chore: add Nvidia drivers version matching test and bump env [MD-413] (…

…#9567) (cherry picked from commit d187243)
determined-ai · Jun 26, 2024 · bad22b2 · bad22b2
1 parent 9adbe7c
commit bad22b2
Show file tree

Hide file tree

Showing 47 changed files with 216 additions and 178 deletions.
diff --git a/.circleci/real_config.yml b/.circleci/real_config.yml
@@ -277,7 +277,7 @@ commands:
       - when:
           condition: <<parameters.tf2>>
           steps:
-            - run: docker pull determinedai/pytorch-ngc-dev:8c90e80
+            - run: docker pull determinedai/pytorch-ngc-dev:0e43056
 
   login-docker:
     parameters:
@@ -2401,7 +2401,7 @@ jobs:
 
   test-unit-harness-gpu-tf:
     docker:
-      - image: determinedai/tensorflow-ngc-dev:8c90e80
+      - image: determinedai/tensorflow-ngc-dev:0e43056
     resource_class: determined-ai/container-runner-gpu
     steps:
       - run: mkdir -p ~/.ssh && ssh-keyscan github.com >> ~/.ssh/known_hosts
@@ -2428,7 +2428,7 @@ jobs:
 
   test-unit-harness-pytorch2-gpu:
     docker:
-      - image: determinedai/pytorch-ngc-dev:8c90e80
+      - image: determinedai/pytorch-ngc-dev:0e43056
     resource_class: determined-ai/container-runner-gpu
     steps:
       - run: mkdir -p ~/.ssh && ssh-keyscan github.com >> ~/.ssh/known_hosts
@@ -2455,7 +2455,7 @@ jobs:
 
   test-unit-harness-pytorch2-cpu:
     docker:
-      - image: determinedai/pytorch-ngc-dev:8c90e80
+      - image: determinedai/pytorch-ngc-dev:0e43056
     steps:
       - run: mkdir -p ~/.ssh && ssh-keyscan github.com >> ~/.ssh/known_hosts
       - checkout
@@ -2481,7 +2481,7 @@ jobs:
 
   test-unit-harness-gpu-parallel:
     docker:
-      - image: determinedai/pytorch-ngc-dev:8c90e80
+      - image: determinedai/pytorch-ngc-dev:0e43056
     resource_class: determined-ai/container-runner-multi-gpu
     steps:
       - run: mkdir -p ~/.ssh && ssh-keyscan github.com >> ~/.ssh/known_hosts
@@ -2508,7 +2508,7 @@ jobs:
 
   test-unit-harness-gpu-deepspeed:
     docker:
-      - image: determinedai/pytorch-ngc-dev:8c90e80
+      - image: determinedai/pytorch-ngc-dev:0e43056
     resource_class: determined-ai/container-runner-gpu
     steps:
       - run: mkdir -p ~/.ssh && ssh-keyscan github.com >> ~/.ssh/known_hosts
@@ -3747,7 +3747,7 @@ jobs:
         type: string
         default: "1"
       environment-image:
-        default: determinedai/pytorch-ngc-dev:8c90e80
+        default: determinedai/pytorch-ngc-dev:0e43056
         type: string
       accel-node-taints:
         type: string

diff --git a/.circleci/scripts/pull_image_daemonset.yaml b/.circleci/scripts/pull_image_daemonset.yaml
@@ -13,7 +13,7 @@ spec:
     spec:
       containers:
         - name: pull-docker-daemonset
-          image: determinedai/pytorch-ngc-dev:8c90e80
+          image: determinedai/pytorch-ngc-dev:0e43056
           command: ["/bin/bash"]
           args: ["echo", "test"]
           resources:

diff --git a/.github/CODEOWNERS b/.github/CODEOWNERS
@@ -35,6 +35,7 @@
 /e2e_tests/tests/requirements.txt @determined-ai/model-dev
 /e2e_tests/tests/experiment       @determined-ai/model-dev
 /e2e_tests/tests/nightly          @determined-ai/model-dev
+/e2e_tests/tests/environment      @determined-ai/model-dev
 
