Use a consistent container name prefix with container_scope=job

This fixes the common use-case of attaching to a running container
that did not use --container-name and thus is using the default name
of "$SLURM_JOB_ID.$SLURM_STEP_ID".

Signed-off-by: Felix Abecassis <fabecassis@nvidia.com>

Author: flx42

pyxis_slurmd.c

@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2019-2020, NVIDIA CORPORATION. All rights reserved.
+ * Copyright (c) 2019-2023, NVIDIA CORPORATION. All rights reserved.
  */
 
 #include <errno.h>
@@ -54,7 +54,7 @@ static int pyxis_container_cleanup(uid_t uid, gid_t gid, uint32_t jobid)
 	int ret;
 	FILE *fp = NULL;
 	char *name = NULL;
-	uint32_t id, stepid;
+	uint32_t id;
 	int n;
 	int rv = -1;
 
@@ -67,8 +67,7 @@ static int pyxis_container_cleanup(uid_t uid, gid_t gid, uint32_t jobid)
 
 	while ((name = get_line_from_file(fp)) != NULL) {
 		/* Remove named and unnamed pyxis containers for this job */
-		if (sscanf(name, "pyxis_%u.%u%n", &id, &stepid, &n) == 2 ||
-		    sscanf(name, "pyxis_%u_%*s%n", &id, &n) == 1) {
+		if (sscanf(name, "pyxis_%u_%*s%n", &id, &n) == 1) {
 			if (strlen(name) != n || id != jobid)
 				continue;
 

pyxis_slurmstepd.c

@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2019-2022, NVIDIA CORPORATION. All rights reserved.
+ * Copyright (c) 2019-2023, NVIDIA CORPORATION. All rights reserved.
  */
 
 #include <linux/limits.h>
@@ -1001,7 +1001,10 @@ int slurm_spank_user_init(spank_t sp, int ac, char **av)
 		context.container.name = container_name;
 		container_name = NULL;
 	} else {
-		ret = xasprintf(&context.container.name, "pyxis_%u.%u", context.job.jobid, context.job.stepid);
+		if (context.config.container_scope == SCOPE_JOB)
+			ret = xasprintf(&context.container.name, "pyxis_%u_%u.%u", context.job.jobid, context.job.jobid, context.job.stepid);
+		else
+			ret = xasprintf(&context.container.name, "pyxis_%u.%u", context.job.jobid, context.job.stepid);
 		if (ret < 0)
 			goto fail;
 		context.container.temporary_rootfs = true;

tests/exec.bats

@@ -30,6 +30,19 @@ function teardown() {
     run_srun --overlap --container-name=exec-test findmnt /mymnt
 }
 
+@test "attach to running unnamed container" {
+    run_srun sh -c 'echo $SLURM_JOB_ID'
+    job_id="${lines[-1]}"
+    run_srun sh -c 'echo $SLURM_STEP_ID'
+    step_id="${lines[-1]}"
+    container_name="${job_id}.$((step_id + 1))"
+
+    run_srun --container-image=ubuntu:22.04 sh -c 'sleep 10s' &
+
+    sleep 3s
+    run_srun --overlap --container-name="${container_name}" true
+}
+
 @test "attach to running container after directory change" {
     run_srun --container-image=ubuntu:20.04 --container-name=exec-test bash -c "cd /var && sleep 30s" &