Add flags argument to slurm_kill_job_step()

job_step_kill_msg_t already has a flags field to populate. Bug 11360
ubccr-slurm-simulator · Mar 22, 2023 · 165f410 · 165f410
1 parent 293656b
commit 165f410
Show file tree

Hide file tree

Showing 22 changed files with 46 additions and 41 deletions.
diff --git a/RELEASE_NOTES b/RELEASE_NOTES
@@ -36,3 +36,4 @@ API CHANGES
  -- cli_filter/lua - return nil for unset time options rather than the string
     "2982616-04:14:00" (which is the internal macro "NO_VAL" represented as
     time string).
+  - "flags" argument was added to slurm_kill_job_step().
diff --git a/contribs/perlapi/libslurm/perl/Slurm.xs b/contribs/perlapi/libslurm/perl/Slurm.xs
@@ -484,16 +484,15 @@ slurm_kill_job(slurm_t self, uint32_t job_id, uint16_t signal, uint16_t batch_fl
 	C_ARGS:
 		job_id, signal, batch_flag
 
-int
-slurm_kill_job_step(slurm_t self, uint32_t job_id, uint32_t step_id, uint16_t signal)
+int slurm_kill_job_step(slurm_t self, uint32_t job_id, uint32_t step_id, uint16_t signal, uint16_t flags)
 	INIT:
 		if (self); /* this is needed to avoid a warning about
 			      unused variables.  But if we take slurm_t self
 			      out of the mix Slurm-> doesn't work,
 			      only Slurm::
 			    */
 	C_ARGS:
-		job_id, step_id, signal
+		job_id, step_id, signal, flags
 
 int
 slurm_signal_job(slurm_t self, uint32_t job_id, uint16_t signal)

diff --git a/slurm/slurm.h b/slurm/slurm.h
@@ -3615,11 +3615,11 @@ extern int slurm_kill_job(uint32_t job_id, uint16_t signal, uint16_t flags);
  * IN job_id  - the job's id
  * IN step_id - the job step's id
  * IN signal  - signal number
+ * IN flags - see KILL_* or 0 for no flags
  * RET SLURM_SUCCESS on success, otherwise return SLURM_ERROR with errno set
  */
-extern int slurm_kill_job_step(uint32_t job_id,
-			       uint32_t step_id,
-			       uint16_t signal);
+extern int slurm_kill_job_step(uint32_t job_id, uint32_t step_id,
+			       uint16_t signal, uint16_t flags);
 /*
  * slurm_kill_job2 - send REQUEST_KILL_JOB msg to an existing job or step.
  * IN job_id - the job's id (in a string format)

diff --git a/src/api/cancel.c b/src/api/cancel.c
@@ -108,8 +108,8 @@ slurm_kill_job (uint32_t job_id, uint16_t signal, uint16_t flags)
  * IN signal     - signal number
  * RET SLURM_SUCCESS on success, otherwise return SLURM_ERROR with errno set
  */
-extern int
-slurm_kill_job_step (uint32_t job_id, uint32_t step_id, uint16_t signal)
+extern int slurm_kill_job_step(uint32_t job_id, uint32_t step_id,
+			       uint16_t signal, uint16_t flags)
 {
 	int rc;
 	slurm_msg_t msg;
@@ -125,7 +125,7 @@ slurm_kill_job_step (uint32_t job_id, uint32_t step_id, uint16_t signal)
 	req.step_id.step_id = step_id;
 	req.step_id.step_het_comp = NO_VAL;
 	req.signal      = signal;
-	req.flags	= 0;
+	req.flags = flags;
 	msg.msg_type    = REQUEST_CANCEL_JOB_STEP;
 	msg.data        = &req;
 

diff --git a/src/api/slurm_pmi.c b/src/api/slurm_pmi.c
@@ -388,5 +388,5 @@ void slurm_pmi_finalize(void)
 extern int slurm_pmi_kill_job_step(uint32_t job_id, uint32_t step_id,
 				   uint16_t signal)
 {
-	return slurm_kill_job_step(job_id, step_id, signal);
+	return slurm_kill_job_step(job_id, step_id, signal, 0);
 }
diff --git a/src/api/step_launch.c b/src/api/step_launch.c
@@ -624,7 +624,7 @@ static void _step_abort(slurm_step_ctx_t *ctx)
 
 	if (!sls->abort_action_taken) {
 		slurm_kill_job_step(ctx->job_id, ctx->step_resp->job_step_id,
-				    SIGKILL);
+				    SIGKILL, 0);
 		sls->abort_action_taken = true;
 	}
 }
@@ -694,7 +694,7 @@ void slurm_step_launch_wait_finish(slurm_step_ctx_t *ctx)
 				slurm_kill_job_step(ctx->job_id,
 						    ctx->step_resp->
 						    job_step_id,
-						    SIGKILL);
+						    SIGKILL, 0);
 				sls->abort_action_taken = true;
 			}
 			if (!time_set) {
@@ -725,7 +725,7 @@ void slurm_step_launch_wait_finish(slurm_step_ctx_t *ctx)
 				 */
 				slurm_kill_job_step(ctx->job_id,
 						    ctx->step_resp->job_step_id,
-						    SIGKILL);
+						    SIGKILL, 0);
 				client_io_handler_abort(sls->io);
 				break;
 			} else if (errnum != 0) {

diff --git a/src/plugins/mpi/pmi2/pmi1.c b/src/plugins/mpi/pmi2/pmi1.c
@@ -192,7 +192,7 @@ _handle_barrier_in(int fd, int lrank, client_req_t *req)
 			/* cancel the step to avoid tasks hang */
 			slurm_kill_job_step(job_info.step_id.job_id,
 					    job_info.step_id.step_id,
-					    SIGKILL);
+					    SIGKILL, 0);
 		} else {
 			waiting_kvs_resp = 1;
 		}
@@ -228,7 +228,7 @@ _handle_abort(int fd, int lrank, client_req_t *req)
 	debug3("mpi/pmi2: in _handle_abort");
 	/* no response needed. just cancel the job */
 	slurm_kill_job_step(job_info.step_id.job_id, job_info.step_id.step_id,
-			    SIGKILL);
+			    SIGKILL, 0);
 	debug3("mpi/pmi2: out _handle_abort");
 	return SLURM_SUCCESS;
 }

diff --git a/src/plugins/mpi/pmi2/pmi2.c b/src/plugins/mpi/pmi2/pmi2.c
@@ -188,7 +188,7 @@ _handle_abort(int fd, int lrank, client_req_t *req)
 	/* no response needed. just cancel the job step if required */
 	if (is_world) {
 		slurm_kill_job_step(job_info.step_id.job_id,
-				    job_info.step_id.step_id, SIGKILL);
+				    job_info.step_id.step_id, SIGKILL, 0);
 	}
 	return rc;
 }
@@ -308,7 +308,7 @@ _handle_kvs_fence(int fd, int lrank, client_req_t *req)
 			/* cancel the step to avoid tasks hang */
 			slurm_kill_job_step(job_info.step_id.job_id,
 					    job_info.step_id.step_id,
-					    SIGKILL);
+					    SIGKILL, 0);
 		} else {
 			waiting_kvs_resp = 1;
 		}

diff --git a/src/plugins/mpi/pmi2/ring.c b/src/plugins/mpi/pmi2/ring.c
@@ -232,7 +232,7 @@ static int pmix_stepd_send(const char* buf, uint32_t size, int rank)
 			/* cancel the step to avoid tasks hang */
 			slurm_kill_job_step(job_info.step_id.job_id,
 					    job_info.step_id.step_id,
-					    SIGKILL);
+					    SIGKILL, 0);
 		}
 
