diff --git a/src/tests/ftest/soak/faults.yaml b/src/tests/ftest/soak/faults.yaml index da1f161ecac..940b9ef569a 100644 --- a/src/tests/ftest/soak/faults.yaml +++ b/src/tests/ftest/soak/faults.yaml @@ -5,14 +5,18 @@ hosts: # server_partition: daos_server client_partition: daos_client # client_reservation: daos-test + orterun: allow_run_as_root: true + # This timeout must be longer than the test_timeout param (+15minutes) # 1 hour test timeout: 24H15M + setup: start_servers: true start_agents: true + server_config: name: daos_server control_log_mask: INFO @@ -35,19 +39,26 @@ server_config: env_vars: - FI_UNIVERSE_SIZE=16383 storage: auto + # pool_params - attributes of the pools to create; Currently only create one pool_jobs: size: 75% + properties: rd_fac:0,space_rb:0 + pool_reserved: size: 10% + properties: rd_fac:0,space_rb:0 + container: type: POSIX properties: cksum:crc16,cksum_size:16384,srv_cksum:on + container_reserved: type: POSIX properties: cksum:crc16,cksum_size:16384,srv_cksum:on,rd_fac:1 file_oclass: EC_2P1GX dir_oclass: RP_2GX + faults: fault_list: # - DAOS_DTX_LOST_RPC_REQUEST @@ -82,6 +93,7 @@ soak_faults: test_soak_faults: 24 joblist: - ior_faults + # Commandline parameters # Benchmark and application params # IOR params -a DFS and -a MPIIO @@ -118,11 +130,14 @@ ior_faults: dfuse: mount_dir: "/tmp/soak_dfuse_ior/" disable_caching: true + hdf5_vol: plugin_path: "/usr/lib64/mpich/lib" + events: - "mce: [Hardware Error]: Machine check events logged" - "Package temperature above threshold" + monitor: - "/usr/bin/free -h" - "/usr/bin/vmstat -w" diff --git a/src/tests/ftest/soak/harassers.yaml b/src/tests/ftest/soak/harassers.yaml index bad690909b5..c609fd34ada 100644 --- a/src/tests/ftest/soak/harassers.yaml +++ b/src/tests/ftest/soak/harassers.yaml @@ -5,13 +5,17 @@ hosts: # server_partition: daos_server client_partition: daos_client # client_reservation: daos-test + orterun: allow_run_as_root: true + # This timeout must be longer than the test_timeout param (+15minutes) timeout: 24H30M + setup: start_servers: true start_agents: true + server_config: name: daos_server control_log_mask: INFO @@ -37,25 +41,32 @@ server_config: - FI_UNIVERSE_SIZE=16383 - DD_MASK=rebuild storage: auto + # pool_params - attributes of the pools to create; Currently only create one pool_jobs: size: 90% rebuild_timeout: 600 pool_query_timeout: 120 + properties: rd_fac:0,space_rb:0 + pool_reserved: size: 5% rebuild_timeout: 600 pool_query_timeout: 120 + properties: rd_fac:0,space_rb:0 + container: type: POSIX properties: cksum:crc16,cksum_size:16384,srv_cksum:on daos_timeout: 30 + container_reserved: type: POSIX properties: cksum:crc16,cksum_size:16384,srv_cksum:on,rd_fac:2 file_oclass: EC_2P2GX dir_oclass: RP_3GX daos_timeout: 30 + # test_params - Defines the type of test to run and how long it runs # It also defines how many pools and jobs to create # name: The name of the Avocado testcase @@ -105,6 +116,7 @@ soak_harassers: enable_remote_logging: false enable_scrubber: false enable_rebuild_logmasks: false + # Commandline parameters # Benchmark and application params # IOR params -a DFS and -a MPIIO @@ -141,6 +153,7 @@ ior_harasser: dfuse: mount_dir: "/tmp/soak_dfuse_ior" disable_caching: true + fio_harasser: api: - POSIX @@ -173,6 +186,7 @@ fio_harasser: dfuse: mount_dir: "/tmp/soak_dfuse_fio/" disable_caching: true + mdtest_harasser: # maximum timeout for a single job in test in minutes job_timeout: 30 @@ -205,13 +219,17 @@ mdtest_harasser: dfuse: mount_dir: "/tmp/soak_dfuse_mdtest/" disable_caching: true + hdf5_vol: plugin_path: "/usr/lib64/mpich/lib" + events: - "mce: [Hardware Error]: Machine check events logged" - "Package temperature above threshold" + monitor: - "/usr/bin/free -h" - "/usr/bin/vmstat -w" - "ps -C daos_engine -o %mem,%cpu,cmd" + enable_telemetry: true diff --git a/src/tests/ftest/soak/smoke.yaml b/src/tests/ftest/soak/smoke.yaml index 3257b47e368..39fcb042128 100644 --- a/src/tests/ftest/soak/smoke.yaml +++ b/src/tests/ftest/soak/smoke.yaml @@ -5,16 +5,21 @@ hosts: # server_partition: daos_server client_partition: daos_client # client_reservation: daos-test + orterun: allow_run_as_root: true + mpi_module: mpi/mpich-x86_64 + enable_sudo: true + # This timeout must be longer than the test_timeout param (+15minutes) # 24 Min test timeout: 30M setup: start_servers: true start_agents: true + server_config: name: daos_server control_log_mask: INFO @@ -37,21 +42,30 @@ server_config: env_vars: - FI_UNIVERSE_SIZE=2048 storage: auto + # pool_params - attributes of the pools to create; Currently only create one pool_jobs: size: 90% + properties: rd_fac:0,space_rb:0 + pool_reserved: size: 5% + properties: rd_fac:0,space_rb:0 + pool_racer: size: 5% + properties: rd_fac:0,space_rb:0 + container: type: POSIX properties: cksum:crc16,cksum_size:16384,srv_cksum:on + container_reserved: type: POSIX properties: cksum:crc16,cksum_size:16384,srv_cksum:on,rd_fac:1 file_oclass: EC_2P1GX dir_oclass: RP_2GX + # test_params - Defines the type of test to run and how long it runs # It also defines how many pools and jobs to create # name: The name of the Avocado testcase @@ -84,6 +98,7 @@ smoke: enable_intercept_lib: true enable_remote_logging: false enable_scrubber: true + # Commandline parameters # Benchmark and application params # IOR params -a DFS and -a MPIIO @@ -121,6 +136,7 @@ ior_smoke: disable_caching: true thread_count: 8 cores: '0-7' + fio_smoke: api: - POSIX @@ -156,8 +172,10 @@ fio_smoke: disable_caching: true thread_count: 8 cores: '0-7' + daos_racer: runtime: 120 + vpic_smoke: job_timeout: 10 nodesperjob: @@ -177,6 +195,7 @@ vpic_smoke: cores: '0-7' oclass: - ["EC_2P1GX", "RP_2GX"] + lammps_smoke: job_timeout: 10 nodesperjob: @@ -196,6 +215,7 @@ lammps_smoke: cores: '0-7' oclass: - ["EC_2P1GX", "RP_2GX"] + mdtest_smoke: # maximum timeout for a single job in test in minutes job_timeout: 10 @@ -227,6 +247,7 @@ mdtest_smoke: disable_caching: true thread_count: 8 cores: '0-7' + macsio_smoke: job_timeout: 10 nodesperjob: @@ -251,6 +272,7 @@ macsio_smoke: disable_caching: true thread_count: 8 cores: '0-7' + datamover_smoke: job_timeout: 10 nodesperjob: @@ -284,13 +306,17 @@ datamover_smoke: - 4G test_file: "daos:/testFile" dfs_destroy: false + hdf5_vol: plugin_path: "/usr/lib64/mpich/lib" + events: - "mce: [Hardware Error]: Machine check events logged" - "Package temperature above threshold" + monitor: - "/usr/bin/free -h" - "/usr/bin/vmstat -w" - "ps -C daos_engine -o %mem,%cpu,cmd" + enable_telemetry: true diff --git a/src/tests/ftest/soak/stress.yaml b/src/tests/ftest/soak/stress.yaml index c53d8b02bea..20145d870b5 100644 --- a/src/tests/ftest/soak/stress.yaml +++ b/src/tests/ftest/soak/stress.yaml @@ -5,8 +5,10 @@ hosts: # server_partition: daos_server client_partition: daos_client # client_reservation: daos-test + orterun: allow_run_as_root: true + # This timeout must be longer than the test_timeout param (+15minutes) # 48 hour test timeouts: @@ -14,9 +16,11 @@ timeouts: test_soak_stress_48h: 48H30M test_soak_stress_24h: 24H30M test_soak_stress_12h: 12H30M + setup: start_servers: true start_agents: true + server_config: name: daos_server control_log_mask: INFO @@ -42,21 +46,30 @@ server_config: - FI_UNIVERSE_SIZE=16383 - DAOS_MD_CAP=1024 storage: auto + # pool_params - attributes of the pools to create; Currently only create one pool_jobs: size: 90% + properties: rd_fac:0,space_rb:0 + pool_reserved: size: 