pgmajfault is what is reported in proc, so we use