Merge branch 'slurm-23.02'

NEWS

@@ -87,6 +87,8 @@ documents those changes that are of interest to users and administrators.
     slurm_spank_job_epilog().
  -- Fix srun < 23.02 always getting an "exact" core allocation.
  -- Prevent scontrol < 23.02 from setting MaxCPUsPerSocket to 0.
+ -- Add ScronParameters=explicit_scancel and corresponding scancel --cron
+    option.
 
 * Changes in Slurm 23.02.0
 ==========================

doc/man/man1/scancel.1

@@ -65,6 +65,12 @@ This is the default behavior on architectures using front end nodes (e.g.
 Cray ALPS computers) or when the \fB\-\-clusters\fR option is used.
 .IP
 
+.TP
+\fB\-c\fR, \fB\-\-cron\fR
+Confirm request to cancel a job submitted by scrontab.  This option only has
+effect with the "explicit_scancel" option is set in \fBScronParameters\fR.
+.IP
+
 .TP
 \fB\-f\fR, \fB\-\-full\fR
 By default, signals other than SIGKILL are not sent to the batch step (the shell
@@ -240,6 +246,11 @@ environment variables, along with their corresponding options, are listed below.
 \fB\-\-ctld\fR
 .IP
 
+.TP
+\fBSCANCEL_CRON\fR
+\fB\-c, \-\-cron\fR
+.IP
+
 .TP
 \fBSCANCEL_FULL\fR
 \fB\-f, \-\-full\fR

doc/man/man5/slurm.conf.5

@@ -4434,6 +4434,12 @@ Multiple options may be comma separated.
 .TP
 \fBenable\fR
 Enable the use of scrontab to submit and manage periodic repeating jobs.
+.IP
+
+.TP
+\fBexplicit_scancel\fR
+When cancelling an scrontab job, require the user to explicitly request
+cancelling the job with the --cron flag in scancel.
 .RE
 .IP
 

slurm/slurm.h

@@ -3596,7 +3596,8 @@ typedef struct job_step_kill_msg {
 #define KILL_NO_SIBS     SLURM_BIT(7) /* Don't kill other sibling jobs */
 #define KILL_JOB_RESV    SLURM_BIT(8) /* Job is willing to run on nodes in a
 				       * magnetic reservation. */
-#define KILL_NO_SIG_FAIL     SLURM_BIT(9) /* Don't fail job due to signal (steps only) */
+#define KILL_NO_CRON     SLURM_BIT(9) /* request killing cron Jobs */
+#define KILL_NO_SIG_FAIL SLURM_BIT(10) /* Don't fail job due to signal (steps only) */
 
 /* Use top bit of uint16_t in conjuction with KILL_* flags to indicate signal
  * has been sent to job previously. Does not need to be passed to slurmd. */

slurm/slurm_errno.h

@@ -181,6 +181,7 @@ typedef enum {
 	ESLURM_PRIO_RESET_FAIL,
 	ESLURM_CANNOT_MODIFY_CRON_JOB,
 	ESLURM_INVALID_JOB_CONTAINER_CHANGE,
+	ESLURM_CANNOT_CANCEL_CRON_JOB,
 
 	ESLURM_INVALID_MCS_LABEL =			2099,
 	ESLURM_BURST_BUFFER_WAIT =			2100,

src/common/slurm_errno.c

@@ -316,6 +316,8 @@ slurm_errtab_t slurm_errtab[] = {
 	  "Cannot modify scrontab jobs through scontrol"	},
 	{ ERRTAB_ENTRY(ESLURM_INVALID_JOB_CONTAINER_CHANGE),
 	  "JobContainerType change requires restart of all Slurm daemons and commands to take effect" },
+	{ ERRTAB_ENTRY(ESLURM_CANNOT_CANCEL_CRON_JOB),
+	  "Cannot cancel scrontab jobs without --cron flag."	},
 	{ ERRTAB_ENTRY(ESLURM_INVALID_MCS_LABEL),
 	  "Invalid mcs_label specified"				},
 	{ ERRTAB_ENTRY(ESLURM_BURST_BUFFER_WAIT),

src/scancel/opt.c

@@ -187,6 +187,7 @@ static void _opt_default(void)
 #else
 	opt.ctld	= false;
 #endif
+	opt.cron = false;
 	opt.full	= false;
 	opt.hurry	= false;
 	opt.interactive	= false;
@@ -247,6 +248,9 @@ static void _opt_env(void)
 	if (getenv("SCANCEL_CTLD"))
 		opt.ctld = true;
 