 # Backend owns some e2e tests.
 /e2e_tests/tests/cluster  @determined-ai/backend

diff --git a/docs/model-dev-guide/prepare-container/custom-env.rst b/docs/model-dev-guide/prepare-container/custom-env.rst
@@ -114,9 +114,9 @@ Default Images
    -  -  Environment
       -  File Name
    -  -  CPUs
-      -  ``determinedai/pytorch-ngc-dev:8c90e80``
+      -  ``determinedai/pytorch-ngc-dev:0e43056``
    -  -  NVIDIA GPUs
-      -  ``determinedai/pytorch-ngc-dev:8c90e80``
+      -  ``determinedai/pytorch-ngc-dev:0e43056``
    -  -  AMD GPUs
       -  ``determinedai/environments:rocm-5.0-pytorch-1.10-tf-2.7-rocm-0.26.4``
 
@@ -155,7 +155,7 @@ Example Dockerfile that installs custom ``conda``-, ``pip``-, and ``apt``-based
 .. code:: bash
 
    # Determined Image
-   FROM determinedai/tensorflow-ngc-dev:8c90e80
+   FROM determinedai/tensorflow-ngc-dev:0e43056
 
    # Custom Configuration
    RUN apt-get update && \
@@ -216,7 +216,7 @@ environments using :ref:`custom images <custom-docker-images>`:
 .. code:: bash
 
    # Determined Image
-   FROM determinedai/pytorch-ngc-dev:8c90e80
+   FROM determinedai/pytorch-ngc-dev:0e43056
 
    # Create a virtual environment
    RUN conda create -n myenv python=3.8

diff --git a/docs/model-dev-guide/prepare-container/tensorflow-support.rst b/docs/model-dev-guide/prepare-container/tensorflow-support.rst
@@ -20,7 +20,7 @@ Determined supports both TensorFlow 1 and 2. The version of TensorFlow used for
 experiment is controlled by the configured container image. Determined provides prebuilt Docker
 images that include TensorFlow 2+, 1.15, and 2.8, respectively:
 
--  ``determinedai/tensorflow-ngc-dev:8c90e80``
+-  ``determinedai/tensorflow-ngc-dev:0e43056``
 -  ``determinedai/environments:cuda-10.2-pytorch-1.7-tf-1.15-gpu-0.21.2``
 -  ``determinedai/environments:cuda-11.2-tf-2.8-gpu-0.29.1``
 

diff --git a/docs/reference/deploy/helm-config-reference.rst b/docs/reference/deploy/helm-config-reference.rst
@@ -197,13 +197,13 @@
 
    -  ``cpuImage``: Sets the default Docker image for all non-GPU tasks. If a Docker image is
       specified in the :ref:`experiment config <exp-environment-image>` this default is overriden.
-      Defaults to: ``determinedai/pytorch-ngc-dev:8c90e80``.
+      Defaults to: ``determinedai/pytorch-ngc-dev:0e43056``.
 
    -  ``startupHook``: An optional inline script that will be executed as part of task set up.
 
    -  ``gpuImage``: Sets the default Docker image for all GPU tasks. If a Docker image is specified
       in the :ref:`experiment config <exp-environment-image>` this default is overriden. Defaults
-      to: ``determinedai/pytorch-ngc-dev:8c90e80``.
+      to: ``determinedai/pytorch-ngc-dev:0e43056``.
 
    -  ``logPolicies``: Sets log policies for trials. For details, visit :ref:`log_policies
       <experiment-config-min-validation-period>`.

diff --git a/docs/reference/deploy/master-config-reference.rst b/docs/reference/deploy/master-config-reference.rst
@@ -89,12 +89,12 @@ configure different container images for NVIDIA GPU tasks using the ``cuda`` key
 Determined 0.17.6), CPU tasks using ``cpu`` key, and ROCm (AMD GPU) tasks using the ``rocm`` key.
 Default values:
 
--  ``determinedai/pytorch-ngc-dev:8c90e80`` for NVIDIA GPUs and for CPUs.
+-  ``determinedai/pytorch-ngc-dev:0e43056`` for NVIDIA GPUs and for CPUs.
 -  ``determinedai/environments:rocm-5.0-pytorch-1.10-tf-2.7-rocm-0.26.4`` for ROCm.
 
