Attachment #540 for ticket #447




#define PATH_MAX 256
#endif

/* One slurmstepd could be in the process of creating cgroups while another
 * slurmstepd is simultaneoulsy deleting cgroups for another step for that
 * same user. MAX_CGROUP_RETRY defines how many times that we retry creating
 * the user and job cgroup on an error of ENOENT. */
#ifndef MAX_CGROUP_RETRY
#define MAX_CGROUP_RETRY 8
#endif

static slurm_cgroup_conf_t slurm_cgroup_conf;

static char user_cgroup_path[PATH_MAX];

{
	/* we do it here as we do not have access to the conf structure */
	/* in libslurm (src/common/xcgroup.c) */
	int retry_count = 0;	/* See MAX_CGROUP_RETRY description above */
	xcgroup_t slurm_cg;
	char* pre = (char*) xstrdup(slurm_cgroup_conf.cgroup_prepend);
#ifdef MULTIPLE_SLURMD

		return SLURM_ERROR;
	}

	xcgroup_lock(&user_freezer_cg);
	if ((xcgroup_instanciate(&user_freezer_cg) != XCGROUP_SUCCESS) ||
	    (xcgroup_instanciate(&job_freezer_cg)  != XCGROUP_SUCCESS) ||
	    (xcgroup_instanciate(&step_freezer_cg) != XCGROUP_SUCCESS)) {
		xcgroup_unlock(&user_freezer_cg);
			goto retry;
		xcgroup_destroy(&user_freezer_cg);
		xcgroup_destroy(&job_freezer_cg);
		xcgroup_destroy(&step_freezer_cg);
		return SLURM_ERROR;
	}
	xcgroup_unlock(&user_freezer_cg);

	/* inhibit release agent for the step cgroup thus letting 
	 * slurmstepd being able to add new pids to the container 


int _slurm_cgroup_destroy(void)
{
	xcgroup_lock(&user_freezer_cg);
	if (jobstep_cgroup_path[0] != '\0') {
		if ( xcgroup_delete(&step_freezer_cg) != XCGROUP_SUCCESS ) {
			xcgroup_unlock(&user_freezer_cg);
			return SLURM_ERROR;
		}
		xcgroup_destroy(&step_freezer_cg);
	}


		xcgroup_delete(&user_freezer_cg);
		xcgroup_destroy(&user_freezer_cg);
	}
	xcgroup_unlock(&user_freezer_cg);
	xcgroup_ns_destroy(&freezer_ns);

	return SLURM_SUCCESS;

Return to ticket 447

Lines 109-122 const uint32_t plugin_version = 91; Link Here

(-)a/src/plugins/proctrack/cgroup/proctrack_cgroup.c (-14 / +9 lines)
109	#define PATH_MAX 256	109	#define PATH_MAX 256
110	#endif	110	#endif
111		111
112	/* One slurmstepd could be in the process of creating cgroups while another
113	* slurmstepd is simultaneoulsy deleting cgroups for another step for that
114	* same user. MAX_CGROUP_RETRY defines how many times that we retry creating
115	* the user and job cgroup on an error of ENOENT. */
116	#ifndef MAX_CGROUP_RETRY
117	#define MAX_CGROUP_RETRY 8
118	#endif
119
120	static slurm_cgroup_conf_t slurm_cgroup_conf;	112	static slurm_cgroup_conf_t slurm_cgroup_conf;
121		113
122	static char user_cgroup_path[PATH_MAX];	114	static char user_cgroup_path[PATH_MAX];
Lines 151-157 int _slurm_cgroup_create(slurmd_job_t *job, uint64_t id, uid_t uid, gid_t gid) Link Here
151	{	143	{
152	/* we do it here as we do not have access to the conf structure */	144	/* we do it here as we do not have access to the conf structure */
153	/* in libslurm (src/common/xcgroup.c) */	145	/* in libslurm (src/common/xcgroup.c) */
154	int retry_count = 0; /* See MAX_CGROUP_RETRY description above */
155	xcgroup_t slurm_cg;	146	xcgroup_t slurm_cg;
156	char* pre = (char*) xstrdup(slurm_cgroup_conf.cgroup_prepend);	147	char* pre = (char*) xstrdup(slurm_cgroup_conf.cgroup_prepend);
157	#ifdef MULTIPLE_SLURMD	148	#ifdef MULTIPLE_SLURMD
Lines 243-258 int _slurm_cgroup_create(slurmd_job_t *job, uint64_t id, uid_t uid, gid_t gid) Link Here
243	return SLURM_ERROR;	234	return SLURM_ERROR;
244	}	235	}
245		236
246	retry: if ((xcgroup_instanciate(&user_freezer_cg) != XCGROUP_SUCCESS) \|\|	237	xcgroup_lock(&user_freezer_cg);
		238	if ((xcgroup_instanciate(&user_freezer_cg) != XCGROUP_SUCCESS) \|\|
247	(xcgroup_instanciate(&job_freezer_cg) != XCGROUP_SUCCESS) \|\|	239	(xcgroup_instanciate(&job_freezer_cg) != XCGROUP_SUCCESS) \|\|
248	(xcgroup_instanciate(&step_freezer_cg) != XCGROUP_SUCCESS)) {	240	(xcgroup_instanciate(&step_freezer_cg) != XCGROUP_SUCCESS)) {
249	if ((errno == ENOENT) && (++retry_count <= MAX_CGROUP_RETRY))	241	xcgroup_unlock(&user_freezer_cg);
250	goto retry;
251	xcgroup_destroy(&user_freezer_cg);	242	xcgroup_destroy(&user_freezer_cg);
252	xcgroup_destroy(&job_freezer_cg);	243	xcgroup_destroy(&job_freezer_cg);
253	xcgroup_destroy(&step_freezer_cg);	244	xcgroup_destroy(&step_freezer_cg);
254	return SLURM_ERROR;	245	return SLURM_ERROR;
255	}	246	}
		247	xcgroup_unlock(&user_freezer_cg);
256		248
257	/* inhibit release agent for the step cgroup thus letting	249	/* inhibit release agent for the step cgroup thus letting
258	* slurmstepd being able to add new pids to the container	250	* slurmstepd being able to add new pids to the container
Lines 264-272 retry: if ((xcgroup_instanciate(&user_freezer_cg) != XCGROUP_SUCCESS) \|\| Link Here
264		256
265	int _slurm_cgroup_destroy(void)	257	int _slurm_cgroup_destroy(void)
266	{	258	{
		259	xcgroup_lock(&user_freezer_cg);
267	if (jobstep_cgroup_path[0] != '\0') {	260	if (jobstep_cgroup_path[0] != '\0') {
268	if ( xcgroup_delete(&step_freezer_cg) != XCGROUP_SUCCESS )	261	if ( xcgroup_delete(&step_freezer_cg) != XCGROUP_SUCCESS ) {
		262	xcgroup_unlock(&user_freezer_cg);
269	return SLURM_ERROR;	263	return SLURM_ERROR;
		264	}
270	xcgroup_destroy(&step_freezer_cg);	265	xcgroup_destroy(&step_freezer_cg);
271	}	266	}
272		267
Lines 279-285 int _slurm_cgroup_destroy(void) Link Here
279	xcgroup_delete(&user_freezer_cg);	274	xcgroup_delete(&user_freezer_cg);
280	xcgroup_destroy(&user_freezer_cg);	275	xcgroup_destroy(&user_freezer_cg);
281	}	276	}
282		277	xcgroup_unlock(&user_freezer_cg);
283	xcgroup_ns_destroy(&freezer_ns);	278	xcgroup_ns_destroy(&freezer_ns);
284		279
285	return SLURM_SUCCESS;	280	return SLURM_SUCCESS;