+	if (getenv("SCANCEL_CRON"))
+		opt.cron = true;
+
 	if ( (val=getenv("SCANCEL_FULL")) ) {
 		if (xstrcasecmp(val, "true") == 0)
 			opt.full       = true;
@@ -338,6 +342,7 @@ static void _opt_args(int argc, char **argv)
 		{"account",	required_argument, 0, 'A'},
 		{"batch",	no_argument,       0, 'b'},
 		{"ctld",	no_argument,	   0, OPT_LONG_CTLD},
+		{"cron",	no_argument,	   0, 'c'},
 		{"full",	no_argument,       0, 'f'},
 		{"help",        no_argument,       0, OPT_LONG_HELP},
 		{"hurry",       no_argument,       0, 'H'},
@@ -363,7 +368,8 @@ static void _opt_args(int argc, char **argv)
 		{NULL,          0,                 0, 0}
 	};
 
-	while ((opt_char = getopt_long(argc, argv, "A:bfHiM:n:p:Qq:R:s:t:u:vVw:",
+	while ((opt_char = getopt_long(argc, argv,
+				       "A:bcfHiM:n:p:Qq:R:s:t:u:vVw:",
 				       long_options, &option_index)) != -1) {
 		switch (opt_char) {
 		case (int)'?':
@@ -381,6 +387,9 @@ static void _opt_args(int argc, char **argv)
 		case OPT_LONG_CTLD:
 			opt.ctld = true;
 			break;
+		case (int)'c':
+			opt.cron = true;
+			break;
 		case (int)'f':
 			opt.full = true;
 			break;
@@ -627,6 +636,7 @@ static void _opt_list(void)
 	info("account        : %s", opt.account);
 	info("batch          : %s", tf_(opt.batch));
 	info("ctld           : %s", tf_(opt.ctld));
+	info("cron           : %s", tf_(opt.cron));
 	info("full           : %s", tf_(opt.full));
 	info("hurry          : %s", tf_(opt.hurry));
 	info("interactive    : %s", tf_(opt.interactive));
@@ -697,6 +707,7 @@ static void _help(void)
 	printf("  -A, --account=account           act only on jobs charging this account\n");
 	printf("  -b, --batch                     signal batch shell for specified job\n");
 /*	printf("      --ctld                      send request directly to slurmctld\n"); */
+	printf("  -c, --cron                      cancel an scrontab job\n");
 	printf("  -f, --full                      signal batch shell and all steps for specified job\n");
 	printf("  -H, --hurry                     avoid burst buffer stage out\n");
 	printf("  -i, --interactive               require response from user for each job\n");

src/scancel/scancel.c

@@ -729,6 +729,18 @@ _cancel_job_id (void *ci)
 		flags |= KILL_JOB_BATCH;
 		job_type = "batch ";
 	}
+
+	/*
+	 * With the introduction of the ScronParameters=explicit_scancel option,
+	 * scancel requests for a cron job should be rejected unless the --cron
+	 * flag is specified.
+	 * To prevent introducing this option from influencing anything other
+	 * than user requests, it has been set up so that when KILL_NO_CRON is
+	 * set when explicit_scancel is also set, the request will be rejected.
+	 */
+	if (!opt.cron)
+		flags |= KILL_NO_CRON;
+
 	if (opt.full) {
 		flags |= KILL_FULL_JOB;
 		job_type = "full ";

src/scancel/scancel.h

@@ -49,6 +49,7 @@ typedef struct scancel_options {
 	char *sibling;		/* --sibling=<sib_name>		*/
 	bool ctld;		/* --ctld			*/
 	List clusters;          /* --cluster=cluster_name -Mcluster-name */
+	bool cron;		/* --cron */
 	bool full;		/* --full, -f			*/
 	bool hurry;		/* --hurry, -H			*/
 	bool interactive;	/* --interactive, -i		*/

src/slurmctld/job_mgr.c

@@ -5433,6 +5433,16 @@ extern int job_signal(job_record_t *job_ptr, uint16_t signal,
 	if (job_ptr->bit_flags & CRON_JOB) {
 		cron_entry_t *entry =
 			(cron_entry_t *) job_ptr->details->crontab_entry;
+		/*
+		 * The KILL_NO_CRON flag being set here is indicating that the
+		 * user has NOT specifically requested killing scrontab jobs. To
+		 * avoid interfering with other possible ways of killing jobs,
+		 * the KILL_NO_CRON flag being unset must mean that killing cron
+		 * jobs is permitted.
+		 */
+		if (xstrcasestr(slurm_conf.scron_params, "explicit_scancel") &&
+		    (flags & KILL_NO_CRON))
+			return ESLURM_CANNOT_CANCEL_CRON_JOB;
 		job_ptr->bit_flags |= ~CRON_JOB;
 		error("cancelling cron job, lines %u %u",
 		      entry->line_start, entry->line_end);