 For TensorFlow users, we provide an image that must be referenced in the experiment configuration:
 
--  ``determinedai/tensorflow-ngc-dev:8c90e80`` for NVIDIA GPUs and for CPUs.
+-  ``determinedai/tensorflow-ngc-dev:0e43056`` for NVIDIA GPUs and for CPUs.
 
 ``environment_variables``
 =========================

diff --git a/docs/reference/experiment-config-reference.rst b/docs/reference/experiment-config-reference.rst
@@ -1333,12 +1333,12 @@ Optional. The Docker image to use when executing the workload. This image must b
 container images for NVIDIA GPU tasks using ``cuda`` key (``gpu`` prior to 0.17.6), CPU tasks using
 ``cpu`` key, and ROCm (AMD GPU) tasks using ``rocm`` key. Default values:
 
--  ``determinedai/pytorch-ngc-dev:8c90e80`` for NVIDIA GPUs and for CPUs.
+-  ``determinedai/pytorch-ngc-dev:0e43056`` for NVIDIA GPUs and for CPUs.
 -  ``determinedai/environments:rocm-5.0-pytorch-1.10-tf-2.7-rocm-0.26.4`` for ROCm.
 
 For TensorFlow users, we provide an image that must be referenced in the experiment configuration:
 
--  ``determinedai/tensorflow-ngc-dev:8c90e80`` for NVIDIA GPUs and for CPUs.
+-  ``determinedai/tensorflow-ngc-dev:0e43056`` for NVIDIA GPUs and for CPUs.
 
 When the cluster is configured with :ref:`resource_manager.type: slurm
 <cluster-configuration-slurm>` and ``container_run_type: singularity``, images are executed using

diff --git a/docs/reference/job-config-reference.rst b/docs/reference/job-config-reference.rst
@@ -45,13 +45,13 @@ The following configuration settings are supported:
       different container images for NVIDIA GPU tasks using ``cuda`` key (``gpu`` prior to 0.17.6),
       CPU tasks using ``cpu`` key, and ROCm (AMD GPU) tasks using ``rocm`` key. Default values:
 
-      -  ``determinedai/pytorch-ngc-dev:8c90e80`` for NVIDIA GPUs and for CPUs.
+      -  ``determinedai/pytorch-ngc-dev:0e43056`` for NVIDIA GPUs and for CPUs.
       -  ``determinedai/environments:rocm-5.0-pytorch-1.10-tf-2.7-rocm-0.26.4`` for ROCm.
 
       For TensorFlow users, we provide an image that must be referenced in the experiment
       configuration:
 
-      -  ``determinedai/tensorflow-ngc-dev:8c90e80`` for NVIDIA GPUs and for CPUs.
+      -  ``determinedai/tensorflow-ngc-dev:0e43056`` for NVIDIA GPUs and for CPUs.
 
    -  ``force_pull_image``: Forcibly pull the image from the Docker registry and bypass the Docker
       cache. Defaults to ``false``.

diff --git a/docs/setup-cluster/deploy-cluster/slurm/singularity.rst b/docs/setup-cluster/deploy-cluster/slurm/singularity.rst
@@ -30,9 +30,9 @@ by default in this version of Determined are described below.
    -  -  Environment
       -  File Name
    -  -  CPUs
-      -  ``determinedai/pytorch-ngc-dev:8c90e80``
+      -  ``determinedai/pytorch-ngc-dev:0e43056``
    -  -  NVIDIA GPUs
-      -  ``determinedai/pytorch-ngc-dev:8c90e80``
+      -  ``determinedai/pytorch-ngc-dev:0e43056``
    -  -  AMD GPUs
       -  ``determinedai/environments:rocm-5.0-pytorch-1.10-tf-2.7-rocm-622d512``
 

diff --git a/docs/setup-cluster/gcp/install-gcp.rst b/docs/setup-cluster/gcp/install-gcp.rst
@@ -406,5 +406,5 @@ This command line will spin up a cluster of up to 2 A100s in the ``us-central1-c
       --compute-agent-instance-type a2-highgpu-1g --gpu-num 1 \
       --gpu-type nvidia-tesla-a100 \
       --region us-central1 --zone us-central1-c \
-      --gpu-env-image determinedai/pytorch-ngc-dev:8c90e80 \
-      --cpu-env-image determinedai/pytorch-ngc-dev:8c90e80
+      --gpu-env-image determinedai/pytorch-ngc-dev:0e43056 \
+      --cpu-env-image determinedai/pytorch-ngc-dev:0e43056
diff --git a/docs/setup-cluster/slurm/singularity.rst b/docs/setup-cluster/slurm/singularity.rst
@@ -30,9 +30,9 @@ by default in this version of Determined are described below.
    -  -  Environment
       -  File Name
    -  -  CPUs
-      -  ``determinedai/pytorch-ngc-dev:8c90e80``
+      -  ``determinedai/pytorch-ngc-dev:0e43056``
    -  -  NVIDIA GPUs
-      -  ``determinedai/pytorch-ngc-dev:8c90e80``
+      -  ``determinedai/pytorch-ngc-dev:0e43056``
    -  -  AMD GPUs
       -  ``determinedai/environments:rocm-5.0-pytorch-1.10-tf-2.7-rocm-622d512``
 

diff --git a/docs/setup-cluster/slurm/slurm-requirements.rst b/docs/setup-cluster/slurm/slurm-requirements.rst
@@ -510,7 +510,7 @@ platform. There may be additional per-user configuration that is required.
 
    .. code:: bash
 
-      image=determinedai/pytorch-ngc-dev:8c90e80
+      image=determinedai/pytorch-ngc-dev:0e43056
       cd /shared/enroot/images
       enroot import docker://$image
       enroot create /shared/enroot/images/${image//[\/:]/\+}.sqsh

diff --git a/e2e_tests/tests/config.py b/e2e_tests/tests/config.py
@@ -14,12 +14,12 @@
 MAX_TRIAL_BUILD_SECS = 90
 
 
-DEFAULT_TF2_CPU_IMAGE = "determinedai/tensorflow-ngc-dev:8c90e80"
-DEFAULT_TF2_GPU_IMAGE = "determinedai/tensorflow-ngc-dev:8c90e80"
-DEFAULT_PT_CPU_IMAGE = "determinedai/pytorch-tensorflow-cpu-dev:8c90e80"
-DEFAULT_PT_GPU_IMAGE = "determinedai/pytorch-tensorflow-cuda-dev:8c90e80"
-DEFAULT_PT2_CPU_IMAGE = "determinedai/pytorch-ngc-dev:8c90e80"
-DEFAULT_PT2_GPU_IMAGE = "determinedai/pytorch-ngc-dev:8c90e80"
+DEFAULT_TF2_CPU_IMAGE = "determinedai/tensorflow-ngc-dev:0e43056"
+DEFAULT_TF2_GPU_IMAGE = "determinedai/tensorflow-ngc-dev:0e43056"
+DEFAULT_PT_CPU_IMAGE = "determinedai/pytorch-tensorflow-cpu-dev:0e43056"
+DEFAULT_PT_GPU_IMAGE = "determinedai/pytorch-tensorflow-cuda-dev:0e43056"
+DEFAULT_PT2_CPU_IMAGE = "determinedai/pytorch-ngc-dev:0e43056"
+DEFAULT_PT2_GPU_IMAGE = "determinedai/pytorch-ngc-dev:0e43056"
 
 TF2_CPU_IMAGE = os.environ.get("TF2_CPU_IMAGE") or DEFAULT_TF2_CPU_IMAGE
 TF2_GPU_IMAGE = os.environ.get("TF2_GPU_IMAGE") or DEFAULT_TF2_GPU_IMAGE

diff --git a/e2e_tests/tests/environment/test_nvidia_driver.py b/e2e_tests/tests/environment/test_nvidia_driver.py
@@ -0,0 +1,37 @@
+import re
+
+import pytest
+
+from tests import api_utils
+from tests import command as cmd
+
+
+@pytest.mark.e2e_gpu
+@pytest.mark.gpu_required
+def test_nvidia_drivers_version_matching() -> None:
+    sess = api_utils.user_session()
+
+    with cmd.interactive_command(sess, ["shell", "start"]) as shell:
+        shell.stdin.write(b"nvidia-smi\n")
+        shell.stdin.write(b"nv-fabricmanager -v\n")
+        # Exit the shell, so we can read output below until EOF instead of timeout
+        shell.stdin.write(b"exit\n")
+        shell.stdin.close()
+
+        lines = ""
+        for line in shell.stdout:
+            lines += line
+
+        m = re.search(r"Driver Version: ([\d.]+)", lines)
+        if not m:
+            pytest.fail(f"Did not find Nvidia driver version in shell output.\n {lines}\n")
+        driver_version = m.group(1)
+
+        m = re.search(r"Fabric Manager version is[\s:]*([\d.]+)", lines)
+        if not m:
+            pytest.fail(f"Did not find fabric manager version in shell output.\n {lines}\n")
+        fabric_manager_version = m.group(1)
+
+        assert (
+            driver_version == fabric_manager_version
+        ), f"nvidia driver {driver_version} doesn't match fabric manager {fabric_manager_version}"
diff --git a/e2e_tests/tests/fixtures/ports-proxy/config.yaml b/e2e_tests/tests/fixtures/ports-proxy/config.yaml
@@ -23,7 +23,7 @@ max_restarts: 0
 
 # Hardcode the image because the new image has a bug. TODO fix this when the image bug is fixed.
 environment:
-  image: determinedai/pytorch-tensorflow-cpu-dev:8c90e80
+  image: determinedai/pytorch-tensorflow-cpu-dev:0e43056
   proxy_ports:
     - proxy_port: 8000
       proxy_tcp: false

diff --git a/examples/computer_vision/iris_tf_keras/adaptive.yaml b/examples/computer_vision/iris_tf_keras/adaptive.yaml
@@ -4,8 +4,8 @@ data:
   test_url: http://download.tensorflow.org/data/iris_test.csv
 environment:
   image:
-    cpu: determinedai/tensorflow-ngc-dev:8c90e80
-    gpu: determinedai/tensorflow-ngc-dev:8c90e80
+    cpu: determinedai/tensorflow-ngc-dev:0e43056
+    gpu: determinedai/tensorflow-ngc-dev:0e43056
 hyperparameters:
   learning_rate:
     type: log

diff --git a/examples/computer_vision/iris_tf_keras/const.yaml b/examples/computer_vision/iris_tf_keras/const.yaml
@@ -4,8 +4,8 @@ data:
   test_url: http://download.tensorflow.org/data/iris_test.csv
 environment:
   image:
-    cpu: determinedai/tensorflow-ngc-dev:8c90e80
-    gpu: determinedai/tensorflow-ngc-dev:8c90e80
+    cpu: determinedai/tensorflow-ngc-dev:0e43056
+    gpu: determinedai/tensorflow-ngc-dev:0e43056
 hyperparameters:
   learning_rate: 1.0e-4
   learning_rate_decay: 1.0e-6

diff --git a/examples/computer_vision/iris_tf_keras/distributed.yaml b/examples/computer_vision/iris_tf_keras/distributed.yaml
@@ -4,8 +4,8 @@ data:
   test_url: http://download.tensorflow.org/data/iris_test.csv
 environment:
   image:
-    cpu: determinedai/tensorflow-ngc-dev:8c90e80
-    gpu: determinedai/tensorflow-ngc-dev:8c90e80
+    cpu: determinedai/tensorflow-ngc-dev:0e43056
+    gpu: determinedai/tensorflow-ngc-dev:0e43056
 hyperparameters:
   learning_rate: 1.0e-4
   learning_rate_decay: 1.0e-6

diff --git a/examples/computer_vision/iris_tf_keras/iris_tf_keras_cancelable.yaml b/examples/computer_vision/iris_tf_keras/iris_tf_keras_cancelable.yaml
@@ -4,8 +4,8 @@ data:
   test_url: http://download.tensorflow.org/data/iris_test.csv
 environment:
   image:
-    cpu: determinedai/tensorflow-ngc-dev:8c90e80
-    gpu: determinedai/tensorflow-ngc-dev:8c90e80
+    cpu: determinedai/tensorflow-ngc-dev:0e43056
+    gpu: determinedai/tensorflow-ngc-dev:0e43056
 resources:
   slots_per_trial: 8
   resource_pool: defq_GPU_cancelable

diff --git a/examples/computer_vision/iris_tf_keras/iris_tf_keras_high_priority.yaml b/examples/computer_vision/iris_tf_keras/iris_tf_keras_high_priority.yaml
@@ -4,8 +4,8 @@ data:
   test_url: http://download.tensorflow.org/data/iris_test.csv
 environment:
   image:
-    cpu: determinedai/tensorflow-ngc-dev:8c90e80
-    gpu: determinedai/tensorflow-ngc-dev:8c90e80
+    cpu: determinedai/tensorflow-ngc-dev:0e43056
+    gpu: determinedai/tensorflow-ngc-dev:0e43056
 resources:
   slots_per_trial: 8
   resource_pool: defq_GPU_hipri

diff --git a/examples/deepspeed_autotune/torchvision/core_api/deepspeed.yaml b/examples/deepspeed_autotune/torchvision/core_api/deepspeed.yaml
@@ -2,7 +2,7 @@ name: torchvision dsat core_api
 max_restarts: 0
 environment:
   image:
-    gpu: determinedai/pytorch-ngc-dev:8c90e80
+    gpu: determinedai/pytorch-ngc-dev:0e43056
 resources:
   slots_per_trial: 2
   shm_size: 4294967296 # 4 GiB.

diff --git a/examples/deepspeed_autotune/torchvision/deepspeed_trial/deepspeed.yaml b/examples/deepspeed_autotune/torchvision/deepspeed_trial/deepspeed.yaml
@@ -2,7 +2,7 @@ name: torchvision dsat deepspeed_trial
 max_restarts: 0
 environment:
   image:
-    gpu: determinedai/pytorch-ngc-dev:8c90e80
+    gpu: determinedai/pytorch-ngc-dev:0e43056
 resources:
   slots_per_trial: 2
   shm_size: 4294967296 # 4 GiB.

diff --git a/examples/hf_trainer_api/hf_image_classification/deepspeed.yaml b/examples/hf_trainer_api/hf_image_classification/deepspeed.yaml
@@ -6,7 +6,7 @@ environment:
     # You may need to modify this to match your network configuration.
     - NCCL_SOCKET_IFNAME=ens,eth,ib
   image:
-    gpu: determinedai/pytorch-ngc-dev:8c90e80
+    gpu: determinedai/pytorch-ngc-dev:0e43056
 resources:
   slots_per_trial: 2
 searcher:

diff --git a/examples/hf_trainer_api/hf_language_modeling/deepspeed.yaml b/examples/hf_trainer_api/hf_language_modeling/deepspeed.yaml
@@ -6,7 +6,7 @@ environment:
     # You may need to modify this to match your network configuration.
     - NCCL_SOCKET_IFNAME=ens,eth,ib
   image:
-    gpu: determinedai/pytorch-ngc-dev:8c90e80
+    gpu: determinedai/pytorch-ngc-dev:0e43056
 resources:
   slots_per_trial: 2
 searcher: