wandb · dmitryduev · Aug 19, 2022 · Aug 16, 2022 · Aug 16, 2022 · Aug 16, 2022
diff --git a/.circleci/config.yml b/.circleci/config.yml
@@ -1,9 +1,9 @@
 version: 2.1
 
 orbs:
-  win: circleci/windows@2.4.0
-  slack: circleci/slack@4.9.3
-  go: circleci/go@1.3.0
+  win: circleci/windows@5.0.0
+  slack: circleci/slack@4.10.1
+  go: circleci/go@1.7.1
   gke: circleci/gcp-gke@1.4.0
 
 parameters:
@@ -79,6 +79,9 @@ parameters:
   manual_nightly_execute_shard_kfp:
     type: boolean
     default: false
+  manual_nightly_execute_shard_standalone_gpu_win:
+    type: boolean
+    default: false
   wandb_server_tag:
     type: string
     default: "master"
@@ -323,7 +326,7 @@ jobs:
         default: 3
       python_version_minor:
         type: integer
-        default: 7
+        default: 9
       toxenv:
         type: string
       parallelism:
@@ -332,31 +335,56 @@ jobs:
       xdist:
         type: integer
         default: 3
+      machine_executor:
+        type: string
+        default: "default"  # "default" or "server-2019-cuda"
+      executor_size:
+        type: string
+        default: "large"  # could only be "medium" for "server-2019-cuda"
+      execute:
+        type: boolean
+        default: true
     executor:
-      name: win/default
-      size: "large"
+      name: win/<< parameters.machine_executor >>
+      size: << parameters.executor_size >>
     parallelism: << parameters.parallelism >>
     steps:
       - checkout
-      - run:
-          name: Install python dependencies
-          command: |
-            pip install tox==<< pipeline.parameters.tox_version >>
-      - run:
-          name: Temporary conda hack
-          shell: bash.exe
-          command: |
-            cp /c/tools/miniconda3/python* /c/tools/miniconda3/lib/venv/scripts/nt/
-      - run:
-          name: Run tests
-          shell: bash.exe
-          command: |
-            echo $GCLOUD_SERVICE_KEY > key.json
-            gcloud auth activate-service-account --key-file=key.json
-            yes | gcloud auth configure-docker
-            CI_PYTEST_PARALLEL=<< parameters.xdist >> CI_PYTEST_SPLIT_ARGS="--splits $CIRCLE_NODE_TOTAL --group $(( $CIRCLE_NODE_INDEX + 1 ))" tox -v -e << parameters.toxenv >>
-          no_output_timeout: 10m
-      - save-test-results
+      - when:
+          condition: << parameters.execute >>
+          steps:
+            - run:
+                name: Install python dependencies
+                shell: bash.exe
+                command: |
+                  pip install tox==<< pipeline.parameters.tox_version >>
+            - run:
+                name: Temporary conda hack
+                shell: bash.exe
+                command: |
+                  cp /c/tools/miniconda3/python* /c/tools/miniconda3/lib/venv/scripts/nt/
+            - when:
+                condition:
+                  equal: [ "server-2019-cuda", << parameters.machine_executor >> ]
+                steps:
+                  - run:
+                      name: Update tox.ini on a GPU machine to install the proper pytorch version
+                      shell: bash.exe
+                      command: |
+                        CUDA_VERSION=`ls "/c/Program Files/NVIDIA GPU Computing Toolkit/CUDA"`
+                        CUDA_VERSION_NO_DOT=`echo "${CUDA_VERSION//./}"`
+                        v=${CUDA_VERSION_NO_DOT:1}
+                        sed -i -e "s/whl\/cpu/whl\/cu$v/g" tox.ini
+            - run:
+                name: Run tests
+                shell: bash.exe
+                command: |
+                  echo $GCLOUD_SERVICE_KEY > key.json
+                  gcloud auth activate-service-account --key-file=key.json
+                  yes | gcloud auth configure-docker
+                  DATE=$(date -u +%Y%m%d) CI_PYTEST_PARALLEL=<< parameters.xdist >> CI_PYTEST_SPLIT_ARGS="--splits $CIRCLE_NODE_TOTAL --group $(( $CIRCLE_NODE_INDEX + 1 ))" tox -v -e << parameters.toxenv >>
+                no_output_timeout: 10m
+            - save-test-results
 
   mac:
     parameters:
@@ -918,6 +946,20 @@ workflows:
           context: slack-secrets
           notify_on_failure: true
       #
+      # standalone GPU tests on Windows
+      #
+      - win:
+          machine_executor: "server-2019-cuda"
+          executor_size: "medium"
+          matrix:
+            parameters:
+              python_version_major: [ 3 ]
+              python_version_minor: [ 9 ]
+          name: "func-s_standalone_gpu-win-py<<matrix.python_version_major>><<matrix.python_version_minor>>"
+          toxenv: "standalone-gpu-py<<matrix.python_version_major>><<matrix.python_version_minor>>"
+          parallelism: 2
+          xdist: 1
+      #
       # standalone tests on gke
       #
       - slack_notify:
@@ -1107,7 +1149,7 @@ workflows:
           matrix:
             parameters:
               python_version_major: [3]
-              python_version_minor: [7]
+              python_version_minor: [9]
           name: "unit-s_base-win-py<<matrix.python_version_major>><<matrix.python_version_minor>>"
           toxenv: "py<<matrix.python_version_major>><<matrix.python_version_minor>>,wincovercircle -- --timeout 300 tests/unit_tests"
       - mac:
@@ -1138,7 +1180,7 @@ workflows:
           matrix:
             parameters:
               python_version_major: [3]
-              python_version_minor: [7]
+              python_version_minor: [9]
           name: "unit-s_nb-win-py<<matrix.python_version_major>><<matrix.python_version_minor>>"
           toxenv: "unit-s_nb-py<<matrix.python_version_major>><<matrix.python_version_minor>>"
           parallelism: 1
@@ -1195,6 +1237,21 @@ workflows:
           context: slack-secrets
           notify_on_failure: << pipeline.parameters.manual_nightly_slack_notify >>
       #
+      # standalone GPU tests on Windows
+      #
+      - win:
+          machine_executor: "server-2019-cuda"
+          executor_size: "medium"
+          execute: << pipeline.parameters.manual_nightly_execute_shard_standalone_gpu_win >>
+          matrix:
+            parameters:
+              python_version_major: [ 3 ]
+              python_version_minor: [ 9 ]
+          name: "func-s_standalone_gpu-win-py<<matrix.python_version_major>><<matrix.python_version_minor>>"
+          toxenv: "standalone-gpu-py<<matrix.python_version_major>><<matrix.python_version_minor>>"
+          parallelism: 2
+          xdist: 1
+      #
       # stanalone tests on gke
       #
       - slack_notify:

diff --git a/tests/standalone_tests/artifact_tests.yea b/tests/standalone_tests/artifact_tests.yea
@@ -5,6 +5,7 @@ tag:
   platforms:
     - linux
     - mac
+    - win
 plugin:
   - wandb
 assert:

diff --git a/tests/standalone_tests/basic.yea b/tests/standalone_tests/basic.yea
@@ -6,6 +6,7 @@ tag:
   platforms:
     - linux
     - mac
+    - win
 plugin:
   - wandb
 assert:

diff --git a/tests/standalone_tests/basic_plots.yea b/tests/standalone_tests/basic_plots.yea
@@ -4,6 +4,7 @@ tag:
   platforms:
     - linux
     - mac
+    - win
 plugin:
   - wandb
 assert:

diff --git a/tests/standalone_tests/mixed_keras.yea b/tests/standalone_tests/mixed_keras.yea
@@ -4,6 +4,7 @@ tag:
   platforms:
     - linux
     - mac
+    - win
 plugin:
   - wandb
 assert:

diff --git a/tests/standalone_tests/tweets.yea b/tests/standalone_tests/tweets.yea
@@ -5,6 +5,7 @@ tag:
   platforms:
     - linux
     - mac
+    - win
 plugin:
   - wandb
 assert:

diff --git a/tools/circleci-tool.py b/tools/circleci-tool.py
@@ -44,6 +44,15 @@
 
 CIRCLECI_API_TOKEN = "CIRCLECI_TOKEN"
 
+NIGHTLY_SHARDS = (
+    "standalone-cpu",
+    "standalone-gpu",
+    "standalone-tpu",
+    "standalone-local",
+    "kfp",
+    "standalone-gpu-win",
+)
+
 platforms_dict = dict(linux="test", lin="test", mac="mac", win="win")
 platforms_short_dict = dict(linux="lin", lin="lin", mac="mac", win="win")
 py_name_dict = dict(
@@ -154,13 +163,7 @@ def trigger(args):
 def trigger_nightly(args):
     url = "https://circleci.com/api/v2/project/gh/wandb/wandb/pipeline"
 
-    default_shards = {
-        "standalone-cpu",
-        "standalone-gpu",
-        "standalone-tpu",
-        "standalone-local",
-        "kfp",
-    }
+    default_shards = set(NIGHTLY_SHARDS)
     shards = {
         f"manual_nightly_execute_shard_{shard.replace('-', '_')}": False
         for shard in default_shards
@@ -320,14 +323,8 @@ def process_args():
     )
     parse_trigger_nightly.add_argument(
         "--shards",
-        default=(
-            "standalone-cpu,"
-            "standalone-gpu,"
-            "standalone-tpu,"
-            "standalone-local,"
-            "kfp"
-        ),
-        help="comma-separated shards (standalone-{cpu,gpu,tpu,local},kfp)",
+        default=",".join(NIGHTLY_SHARDS),
+        help="comma-separated shards (standalone-{cpu,gpu,tpu,local,gpu-win},kfp)",
     )
     parse_trigger_nightly.add_argument(
         "--wait", action="store_true", help="Wait for finish or error"

diff --git a/tox.ini b/tox.ini
@@ -84,7 +84,7 @@ passenv =
 commands =
     s_nb: ipython kernel install --user --name=wandb_python
     mkdir -p test-results
-    python -m pytest {env:CI_PYTEST_SPLIT_ARGS:} -n={env:CI_PYTEST_PARALLEL:{env:WB_UNIT_PARALLEL:4}} --durations=20 --junitxml=test-results/junit.xml --cov-config=.coveragerc --cov --cov-report= --no-cov-on-fail {posargs:tests/unit_tests_old/tests_{env:WB_UNIT_SHARD}/}
+    python -m pytest {env:CI_PYTEST_SPLIT_ARGS:} -n={env:CI_PYTEST_PARALLEL:{env:WB_UNIT_PARALLEL:4}} --durations=20 --junitxml=test-results/junit.xml --cov-config=.coveragerc --cov --cov-report= --no-cov-on-fail --timeout 300 {posargs:tests/unit_tests_old/tests_{env:WB_UNIT_SHARD}/}
 
 [testenv:dev]
 usedevelop = true
@@ -384,11 +384,11 @@ whitelist_externals =
     bash.exe
 commands =
     bash.exe -c 'mkdir -p cover-results'
-    bash.exe -c 'C:/Users/circleci/project/.tox/wincovercircle/Scripts/python -m coverage combine C:/Users/circleci/project/.tox/py37/.coverage*'
-    bash.exe -c 'C:/Users/circleci/project/.tox/wincovercircle/Scripts/python -m coverage xml'
+    bash.exe -c '~/project/.tox/wincovercircle/Scripts/python.exe -m coverage combine ~/project/.tox/py39/.coverage*'
+    bash.exe -c '~/project/.tox/wincovercircle/Scripts/python.exe -m coverage xml'
     bash.exe -c 'cp .coverage coverage.xml cover-results/'
-    bash.exe -c 'C:/Users/circleci/project/.tox/wincovercircle/Scripts/python -m coverage report --ignore-errors --skip-covered --omit "wandb/vendor/*"'
-    bash.exe -c 'C:/Users/circleci/project/.tox/wincovercircle/Scripts/python -m codecov -e TOXENV -F unittest'
+    bash.exe -c '~/project/.tox/wincovercircle/Scripts/python.exe -m coverage report --ignore-errors --skip-covered --omit "wandb/vendor/*"'
+    bash.exe -c '~/project/.tox/wincovercircle/Scripts/python.exe -m codecov -e TOXENV -F unittest'
 
 [testenv:cover]
 skip_install = true

diff --git a/wandb/sdk/internal/meta.py b/wandb/sdk/internal/meta.py
@@ -180,7 +180,7 @@ def _setup_sys(self):
             pynvml.nvmlInit()
             self.data["gpu"] = pynvml.nvmlDeviceGetName(
                 pynvml.nvmlDeviceGetHandleByIndex(0)
-            ).decode("utf8")
+            )
             self.data["gpu_count"] = pynvml.nvmlDeviceGetCount()
         except pynvml.NVMLError:
             pass