Attachment #20321 for ticket #12004

Lines 1060-1067 Note that the \fBHealthCheckProgram\fR will be executed at the same time Link Here

(-)a/doc/man/man5/slurm.conf.5 (-2 / +2 lines)
1060	on all nodes to minimize its impact upon parallel programs.	1060	on all nodes to minimize its impact upon parallel programs.
1061	This program is will be killed if it does not terminate normally within	1061	This program is will be killed if it does not terminate normally within
1062	60 seconds.	1062	60 seconds.
1063	This program will also be executed when the slurmd daemon is first started and	1063	This program will also be executed when the slurmd daemon is first started
1064	before it registers with the slurmctld daemon.	1064	(with a 20 minute timeout) before it registers with the slurmctld daemon.
1065	By default, no program will be executed.	1065	By default, no program will be executed.
1066		1066
1067	.TP	1067	.TP

Lines 3042-3048 static void _rpc_health_check(slurm_msg_t *msg) Link Here

(-)a/src/slurmd/slurmd/req.c (-1 / +1 lines)
3042	}	3042	}
3043		3043
3044	if (rc == SLURM_SUCCESS)	3044	if (rc == SLURM_SUCCESS)
3045	rc = run_script_health_check();	3045	rc = run_script_health_check(false);
3046		3046
3047	/* Take this opportunity to enforce any job memory limits */	3047	/* Take this opportunity to enforce any job memory limits */
3048	_enforce_job_mem_limit();	3048	_enforce_job_mem_limit();




	slurm_conf_install_fork_handlers();
	record_launched_jobs();

	run_script_health_check(true);

	slurm_thread_create_detached(NULL, _registration_engine, NULL);


 * Returns the run result. If the health check program
 * is not defined, returns success immediately.
 */
extern int run_script_health_check(bool startup)
{
	int rc = SLURM_SUCCESS;


		setenvf(&env, "SLURMD_NODENAME", "%s", conf->node_name);

		rc = run_script("health_check", slurm_conf.health_check_program,
				0, startup ? 1200 : 60, env, 0);

		env_array_free(env);
	}

Lines 170-176 int send_registration_msg(uint32_t status, bool startup); Link Here

(-)a/src/slurmd/slurmd/slurmd.h (-2 / +1 lines)
170	int save_cred_state(slurm_cred_ctx_t vctx);	170	int save_cred_state(slurm_cred_ctx_t vctx);
171		171
172	/* Run the health check program if configured */	172	/* Run the health check program if configured */
173	int run_script_health_check(void);	173	int run_script_health_check(bool startup);
174		174
175	/* Handler for SIGTERM; can also be called to shutdown the slurmd. */	175	/* Handler for SIGTERM; can also be called to shutdown the slurmd. */
176	void slurmd_shutdown(int signum);	176	void slurmd_shutdown(int signum);
177	-

Return to ticket 12004

Lines 381-387 main (int argc, char **argv) Link Here

(-)a/src/slurmd/slurmd/slurmd.c (-3 / +3 lines)
381	slurm_conf_install_fork_handlers();	381	slurm_conf_install_fork_handlers();
382	record_launched_jobs();	382	record_launched_jobs();
383		383
384	run_script_health_check();	384	run_script_health_check(true);
385		385
386	slurm_thread_create_detached(NULL, _registration_engine, NULL);	386	slurm_thread_create_detached(NULL, _registration_engine, NULL);
387		387
Lines 2611-2617 static void _resource_spec_fini(void) Link Here
2611	* Returns the run result. If the health check program	2611	* Returns the run result. If the health check program
2612	* is not defined, returns success immediately.	2612	* is not defined, returns success immediately.
2613	*/	2613	*/
2614	extern int run_script_health_check(void)	2614	extern int run_script_health_check(bool startup)
2615	{	2615	{
2616	int rc = SLURM_SUCCESS;	2616	int rc = SLURM_SUCCESS;
2617		2617
Lines 2621-2627 extern int run_script_health_check(void) Link Here
2621	setenvf(&env, "SLURMD_NODENAME", "%s", conf->node_name);	2621	setenvf(&env, "SLURMD_NODENAME", "%s", conf->node_name);
2622		2622
2623	rc = run_script("health_check", slurm_conf.health_check_program,	2623	rc = run_script("health_check", slurm_conf.health_check_program,
2624	0, 60, env, 0);	2624	0, startup ? 1200 : 60, env, 0);
2625		2625
2626	env_array_free(env);	2626	env_array_free(env);
2627	}	2627	}