 		/* didn't succeeded, but we'll retry again,

diff --git a/src/plugins/mpi/pmi2/tree.c b/src/plugins/mpi/pmi2/tree.c
@@ -152,7 +152,7 @@ static int _handle_kvs_fence(int fd, buf_t *buf)
 			/* cancel the step to avoid tasks hang */
 			slurm_kill_job_step(job_info.step_id.job_id,
 					    job_info.step_id.step_id,
-					    SIGKILL);
+					    SIGKILL, 0);
 		} else {
 			if (in_stepd())
 				waiting_kvs_resp = 1;
@@ -213,7 +213,7 @@ static int _handle_kvs_fence_resp(int fd, buf_t *buf)
 	send_kvs_fence_resp_to_clients(rc, errmsg);
 	if (rc != SLURM_SUCCESS) {
 		slurm_kill_job_step(job_info.step_id.job_id,
-				    job_info.step_id.step_id, SIGKILL);
+				    job_info.step_id.step_id, SIGKILL, 0);
 	}
 	return rc;
 

diff --git a/src/plugins/mpi/pmix/mpi_pmix.c b/src/plugins/mpi/pmix/mpi_pmix.c
@@ -231,7 +231,7 @@ extern int mpi_p_slurmstepd_prefork(const stepd_step_rec_t *step, char ***env)
 err_ext:
 	/* Abort the whole job if error! */
 	slurm_kill_job_step(step->step_id.job_id,
-			    step->step_id.step_id, SIGKILL);
+			    step->step_id.step_id, SIGKILL, 0);
 	return ret;
 }
 

diff --git a/src/plugins/mpi/pmix/pmixp_client.c b/src/plugins/mpi/pmix/pmixp_client.c
@@ -783,7 +783,8 @@ extern int pmixp_lib_abort(int status, void *cbfunc, void *cbdata)
 	 */
 	pmixp_abort_propagate(status);
 
-	slurm_kill_job_step(pmixp_info_jobid(), pmixp_info_stepid(), SIGKILL);
+	slurm_kill_job_step(pmixp_info_jobid(), pmixp_info_stepid(), SIGKILL,
+			    0);
 
 	if (abort_cbfunc)
 		abort_cbfunc(PMIX_SUCCESS, cbdata);

diff --git a/src/plugins/mpi/pmix/pmixp_client_v2.c b/src/plugins/mpi/pmix/pmixp_client_v2.c
@@ -210,7 +210,8 @@ static void _errhandler(size_t evhdlr_registration_id,
 	/* FIXME: use proper specificator for nranges */
 	PMIXP_ERROR("Error handler invoked: status = %d, source = [%s:%d]",
 		    (int) status, source->nspace, source->rank);
-	slurm_kill_job_step(pmixp_info_jobid(), pmixp_info_stepid(), SIGKILL);
+	slurm_kill_job_step(pmixp_info_jobid(), pmixp_info_stepid(), SIGKILL,
+			    0);
 }
 
 static pmix_server_module_t slurm_pmix_cb = {

diff --git a/src/plugins/mpi/pmix/pmixp_coll_ring.c b/src/plugins/mpi/pmix/pmixp_coll_ring.c
@@ -629,7 +629,7 @@ int pmixp_coll_ring_check(pmixp_coll_t *coll, pmixp_coll_ring_msg_hdr_t *hdr)
 			    hdr->seq, nodename, hdr->nodeid, coll->seq);
 		pmixp_debug_hang(0); /* enable hang to debug this! */
 		slurm_kill_job_step(pmixp_info_jobid(),
-				    pmixp_info_stepid(), SIGKILL);
+				    pmixp_info_stepid(), SIGKILL, 0);
 		xfree(nodename);
 		return SLURM_SUCCESS;
 	} else if (PMIXP_COLL_REQ_SKIP == rc) {

diff --git a/src/plugins/mpi/pmix/pmixp_coll_tree.c b/src/plugins/mpi/pmix/pmixp_coll_tree.c
@@ -190,7 +190,7 @@ static void _reset_coll(pmixp_coll_t *coll)
 		/* collective is spoiled, reset state */
 		tree->state = PMIXP_COLL_TREE_SYNC;
 		slurm_kill_job_step(pmixp_info_jobid(),
-				    pmixp_info_stepid(), SIGKILL);
+				    pmixp_info_stepid(), SIGKILL, 0);
 	}
 }
 
@@ -610,7 +610,7 @@ static int _progress_ufwd(pmixp_coll_t *coll)
 		/* collective is spoiled, reset state */
 		tree->state = PMIXP_COLL_TREE_SYNC;
 		slurm_kill_job_step(pmixp_info_jobid(),
-				    pmixp_info_stepid(), SIGKILL);
+				    pmixp_info_stepid(), SIGKILL, 0);
 		return false;
 	}
 
@@ -742,7 +742,7 @@ static int _progress_ufwd_sc(pmixp_coll_t *coll)
 		/* collective is spoiled, reset state */
 		tree->state = PMIXP_COLL_TREE_SYNC;
 		slurm_kill_job_step(pmixp_info_jobid(),
-				    pmixp_info_stepid(), SIGKILL);
+				    pmixp_info_stepid(), SIGKILL, 0);
 		return false;
 	}
 
@@ -836,7 +836,7 @@ static int _progress_dfwd(pmixp_coll_t *coll)
 		/* collective is spoiled, reset state */
 		tree->state = PMIXP_COLL_TREE_SYNC;
 		slurm_kill_job_step(pmixp_info_jobid(),
-				    pmixp_info_stepid(), SIGKILL);
+				    pmixp_info_stepid(), SIGKILL, 0);
 		return false;
 	}
 #ifdef PMIXP_COLL_DEBUG
@@ -944,7 +944,7 @@ int pmixp_coll_tree_local(pmixp_coll_t *coll, char *data, size_t size,
 		/* collective is spoiled, reset state */
 		tree->state = PMIXP_COLL_TREE_SYNC;
 		slurm_kill_job_step(pmixp_info_jobid(),
-				    pmixp_info_stepid(), SIGKILL);
+				    pmixp_info_stepid(), SIGKILL, 0);
 		ret = SLURM_ERROR;
 		goto exit;
 	}
@@ -1143,7 +1143,7 @@ int pmixp_coll_tree_child(pmixp_coll_t *coll, uint32_t peerid, uint32_t seq,
 	_reset_coll(coll);
 error2:
 	slurm_kill_job_step(pmixp_info_jobid(),
-			    pmixp_info_stepid(), SIGKILL);
+			    pmixp_info_stepid(), SIGKILL, 0);
 	/* unlock the structure */
 	slurm_mutex_unlock(&coll->lock);
 
@@ -1292,7 +1292,7 @@ int pmixp_coll_tree_parent(pmixp_coll_t *coll, uint32_t peerid, uint32_t seq,
 	_reset_coll(coll);
 error2:
 	slurm_kill_job_step(pmixp_info_jobid(),
-			    pmixp_info_stepid(), SIGKILL);
+			    pmixp_info_stepid(), SIGKILL, 0);
 	slurm_mutex_unlock(&coll->lock);
 
 	return SLURM_ERROR;

diff --git a/src/plugins/mpi/pmix/pmixp_dconn.h b/src/plugins/mpi/pmix/pmixp_dconn.h
@@ -247,7 +247,7 @@ static inline int pmixp_dconn_connect(
 		xfree(nodename);
 		pmixp_debug_hang(0); /* enable hang to debug this! */
 		slurm_kill_job_step(pmixp_info_jobid(),
-				    pmixp_info_stepid(), SIGKILL);
+				    pmixp_info_stepid(), SIGKILL, 0);
 	}
 	return rc;
 }

diff --git a/src/plugins/mpi/pmix/pmixp_debug.h b/src/plugins/mpi/pmix/pmixp_debug.h
@@ -63,7 +63,7 @@
 #define PMIXP_ABORT(format, args...) {				\
 	PMIXP_ERROR(format, ##args);				\
 	slurm_kill_job_step(pmixp_info_jobid(),			\
-			    pmixp_info_stepid(), SIGKILL);	\
+			    pmixp_info_stepid(), SIGKILL, 0);	\
 }
 
 #define PMIXP_ERROR_NO(err, format, args...) {			\

diff --git a/src/plugins/mpi/pmix/pmixp_server.c b/src/plugins/mpi/pmix/pmixp_server.c
@@ -833,7 +833,7 @@ static void _process_server_request(pmixp_base_hdr_t *hdr, buf_t *buf)
 				    hdr->seq, nodename, hdr->nodeid, coll->seq);
 			pmixp_debug_hang(0); /* enable hang to debug this! */
 			slurm_kill_job_step(pmixp_info_jobid(),
-					    pmixp_info_stepid(), SIGKILL);
+					    pmixp_info_stepid(), SIGKILL, 0);
 			xfree(nodename);
 			break;
 		} else if (PMIXP_COLL_REQ_SKIP == rc) {

diff --git a/src/scancel/scancel.c b/src/scancel/scancel.c
@@ -854,7 +854,7 @@ _cancel_step_id (void *ci)
 		START_TIMER;
 		if ((!sig_set) || opt.ctld)
 			error_code = slurm_kill_job_step(job_id, step_id,
-							 cancel_info->sig);
+							 cancel_info->sig, 0);
 		else if (cancel_info->sig == SIGKILL)
 			error_code = slurm_terminate_job_step(job_id, step_id);
 		else

diff --git a/src/srun/launch.c b/src/srun/launch.c
@@ -100,7 +100,8 @@ static int _step_signal(int signal)
 	while ((my_srun_job = (srun_job_t *) list_next(iter))) {
 		info("Terminating %ps", &my_srun_job->step_id);
 		rc2 = slurm_kill_job_step(my_srun_job->step_id.job_id,
-					  my_srun_job->step_id.step_id, signal);
+					  my_srun_job->step_id.step_id, signal,
+					  0);
 		if (rc2)
 			rc = rc2;
 	}
@@ -269,7 +270,7 @@ _handle_openmpi_port_error(const char *tasks, const char *hosts,
 	      hosts, tasks, msg);
 
 	info("Terminating job step %ps", &step_id);
-	slurm_kill_job_step(step_id.job_id, step_id.step_id, SIGKILL);
+	slurm_kill_job_step(step_id.job_id, step_id.step_id, SIGKILL, 0);
 }
 
 static char *_mpir_get_host_name(char *node_name)

diff --git a/src/srun/srun_job.c b/src/srun/srun_job.c
@@ -1610,7 +1610,8 @@ job_force_termination(srun_job_t *job)
 		}
 		if (kill_sent == 1) {
 			/* Try sending SIGKILL through slurmctld */
-			slurm_kill_job_step(job->step_id.job_id, job->step_id.step_id, SIGKILL);
+			slurm_kill_job_step(job->step_id.job_id,
+					    job->step_id.step_id, SIGKILL, 0);
 		}
 	}
 	kill_sent++;
@@ -2294,13 +2295,14 @@ static int _shepherd_spawn(srun_job_t *job, List srun_job_list, bool got_alloc)
 		job_iter  = list_iterator_create(srun_job_list);
 		while ((job = list_next(job_iter))) {
 			(void) slurm_kill_job_step(job->step_id.job_id, job->step_id.step_id,
-						   SIGKILL);
+						   SIGKILL, 0);
 			if (got_alloc)
 				slurm_complete_job(job->step_id.job_id, NO_VAL);
 		}
 		list_iterator_destroy(job_iter);
 	} else {
-		(void) slurm_kill_job_step(job->step_id.job_id, job->step_id.step_id, SIGKILL);
+		(void) slurm_kill_job_step(job->step_id.job_id,
+					   job->step_id.step_id, SIGKILL, 0);
 		if (got_alloc)
 			slurm_complete_job(job->step_id.job_id, NO_VAL);
 	}

diff --git a/src/sview/job_info.c b/src/sview/job_info.c
@@ -625,7 +625,7 @@ static int _cancel_step_id(uint32_t job_id, uint32_t step_id,
 	for (i = 0; i < MAX_CANCEL_RETRY; i++) {
 		/* NOTE: RPC always sent to slurmctld rather than directly
 		 * to slurmd daemons */
-		error_code = slurm_kill_job_step(job_id, step_id, signal);
+		error_code = slurm_kill_job_step(job_id, step_id, signal, 0);
 
 		if (error_code == 0
 		    || (errno != ESLURM_TRANSITION_STATE_NO_UPDATE