deploy/helm/health_checks/tinymax_healthcheck/a3plus.yaml

# Copyright 2024 Google LLC
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
#     http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.

job:
  base_name: "chs-hc"
  # guid: "xckd"  # Can specify a GUID if desired. Otherwise, a random GUID will be generated.
  # check_time: "1590303600"  # Will automatically be set if not given
health_check:
  name: "tinymax"
  image:
    repo: "us-docker.pkg.dev/gce-ai-infra/health-check/tinymax-healthcheck"
    tag: "a3-megagpu-8g_4.4.0"
    pull_policy: "Always"
  test_label:
    name: "aiinfra/tinymax-healthcheck-test"
    value: "true"
  env:
    # A3+ instance type is a3-megagpu-8g
    INSTANCE_TYPE: "a3-megagpu-8g"
    HEALTH_VALIDITY_HOURS: "24"
    DRY_RUN: "true"
    NHOSTS: "1"
    nr: "8"
    # Specific to A3+
    LD_LIBRARY_PATH: /usr/local/nvidia/lib64
    USE_TCPX: "false"
    USE_FASTRAK: "true"
    # Note A3+ (a3-megagpu-8g) has no UNIX_CLIENT_PREFIX
    NCCL_LIB_DIR: /usr/local/nvidia/lib64
    NCCL_FASTRAK_USE_SNAP: "1"
    NCCL_FASTRAK_ENABLE_CONTROL_CHANNEL: "0"
    NCCL_FASTRAK_NUM_FLOWS: "2"
    NCCL_DEBUG: "INFO"
    NCCL_DEBUG_SUBSYS: "INIT,NET"
    JAX_COORDINATOR_ADDRESS: "$(JOBSET_NAME)-$(REPLICATED_JOB_NAME)-0-0.$(JOBSET_NAME)"
    USE_GPUDIRECT: "fastrak"
    GPUS_PER_NODE: "8"
    JAX_COORDINATOR_PORT: "6002"
    NCCL_FASTRAK_CTRL_DEV: "eth0"
    NCCL_FASTRAK_IFNAME: "eth1,eth2,eth3,eth4,eth5,eth6,eth7,eth8"
    NCCL_SOCKET_IFNAME: "eth0"
    NCCL_CROSS_NIC: "0"
    NCCL_ALGO: "Ring,Tree"
    NCCL_PROTO: "Simple"
    NCCL_MIN_NCHANNELS: "4"
    NCCL_DYNAMIC_CHUNK_SIZE: "524288"
    NCCL_P2P_NET_CHUNKSIZE: "524288"
    NCCL_P2P_PCI_CHUNKSIZE: "524288"
    NCCL_P2P_NVL_CHUNKSIZE: "1048576"
    NCCL_BUFFSIZE: "8388608"
    CUDA_VISIBLE_DEVICES: "0,1,2,3,4,5,6,7"
    NCCL_NET_GDR_LEVEL: "PIX"
    NCCL_FASTRAK_ENABLE_HOTPATH_LOGGING: "0"
    NCCL_FASTRAK_USE_LLCM: "1"
    TF_CPP_VMODULE: "profile_guided_latency_estimator=10"
    TF_CPP_MIN_LOG_LEVEL: "0"
    TF_CPP_MAX_LOG_LEVEL: "100"
    XLA_PYTHON_CLIENT_MEM_FRACTION: "0.94"
    CUDA_DEVICE_MAX_CONNECTIONS: "1"
    NVTE_FUSED_ATTN: "1"
    NCCL_NVLS_ENABLE: "0"
    NCCL_TUNER_PLUGIN: "libnccl-tuner.so"
    NCCL_TUNER_CONFIG_PATH: "/usr/local/nvidia/lib64/a3plus_tuner_config.textproto"
    NCCL_SHIMNET_GUEST_CONFIG_CHECKER_CONFIG_FILE: "/usr/local/nvidia/lib64/a3plus_guest_config.textproto"
    NCCL_FASTRAK_PLUGIN_ACCEPT_TIMEOUT_MS: "600000"
  volumeMounts:
  - name: nvidia-install-dir-host
    mountPath: /usr/local/nvidia/lib64
  - name: tcpxo-nccl-plugin-volume
    mountPath: /usr/local/tcpxo
  - name: tcpd-socket
    mountPath: /tmp
  - name: workload-terminated-volume
    mountPath: /usr/share/nemo
  - name: shared-memory
    mountPath: /dev/shm
  - name: varlog
    mountPath: /var/log
# Configuration specific to a3plus
volumes:
- name: nvidia-install-dir-host
  hostPath:
    path: /home/kubernetes/bin/nvidia/lib64
- name: tcpd-socket
  hostPath:
    path: /run/tcpxo
- name: tcpxo-nccl-plugin-volume
  emptyDir: {}
- name: workload-terminated-volume
  emptyDir: {}
- name: shared-memory
  emptyDir:
    medium: "Memory"
    sizeLimit: 250Gi
- name: varlog
  emptyDir: {}
- name: fluentbit-key
  secret:
    secretName: fluentbit-key
    optional: true
initContainers:
  nccl_plugin_installer:
    name: "nccl-plugin-installer"
    image: "us-docker.pkg.dev/gce-ai-infra/gpudirect-tcpxo/nccl-plugin-gpudirecttcpx-dev:v1.0.6"
    imagePullPolicy: "Always"
    volumeMounts:
      name: "tcpxo-nccl-plugin-volume"
      mountPath: "/usr/local/tcpxo"
tcpd_daemon:
  image: "us-docker.pkg.dev/gce-ai-infra/gpudirect-tcpxo/tcpgpudmarxd-dev:v1.0.12"
  imagePullPolicy: "Always"
  command:
  - "bash"
  args:
  - "-c"
  - |
    set -ex
    chmod 755 /fts/entrypoint_rxdm_container.sh
    /fts/entrypoint_rxdm_container.sh --num_hops=2 --num_nics=8  --uid= --alsologtostderr &
    cleanup() {
      echo "Received SIGTERM or workload terminated, starting cleanup..."
      pkill -e "^"tcpgpudmarxd || true
      exit 0
    }
    trap cleanup SIGTERM
    while [ ! -e "/usr/share/nemo/workload_terminated" ]; do sleep 1; done
    cleanup
    sleep 30
  volumeMounts:
  - name: nvidia-install-dir-host
    mountPath: /usr/local/nvidia/lib64
  - name: tcpd-socket
    mountPath: /tmp
  - name: workload-terminated-volume
    mountPath: /usr/share/nemo