Merge pull request #32 from rapidsai/feat/check_nightly_ci

Add an action to check whether nightlies have succeeded recently

.gitignore

@@ -0,0 +1 @@
+*.swp

.pre-commit-config.yaml

@@ -13,3 +13,9 @@ repos:
     rev: v1.7.4
     hooks:
       - id: actionlint-docker
+  - repo: https://github.com/astral-sh/ruff-pre-commit
+    rev: v0.8.3
+    hooks:
+      - id: ruff
+        args: ["--fix"]
+      - id: ruff-format

README.md

@@ -81,4 +81,4 @@ jobs:
         id: telemetry-setup
         # DO NOT change this in PRs
         uses: rapidsai/shared-actions/dispatch-script@main
-```
+```

check_nightly_success/check-nightly-success/action.yaml

@@ -0,0 +1,36 @@
+name: check-nightly-success
+description: Check if the nightlies have succeeded recently.
+inputs:
+  repo:
+    description: "The repository to check"
+    required: true
+    type: string
+  repo_owner:
+    description: "The org that owns the repo (default: rapidsai)"
+    required: false
+    default: "rapidsai"
+    type: string
+  workflow_id:
+    description: "The workflow whose runs to check"
+    required: false
+    default: "test.yaml"
+    type: string
+  max_days_without_success:
+    description: "The number of consecutive days that may go by without a successful CI run"
+    required: false
+    default: 7
+    type: integer
+
+runs:
+  using: composite
+  steps:
+    - name: Run the Python script
+      shell: bash
+      env:
+        REPO: ${{ inputs.repo }}
+        REPO_OWNER: ${{ inputs.repo_owner }}
+        WORKFLOW_ID: ${{ inputs.workflow_id }}
+        MAX_DAYS_WITHOUT_SUCCESS: ${{ inputs.max_days_without_success }}
+      run: |
+          python -m pip install requests
+          python shared-actions/check_nightly_success/check-nightly-success/check.py ${REPO} --repo-owner ${REPO_OWNER} --workflow-id ${WORKFLOW_ID} --max-days-without-success ${MAX_DAYS_WITHOUT_SUCCESS}

check_nightly_success/check-nightly-success/check.py

@@ -0,0 +1,97 @@
+"""Check whether a GHA workflow has run successfully in the last N days."""
+# ruff: noqa: INP001
+
+import argparse
+import itertools
+import os
+import re
+import sys
+from datetime import datetime
+
+import requests
+
+# Constants
+GITHUB_TOKEN = os.environ["RAPIDS_GH_TOKEN"]
+GOOD_STATUSES = {"success"}
+
+
+def main(
+    repo: str,
+    repo_owner: str,
+    workflow_id: str,
+    max_days_without_success: int,
+    num_attempts: int = 5,
+) -> bool:
+    """Check whether a GHA workflow has run successfully in the last N days."""
+    headers = {"Authorization": f"token {GITHUB_TOKEN}"}
+    url = f"https://api.github.com/repos/{repo_owner}/{repo}/actions/workflows/{workflow_id}/runs"
+    exceptions = []
+    for _ in range(num_attempts):
+        try:
+            response = requests.get(url, headers=headers, timeout=10)
+            response.raise_for_status()
+            break
+        except requests.RequestException as e:
+            exceptions.append(e)
+    else:
+        sep = "\n\t"
+        msg = (
+            f"Failed to fetch {url} after {num_attempts} attempts with the following "
+            f"errors: {sep}{'{sep}'.join(exceptions)}"
+        )
+        raise RuntimeError(msg)
+
+    runs = response.json()["workflow_runs"]
+    tz = datetime.fromisoformat(runs[0]["run_started_at"]).tzinfo
+    now = datetime.now(tz=tz)
+
+    branch_ok = {}
+    for branch, branch_runs in itertools.groupby(runs, key=lambda r: r["head_branch"]):
+        if not re.match("branch-[0-9]{2}.[0-9]{2}", branch):
+            continue
+
+        branch_ok[branch] = False
+        for run in sorted(branch_runs, key=lambda r: r["run_started_at"], reverse=True):
+            if (
+                now - datetime.fromisoformat(run["run_started_at"])
+            ).days > max_days_without_success:
+                break
+            if run["conclusion"] in GOOD_STATUSES:
+                branch_ok[branch] = True
+                break
+
+    failed_branches = [k for k, v in branch_ok.items() if not v]
+    if failed_branches:
+        print(  # noqa: T201
+            f"Branches with no successful runs of {workflow_id} in the last "
+            f"{max_days_without_success} days: "
+            f"{', '.join(failed_branches)}",
+        )
+    return len(failed_branches) > 0
+
+
+if __name__ == "__main__":
+    parser = argparse.ArgumentParser()
+    parser.add_argument("repo", type=str, help="Repository name")
+    parser.add_argument(
+        "--repo-owner",
+        default="rapidsai",
+        help="Repository organization/owner",
+    )
+    parser.add_argument("--workflow-id", default="test.yaml", help="Workflow ID")
+    parser.add_argument(
+        "--max-days-without-success",
+        type=int,
+        default=7,
+        help="Maximum number of days without a successful run",
+    )
+    args = parser.parse_args()
+
+    sys.exit(
+        main(
+            args.repo,
+            args.repo_owner,
+            args.workflow_id,
+            args.max_days_without_success,
+        ),
+    )

check_nightly_success/dispatch/action.yml

@@ -0,0 +1,39 @@
+name: dispatch-check-nightly-success
+description: Clone shared-actions and dispatch to the check-nightly-success action.
+inputs:
+  repo:
+    description: "The repository to check"
+    required: true
+    type: string
+  repo_owner:
+    description: "The org that owns the repo (default: rapidsai)"
+    required: false
+    default: "rapidsai"
+    type: string
+  workflow_id:
+    description: "The workflow whose runs to check"
+    required: false
+    default: "test.yaml"
+    type: string
+  max_days_without_success:
+    description: "The number of consecutive days that may go by without a successful CI run"
+    required: false
+    default: 7
+    type: integer
+
+runs:
+  using: 'composite'
+  steps:
+    - name: Clone shared-actions repo
+      uses: actions/checkout@v4
+      with:
+        repository: ${{ env.SHARED_ACTIONS_REPO || 'rapidsai/shared-actions' }}
+        ref: ${{ env.SHARED_ACTIONS_REF || 'main' }}
+        path: ./shared-actions
+    - name: Run check-nightly-success
+      uses: ./shared-actions/check_nightly_success/check-nightly-success
+      with:
+        repo: ${{ inputs.repo }}
+        repo_owner: ${{ inputs.repo_owner }}
+        workflow_id: ${{ inputs.workflow_id }}
+        max_days_without_success: ${{ inputs.max_days_without_success }}

pyproject.toml

@@ -0,0 +1,17 @@
+# Copyright (c) 2024, NVIDIA CORPORATION.
+
+[tool.ruff]
+target-version = "py310"
+
+[tool.ruff.lint]
+select = ["ALL"]
+ignore = [
+    # Incompatible with D211
+    "D203",
+    # Incompatible with D213
+    "D213",
+    # Incompatible with ruff-format
+    "COM812",
+    "ISC001",
+]
+fixable = ["ALL"]