diff --git a/ci/cscs_daint_spack_psmp.yaml b/ci/cscs_daint_spack_psmp.yaml index e1e3cf15ef..824bfb9465 100644 --- a/ci/cscs_daint_spack_psmp.yaml +++ b/ci/cscs_daint_spack_psmp.yaml @@ -19,7 +19,7 @@ variables: build deps daint: extends: .container-builder-cscs-gh200 stage: deps - timeout: 2h + timeout: 6h before_script: - DEPS_IMAGE_TAG=$(cat make_cp2k.sh ci/cscs_daint_${TAG}.yaml ci/docker/build_deps_spack.Dockerfile $(find tools/spack -type f | sort) | sha256sum | head -c 16) - export PERSIST_IMAGE_NAME=${CSCS_REGISTRY_PATH}/base/${REPOSITORY}:${TAG}_${DEPS_IMAGE_TAG} @@ -48,7 +48,7 @@ build cp2k daint: extends: .container-builder-cscs-gh200 stage: cp2k needs: ["build deps daint"] - timeout: 1h + timeout: 6h before_script: - echo ${DOCKERHUB_TOKEN} | podman login docker.io --username ${DOCKERHUB_USERNAME} --password-stdin after_script: @@ -74,7 +74,7 @@ test cp2k daint: extends: .baremetal-runner-daint-gh200 stage: test needs: ["build cp2k daint"] - timeout: 1h + timeout: 6h script: - podman run --shm-size=1g -v ${PWD}:/mnt ${DOCKERHUB_IMAGE_NAME} mpiexec -bind-to socket -genv OMP_NUM_THREADS=${SLURM_CPUS_PER_TASK} -n ${SLURM_NTASKS} cp2k /opt/cp2k/benchmarks/CI/H2O-32_md.inp - podman run --shm-size=1g -v ${PWD}:/mnt ${DOCKERHUB_IMAGE_NAME} mpiexec -bind-to socket -genv OMP_NUM_THREADS=${SLURM_CPUS_PER_TASK} -n ${SLURM_NTASKS} cp2k /opt/cp2k/benchmarks/CI/H2O-32_md_elpa.inp @@ -85,14 +85,14 @@ test cp2k daint: SLURM_JOB_NUM_NODES: 1 SLURM_NTASKS: 8 SLURM_NTASKS_PER_NODE: 8 - SLURM_TIMELIMIT: 10 + SLURM_TIMELIMIT: 15 USE_MPI: YES regression test cp2k daint: extends: .baremetal-runner-daint-gh200 stage: test needs: ["build cp2k daint","test cp2k daint"] - timeout: 1h + timeout: 6h script: - git --no-pager log -1 --pretty="%nCommitSHA:%x20%H%nCommitTime:%x20%ci%nCommitAuthor:%x20%an%nCommitSubject:%x20%s%n" - podman run --shm-size=1g -v ${PWD}:/mnt ${DOCKERHUB_IMAGE_NAME} /opt/cp2k/tests/do_regtest.py --cp2kdatadir /opt/cp2k/install/share/cp2k/data --maxtasks ${SLURM_NTASKS} --mpiexec "mpiexec -bind-to none" --keepalive /opt/cp2k/install/bin ${VERSION} 2>&1 || true @@ -101,14 +101,14 @@ regression test cp2k daint: SLURM_HINT: nomultithread SLURM_JOB_NUM_NODES: 1 SLURM_NTASKS: 288 - SLURM_TIMELIMIT: 40 + SLURM_TIMELIMIT: 60 USE_MPI: YES benchmark cp2k 1 daint node: extends: .container-runner-daint-gh200 stage: benchmark needs: ["build cp2k daint","test cp2k daint"] - timeout: 1h + timeout: 6h image: ${CSCS_IMAGE_NAME} before_script: - export CP2K_DATA_DIR=/opt/cp2k/install/share/cp2k/data @@ -131,7 +131,7 @@ benchmark cp2k 1 daint node elpa: extends: .container-runner-daint-gh200 stage: benchmark needs: ["build cp2k daint","test cp2k daint","benchmark cp2k 1 daint node"] - timeout: 1h + timeout: 6h image: ${CSCS_IMAGE_NAME} before_script: - export CP2K_DATA_DIR=/opt/cp2k/install/share/cp2k/data diff --git a/ci/cscs_daint_spack_psmp_h100.yaml b/ci/cscs_daint_spack_psmp_h100.yaml index 52c27172ca..9e12176800 100644 --- a/ci/cscs_daint_spack_psmp_h100.yaml +++ b/ci/cscs_daint_spack_psmp_h100.yaml @@ -19,7 +19,7 @@ variables: build deps daint: extends: .container-builder-cscs-gh200 stage: deps - timeout: 2h + timeout: 6h before_script: - DEPS_IMAGE_TAG=$(cat make_cp2k.sh ci/cscs_daint_${TAG}.yaml ci/docker/build_deps_spack.Dockerfile $(find tools/spack -type f | sort) | sha256sum | head -c 16) - export PERSIST_IMAGE_NAME=${CSCS_REGISTRY_PATH}/base/${REPOSITORY}:${TAG}_${DEPS_IMAGE_TAG} @@ -48,7 +48,7 @@ build cp2k daint: extends: .container-builder-cscs-gh200 stage: cp2k needs: ["build deps daint"] - timeout: 1h + timeout: 6h before_script: - echo ${DOCKERHUB_TOKEN} | podman login docker.io --username ${DOCKERHUB_USERNAME} --password-stdin after_script: @@ -74,7 +74,7 @@ test cp2k daint: extends: .baremetal-runner-daint-gh200 stage: test needs: ["build cp2k daint"] - timeout: 1h + timeout: 6h script: - podman run --device nvidia.com/gpu=0 --shm-size=1g -v ${PWD}:/mnt ${DOCKERHUB_IMAGE_NAME} mpiexec -bind-to socket -genv OMP_NUM_THREADS=${SLURM_CPUS_PER_TASK} -n ${SLURM_NTASKS} cp2k /opt/cp2k/benchmarks/CI/H2O-32_md.inp - podman run --device nvidia.com/gpu=0 --shm-size=1g -v ${PWD}:/mnt ${DOCKERHUB_IMAGE_NAME} mpiexec -bind-to socket -genv OMP_NUM_THREADS=${SLURM_CPUS_PER_TASK} -n ${SLURM_NTASKS} cp2k /opt/cp2k/benchmarks/CI/H2O-32_md_elpa.inp @@ -85,14 +85,14 @@ test cp2k daint: SLURM_JOB_NUM_NODES: 1 SLURM_NTASKS: 8 SLURM_NTASKS_PER_NODE: 8 - SLURM_TIMELIMIT: 10 + SLURM_TIMELIMIT: 15 USE_MPI: YES regression test cp2k daint: extends: .baremetal-runner-daint-gh200 stage: test needs: ["build cp2k daint"] - timeout: 1h + timeout: 6h script: - git --no-pager log -1 --pretty="%nCommitSHA:%x20%H%nCommitTime:%x20%ci%nCommitAuthor:%x20%an%nCommitSubject:%x20%s%n" - podman run --device nvidia.com/gpu=all --shm-size=1g -v ${PWD}:/mnt ${DOCKERHUB_IMAGE_NAME} /opt/cp2k/tests/do_regtest.py --cp2kdatadir /opt/cp2k/install/share/cp2k/data --maxtasks ${SLURM_NTASKS} --mpiexec "mpiexec -bind-to none" --keepalive /opt/cp2k/install/bin ${VERSION} 2>&1 || true @@ -101,14 +101,14 @@ regression test cp2k daint: SLURM_HINT: nomultithread SLURM_JOB_NUM_NODES: 1 SLURM_NTASKS: 224 - SLURM_TIMELIMIT: 40 + SLURM_TIMELIMIT: 60 USE_MPI: YES benchmark cp2k 1 daint node: extends: .container-runner-daint-gh200 stage: benchmark needs: ["build cp2k daint","test cp2k daint"] - timeout: 1h + timeout: 6h image: ${CSCS_IMAGE_NAME} before_script: - export CP2K_DATA_DIR=/opt/cp2k/install/share/cp2k/data @@ -131,7 +131,7 @@ benchmark cp2k 1 daint node elpa: extends: .container-runner-daint-gh200 stage: benchmark needs: ["build cp2k daint","test cp2k daint","benchmark cp2k 1 daint node"] - timeout: 1h + timeout: 6h image: ${CSCS_IMAGE_NAME} before_script: - export CP2K_DATA_DIR=/opt/cp2k/install/share/cp2k/data diff --git a/ci/cscs_eiger_spack_psmp.yaml b/ci/cscs_eiger_spack_psmp.yaml index 7bd7ef7b5d..1b67cbf3ba 100644 --- a/ci/cscs_eiger_spack_psmp.yaml +++ b/ci/cscs_eiger_spack_psmp.yaml @@ -19,7 +19,7 @@ variables: build deps eiger: extends: .container-builder-cscs-zen2 stage: deps - timeout: 2h + timeout: 6h before_script: - DEPS_IMAGE_TAG=$(cat make_cp2k.sh ci/cscs_eiger_${TAG}.yaml ci/docker/build_deps_spack.Dockerfile $(find tools/spack -type f | sort) | sha256sum | head -c 16) - export PERSIST_IMAGE_NAME=${CSCS_REGISTRY_PATH}/base/${REPOSITORY}:${TAG}_${DEPS_IMAGE_TAG} @@ -48,7 +48,7 @@ build cp2k eiger: extends: .container-builder-cscs-zen2 stage: cp2k needs: ["build deps eiger"] - timeout: 1h + timeout: 6h before_script: - echo ${DOCKERHUB_TOKEN} | podman login docker.io --username ${DOCKERHUB_USERNAME} --password-stdin after_script: @@ -74,7 +74,7 @@ test cp2k eiger: extends: .baremetal-runner-eiger-zen2 stage: test needs: ["build cp2k eiger"] - timeout: 1h + timeout: 6h script: - podman run --shm-size=1g -v ${PWD}:/mnt ${DOCKERHUB_IMAGE_NAME} mpiexec -genv OMP_NUM_THREADS=${SLURM_CPUS_PER_TASK} -n ${SLURM_NTASKS} cp2k /opt/cp2k/benchmarks/CI/H2O-32_md.inp - podman run --shm-size=1g -v ${PWD}:/mnt ${DOCKERHUB_IMAGE_NAME} mpiexec -genv OMP_NUM_THREADS=${SLURM_CPUS_PER_TASK} -n ${SLURM_NTASKS} cp2k /opt/cp2k/benchmarks/CI/H2O-32_md_elpa.inp @@ -86,14 +86,14 @@ test cp2k eiger: SLURM_JOB_NUM_NODES: 1 SLURM_NTASKS: 8 SLURM_NTASKS_PER_NODE: 8 - SLURM_TIMELIMIT: 10 + SLURM_TIMELIMIT: 15 USE_MPI: YES regression test cp2k eiger: extends: .baremetal-runner-eiger-zen2 stage: test needs: ["build cp2k eiger","test cp2k eiger"] - timeout: 1h + timeout: 6h script: - git --no-pager log -1 --pretty="%nCommitSHA:%x20%H%nCommitTime:%x20%ci%nCommitAuthor:%x20%an%nCommitSubject:%x20%s%n" - podman run --shm-size=1g -v ${PWD}:/mnt ${DOCKERHUB_IMAGE_NAME} /opt/cp2k/tests/do_regtest.py --cp2kdatadir /opt/cp2k/install/share/cp2k/data --maxtasks ${SLURM_NTASKS} --mpiexec "mpiexec -bind-to none" --keepalive /opt/cp2k/install/bin ${VERSION} 2>&1 || true @@ -103,14 +103,14 @@ regression test cp2k eiger: SLURM_HINT: nomultithread SLURM_JOB_NUM_NODES: 1 SLURM_NTASKS: 128 - SLURM_TIMELIMIT: 40 + SLURM_TIMELIMIT: 60 USE_MPI: YES benchmark cp2k 1 eiger node: extends: .container-runner-eiger-zen2 stage: benchmark needs: ["build cp2k eiger","test cp2k eiger","regression test cp2k eiger"] - timeout: 1h + timeout: 6h image: ${CSCS_IMAGE_NAME} before_script: - export CP2K_DATA_DIR=/opt/cp2k/install/share/cp2k/data @@ -132,7 +132,7 @@ benchmark cp2k 1 eiger node elpa: extends: .container-runner-eiger-zen2 stage: benchmark needs: ["build cp2k eiger","test cp2k eiger","regression test cp2k eiger","benchmark cp2k 1 eiger node"] - timeout: 1h + timeout: 6h image: ${CSCS_IMAGE_NAME} before_script: - export CP2K_DATA_DIR=/opt/cp2k/install/share/cp2k/data @@ -154,7 +154,7 @@ benchmark cp2k 2 eiger nodes: extends: .container-runner-eiger-zen2 stage: benchmark needs: ["build cp2k eiger","test cp2k eiger","regression test cp2k eiger"] - timeout: 1h + timeout: 6h image: ${CSCS_IMAGE_NAME} before_script: - export CP2K_DATA_DIR=/opt/cp2k/install/share/cp2k/data @@ -177,7 +177,7 @@ benchmark cp2k 2 eiger nodes elpa: extends: .container-runner-eiger-zen2 stage: benchmark needs: ["build cp2k eiger","test cp2k eiger","regression test cp2k eiger","benchmark cp2k 2 eiger nodes"] - timeout: 1h + timeout: 6h image: ${CSCS_IMAGE_NAME} before_script: - export CP2K_DATA_DIR=/opt/cp2k/install/share/cp2k/data diff --git a/ci/cscs_eiger_spack_ssmp.yaml b/ci/cscs_eiger_spack_ssmp.yaml index 75aa1e3ac5..85f417f9c3 100644 --- a/ci/cscs_eiger_spack_ssmp.yaml +++ b/ci/cscs_eiger_spack_ssmp.yaml @@ -18,7 +18,7 @@ variables: build deps eiger: extends: .container-builder-cscs-zen2 stage: deps - timeout: 2h + timeout: 6h before_script: - DEPS_IMAGE_TAG=$(cat make_cp2k.sh ci/cscs_eiger_${TAG}.yaml ci/docker/build_deps_spack.Dockerfile $(find tools/spack -type f | sort) | sha256sum | head -c 16) - export PERSIST_IMAGE_NAME=${CSCS_REGISTRY_PATH}/base/${REPOSITORY}:${TAG}_${DEPS_IMAGE_TAG} @@ -47,7 +47,7 @@ build cp2k eiger: extends: .container-builder-cscs-zen2 stage: cp2k needs: ["build deps eiger"] - timeout: 1h + timeout: 6h before_script: - echo ${DOCKERHUB_TOKEN} | podman login docker.io --username ${DOCKERHUB_USERNAME} --password-stdin after_script: @@ -73,7 +73,7 @@ test cp2k eiger: extends: .baremetal-runner-eiger-zen2 stage: test needs: ["build cp2k eiger"] - timeout: 1h + timeout: 6h script: - podman run --env OMP_NUM_THREADS=${SLURM_CPUS_PER_TASK} --shm-size=1g -v ${PWD}:/mnt ${DOCKERHUB_IMAGE_NAME} cp2k /opt/cp2k/benchmarks/CI/H2O-32_md.inp variables: @@ -84,13 +84,13 @@ test cp2k eiger: SLURM_JOB_NUM_NODES: 1 SLURM_NTASKS: 1 SLURM_NTASKS_PER_NODE: 1 - SLURM_TIMELIMIT: 10 + SLURM_TIMELIMIT: 15 regression test cp2k eiger: extends: .baremetal-runner-eiger-zen2 stage: test needs: ["build cp2k eiger","test cp2k eiger"] - timeout: 1h + timeout: 6h script: - git --no-pager log -1 --pretty="%nCommitSHA:%x20%H%nCommitTime:%x20%ci%nCommitAuthor:%x20%an%nCommitSubject:%x20%s%n" - podman run --shm-size=1g -v ${PWD}:/mnt ${DOCKERHUB_IMAGE_NAME} /opt/cp2k/tests/do_regtest.py --cp2kdatadir /opt/cp2k/install/share/cp2k/data --maxtasks ${SLURM_NTASKS} --mpiexec "mpiexec -bind-to none" --keepalive /opt/cp2k/install/bin ${VERSION} 2>&1 || true @@ -100,4 +100,4 @@ regression test cp2k eiger: SLURM_HINT: nomultithread SLURM_JOB_NUM_NODES: 1 SLURM_NTASKS: 128 - SLURM_TIMELIMIT: 40 + SLURM_TIMELIMIT: 60