5% + properties: rd_fac:0,space_rb:0 + pool_racer: size: 5% + properties: rd_fac:0,space_rb:0 + container: type: POSIX properties: cksum:crc16,cksum_size:16384,srv_cksum:on + container_reserved: type: POSIX properties: cksum:crc16,cksum_size:16384,srv_cksum:on,rd_fac:1 file_oclass: EC_2P1GX dir_oclass: RP_2GX + # test_params - Defines the type of test to run and how long it runs # It also defines how many pools and jobs to create # name: The name of the Avocado testcase @@ -90,6 +103,7 @@ soak_stress: enable_intercept_lib: true enable_remote_logging: false enable_scrubber: true + # Commandline parameters # Benchmark and application params # IOR params -a DFS and -a MPIIO @@ -177,8 +191,10 @@ fio_stress: disable_caching: true thread_count: 8 cores: '0-7' + daos_racer: runtime: 120 + vpic_stress: # Requires opeapi job_timeout: 20 @@ -197,6 +213,7 @@ vpic_stress: disable_caching: true oclass: - ["EC_2P1GX", "RP_2GX"] + lammps_stress: job_timeout: 55 nodesperjob: @@ -216,6 +233,7 @@ lammps_stress: cores: '0-7' oclass: - ["EC_2P1GX", "RP_2GX"] + mdtest_stress: # maximum timeout for a single job in test in minutes job_timeout: 40 @@ -255,6 +273,7 @@ mdtest_stress: disable_caching: true thread_count: 8 cores: '0-7' + macsio_stress: job_timeout: 30 nodesperjob: @@ -286,6 +305,7 @@ macsio_stress: disable_caching: true thread_count: 8 cores: '0-7' + datamover_stress: job_timeout: 10 nodesperjob: @@ -324,13 +344,17 @@ datamover_stress: - 4G test_file: "daos:/testFile" dfs_destroy: false + hdf5_vol: plugin_path: "/usr/lib64/mpich/lib" + events: - "mce: [Hardware Error]: Machine check events logged" - "Package temperature above threshold" + monitor: - "/usr/bin/free -h" - "/usr/bin/vmstat -w" - "ps -C daos_engine -o %mem,%cpu,cmd" + enable_telemetry: true diff --git a/src/tests/ftest/soak/stress_2h.yaml b/src/tests/ftest/soak/stress_2h.yaml index 09789cf2ea6..d2542f63ab0 100644 --- a/src/tests/ftest/soak/stress_2h.yaml +++ b/src/tests/ftest/soak/stress_2h.yaml @@ -5,13 +5,17 @@ hosts: # server_partition: daos_server client_partition: normal client_reservation: daos-test + mpi_module: mpich/gnu-tcp + # This timeout must be longer than the test_timeout param (+15minutes) # 2 hour test timeout: 2H15M + setup: start_servers: false start_agents: false + server_config: name: daos_server control_log_mask: INFO @@ -24,26 +28,35 @@ server_config: storage: auto transport_config: allow_insecure: true + agent_config: transport_config: allow_insecure: true + dmg: transport_config: allow_insecure: true + # pool_params - attributes of the pools to create; Currently only create one pool_jobs: size: 75% + properties: rd_fac:0,space_rb:0 + pool_reserved: size: 10% + properties: rd_fac:0,space_rb:0 + container: type: POSIX properties: cksum:crc16,cksum_size:16384,srv_cksum:on + container_reserved: type: POSIX properties: cksum:crc16,cksum_size:16384,srv_cksum:on,rd_fac:1 file_oclass: SX dir_oclass: SX daos_timeout: 30 + # test_params - Defines the type of test to run and how long it runs # It also defines how many pools and jobs to create # name: The name of the Avocado testcase @@ -66,6 +79,7 @@ soak_stress: - mdtest_stress - vpic_stress - lammps_stress + # Commandline parameters # Benchmark and application params # IOR params -a DFS and -a MPIIO @@ -97,6 +111,7 @@ ior_stress: dfuse: mount_dir: "/tmp/soak_dfuse_ior/" disable_caching: true + mdtest_stress: # maximum timeout for a single job in test in minutes job_timeout: 10 @@ -124,6 +139,7 @@ mdtest_stress: dfuse: mount_dir: "/tmp/soak_dfuse_mdtest/" disable_caching: true + vpic_stress: job_timeout: 20 nodesperjob: @@ -138,6 +154,7 @@ vpic_stress: disable_caching: true oclass: - ["SX","SX"] + lammps_stress: job_timeout: 30 nodesperjob: