DAOS-18586 test: Adding more workload to soak smoke and harasser

src/tests/ftest/soak/harassers.yaml

@@ -10,8 +10,7 @@ orterun:
   allow_run_as_root: true
 
 # This timeout must be longer than the test_timeout param (+15minutes)
-timeout: 24H30M
-
+timeout: 72H30M
 setup:
   start_servers: true
   start_agents: true
@@ -47,14 +46,12 @@ pool_jobs:
   size: 90%
   rebuild_timeout: 600
   pool_query_timeout: 120
-  properties: rd_fac:0,space_rb:0
-
+  properties: "scrub:timed,scrub_freq:120,space_rb:8"
 pool_reserved:
   size: 5%
   rebuild_timeout: 600
   pool_query_timeout: 120
-  properties: rd_fac:0,space_rb:0
-
+  properties: "scrub:timed,scrub_freq:120,space_rb:8"
 container:
   type: POSIX
   properties: cksum:crc16,cksum_size:16384,srv_cksum:on
@@ -84,13 +81,18 @@ soak_harassers:
   # harasser test timeout in hours
   single_test_pool: false
   test_timeout:
-    test_soak_online_harassers: 24
-    test_soak_offline_harassers: 24
+    test_soak_online_harassers: 72
+    test_soak_offline_harassers: 72
   # maximum timeout for a single job in test in minutes
   joblist:
     - ior_harasser
     - fio_harasser
     - mdtest_harasser
+    #- vpic_harasser
+    - lammps_harasser
+    - macsio_harasser
+    - datamover_harasser
+    - e3smio_harasser
   # one harrasser is run during each pass of soak jobs
   harasserlist:
     test_soak_offline_harassers:
@@ -148,8 +150,8 @@ ior_harasser:
     - '4k'
   segment_count: 1
   dfs_oclass:
-    - ["EC_2P2GX", "RP_3GX"]
-    - ["EC_4P2GX", "RP_3GX"]
+    - ["EC_2P2GX"]
+    - ["EC_4P2GX"]
   dfuse:
     mount_dir: "/tmp/soak_dfuse_ior"
     disable_caching: true
@@ -186,7 +188,60 @@ fio_harasser:
   dfuse:
     mount_dir: "/tmp/soak_dfuse_fio/"
     disable_caching: true
-
+vpic_harasser:
+  # Requires opeapi
+  job_timeout: 20
+  nodesperjob:
+    - 1
+  taskspernode:
+    - 32
+  cmdline: "${DAOS_TEST_APP_DIR}/vpic-install/bin/harris.Linux"
+  api:
+    - POSIX
+    - POSIX-LIBIOIL
+    - POSIX-LIBPIL4DFS
+  workdir: "/tmp/soak_dfuse_vpic/"
+  dfuse:
+    mount_dir: "/tmp/soak_dfuse_vpic/"
+    disable_caching: true
+  oclass:
+    - ["EC_2P1GX", "RP_2GX"]
+lammps_harasser:
+  job_timeout: 55
+  nodesperjob:
+    - 8
+  taskspernode:
+    - 32
+  cmdline: "${DAOS_TEST_APP_DIR}/lammps/src/lmp_mpi -i ${DAOS_TEST_APP_DIR}/lammps/bench/in.lj"
+  api:
+    - POSIX
+    - POSIX-LIBIOIL
+    - POSIX-LIBPIL4DFS
+  workdir: "/tmp/soak_dfuse_lammps/"
+  dfuse:
+    mount_dir: "/tmp/soak_dfuse_lammps/"
+    disable_caching: true
+    thread_count: 8
+    cores: '0-7'
+  oclass:
+    - ["EC_2P1GX", "RP_2GX"]
+e3smio_harasser:
+  job_timeout: 20
+  nodesperjob:
+    - 4
+  taskspernode:
+    - 32
+  cmdline: "e3sm_io -r 10 -o can_I_out.h5 -a hdf5 -x canonical -k "
+  api:
+    - POSIX-LIBPIL4DFS
+  workdir: "/tmp/soak_dfuse_e3smio/"
+  dfuse:
+    mount_dir: "/tmp/soak_dfuse_e3smio/"
+    disable_caching: true
+    thread_count: 8
+    cores: '0-7'
+  oclass:
+    - ["EC_2P1GX", "RP_2GX"]
 mdtest_harasser:
   # maximum timeout for a single job in test in minutes
   job_timeout: 30
@@ -219,7 +274,75 @@ mdtest_harasser:
   dfuse:
     mount_dir: "/tmp/soak_dfuse_mdtest/"
     disable_caching: true
-
+macsio_harasser:
+  job_timeout: 30
+  nodesperjob:
+    - 1
+    - 4
+    - 8
+  taskspernode:
+    - 16
+  api:
+    - HDF5
+    - HDF5-VOL
+  interface: hdf5
+  parallel_file_mode: SIF 1
+  filebase: daos
+  units_prefix_system: decimal
+  part_size: 10M
+  avg_num_parts: 2.5
+  num_dumps: 2
+  debug_level: 1
+  oclass:
+    - ["SX","SX"]
+    - ["EC_2P1GX", "RP_2GX"]
+    #  - ["EC_4P1G1", "RP_2GX"]
+    - ["EC_4P2GX", "RP_3GX"]
+    #  - ["EC_8P2G1", "RP_3GX"]
+    #  - ["EC_16P2GX", "RP_3GX"]
+  dfuse:
+    mount_dir: "/tmp/soak_dfuse_macsio/"
+    disable_caching: true
+    thread_count: 8
+    cores: '0-7'
+datamover_harasser:
+  job_timeout: 10
+  nodesperjob:
+    - 8
+  taskspernode:
+    - 16
+  oclass:
+    - ["SX","SX"]
+    - ["EC_2P1GX", "RP_2GX"]
+    #  - ["EC_4P1G1", "RP_2GX"]
+    - ["EC_4P2GX", "RP_3GX"]
+    #  - ["EC_8P2G1", "RP_3GX"]
+    #  - ["EC_16P2GX", "RP_3GX"]
+  dcp:
+    bufsize: 1M
+    chunksize: 1M
+  ior_write:
+    api:
+      - DFS
+    flags: "-w -F -k"
+    signature: "5"
+    transfer_size:
+      - 1M
+    block_size:
+      - 2G
+    test_file: "daos:/testFile"
+    dfs_destroy: false
+  ior_read:
+    api:
+      - DFS
+    flags: "-r -R -F -k"
+    signature: "5"
+    transfer_size:
+      - 1M
+    block_size:
+      - 2G
+    test_file: "daos:/testFile"
+    dfs_destroy: false
 hdf5_vol:
   plugin_path: "/usr/lib64/mpich/lib"
 

src/tests/ftest/soak/smoke.yaml

@@ -46,16 +46,13 @@ server_config:
 # pool_params - attributes of the pools to create; Currently only create one
 pool_jobs:
   size: 90%
-  properties: rd_fac:0,space_rb:0
-
+  properties: "scrub:timed,scrub_freq:120,space_rb:8"
 pool_reserved:
   size: 5%
-  properties: rd_fac:0,space_rb:0
-
+  properties: "scrub:timed,scrub_freq:120,space_rb:8"
 pool_racer:
   size: 5%
-  properties: rd_fac:0,space_rb:0
-
+  properties: "scrub:timed,scrub_freq:120,space_rb:8"
 container:
   type: POSIX
   properties: cksum:crc16,cksum_size:16384,srv_cksum:on
@@ -93,6 +90,7 @@ smoke:
     - lammps_smoke
     - macsio_smoke
     - datamover_smoke
+    - e3smio_smoke
   # num of bytes to write to reserved container
   resv_bytes: 500
   enable_intercept_lib: true
@@ -215,7 +213,23 @@ lammps_smoke:
     cores: '0-7'
   oclass:
     - ["EC_2P1GX", "RP_2GX"]
-
+e3smio_smoke:
+  job_timeout: 20
+  nodesperjob:
+    - 4
+  taskspernode:
+    - 32
+  cmdline: "e3sm_io -r 10 -o can_I_out.h5 -a hdf5 -x canonical -k "
+  api:
+    - POSIX-LIBPIL4DFS
+  workdir: "/tmp/soak_dfuse_e3smio/"
+  dfuse:
+    mount_dir: "/tmp/soak_dfuse_e3smio/"
+    disable_caching: true
+    thread_count: 8
+    cores: '0-7'
+  oclass:
+    - ["EC_2P1GX", "RP_2GX"]
 mdtest_smoke:
   # maximum timeout for a single job in test in minutes
   job_timeout: 10

src/tests/ftest/util/soak_test_base.py

@@ -1,6 +1,6 @@
 """
 (C) Copyright 2019-2024 Intel Corporation.
-(C) Copyright 2025 Hewlett Packard Enterprise Development LP
+(C) Copyright 2025-2026 Hewlett Packard Enterprise Development LP
 
 SPDX-License-Identifier: BSD-2-Clause-Patent
 """
@@ -448,6 +448,8 @@ def job_setup(self, jobs, pool):
                         commands = create_app_cmdline(self, job, pool, ppn, npj)
                     elif "lammps" in job:
                         commands = create_app_cmdline(self, job, pool, ppn, npj)
+                    elif "e3smio" in job:
+                        commands = create_app_cmdline(self, job, pool, ppn, npj)
                     elif "macsio" in job:
                         commands = create_macsio_cmdline(self, job, pool, ppn, npj)
                     elif "datamover" in job:

src/tests/ftest/util/soak_utils.py

@@ -1532,20 +1532,37 @@
             mpirun_cmd.get_params(self)
             env = EnvironmentVariables()
             env["D_LOG_FILE_APPEND_PID"] = "1"
-            if "mpich" in mpi_module:
+            if "mpich" in mpi_module and "e3smio" not in job_spec:
                 # Pass pool and container information to the commands
                 env["DAOS_UNS_PREFIX"] = format_path(pool, self.container[-1])
             if self.enable_il and api == "POSIX-LIBPIL4DFS":
                 env["LD_PRELOAD"] = os.path.join(self.prefix, 'lib64', 'libpil4dfs.so')
             if self.enable_il and api == "POSIX-LIBIOIL":
                 env["LD_PRELOAD"] = os.path.join(self.prefix, 'lib64', 'libioil.so')
+            self.log.info(f"<<APP COMMAND: {app_cmd}")
+            if 'e3sm_io' in app_cmd:
+                env["ROMIO_FSTYPE_FORCE"] = "daos:"
+                env["ROMIO_READ_AGGMETHOD"] = "0"
+                env["ROMIO_HINTS"] = os.path.join(apps_dir, 'romio_hints', 'romio_hints_e3smio')
+                input_file = os.path.join(apps_dir, 'E3SM-IO', 'datasets', 'map_i_case_16p.h5')
+
+                # Copy the file to the dfuse mount directory
+                destination = os.path.join(dfuse.mount_dir.value, os.path.basename(input_file))
+                cp_input_file_cmd = "cp {} {}; ls -l {}".format(
+                    input_file, destination, destination)
+                sbatch_cmds.append(cp_input_file_cmd)
+                self.log.info(f"Copied {input_file} to {destination}")
+
             mpirun_cmd.assign_environment(env, True)
             mpirun_cmd.assign_processes(nodesperjob * ppn)
             mpirun_cmd.ppn.update(ppn)
             mpirun_cmd.hostlist.update("$HOSTLIST")
             if api in ["POSIX", "POSIX-LIBIOIL", "POSIX-LIBPIL4DFS"]:
                 mpirun_cmd.working_dir.update(dfuse.mount_dir.value)
-            cmdline = str(mpirun_cmd)
+            if 'e3sm_io' in app_cmd:
+                cmdline = str(mpirun_cmd) + " " + destination
+            else:
+                cmdline = str(mpirun_cmd)
             sbatch_cmds.append(str(cmdline))
             sbatch_cmds.append("status=$?")
             if api in ["POSIX", "POSIX-LIBIOIL", "POSIX-LIBPIL4DFS"]: