fix: extend timeout for jobset_ttr_drain_restart to 90 minutes

.github/workflows/dag-check.yml

@@ -3,7 +3,6 @@ name: DAG Check
 
 on:
   pull_request:
-    branches: [master]
     types: [opened, synchronize, edited]
 
   push:

.github/workflows/pyink-check.yml

@@ -2,11 +2,12 @@ name: Formatter
 
 on:
   pull_request:
-    branches: [master]
     types: [opened, synchronize, edited]
   push:
     branches: [master]
 
+  workflow_dispatch: {}
+
 jobs:
   format_check:
     runs-on: ubuntu-latest

.github/workflows/pylint-check.yml

@@ -2,12 +2,13 @@ name: Linter
 
 on:
   pull_request:
-    branches: [master]
     types: [opened, synchronize, edited]
 
   push:
     branches: [master]
 
+  workflow_dispatch: {}
+
 jobs:
   linting_check:
     runs-on: ubuntu-latest

.github/workflows/require-checklist.yml

@@ -2,10 +2,13 @@ name: Require Checklist
 on:
   pull_request:
     types: [opened, edited, synchronize]
+
+  workflow_dispatch: {}
+
 jobs:
   check_pr_body:
     runs-on: ubuntu-latest
     steps:
       - uses: mheap/require-checklist-action@v2
         with:
-          requireChecklist: false # If this is true and there are no checklists detected, the action will fail
+          requireChecklist: false # If this is true and there are no checklists detected, the action will fail

.github/workflows/unit-test.yml

@@ -3,7 +3,6 @@ name: Unit Test
 
 on:
   pull_request:
-    branches: [master]
     types: [opened, synchronize, edited]
 
   push:

dags/common/scheduling_helper/scheduling_helper.py

@@ -72,7 +72,7 @@ class DayOfWeek(enum.Enum):
         "tpu_sdk_monitoring_validation": DefaultTimeout,
         "jobset_ttr_kill_process": dt.timedelta(minutes=90),
         "jobset_uptime_validation": dt.timedelta(minutes=90),
-        "jobset_ttr_drain_restart": DefaultTimeout,
+        "jobset_ttr_drain_restart": dt.timedelta(minutes=90),
         "tpu_info_metrics_verification": DefaultTimeout,
         "jobset_ttr_node_reboot": dt.timedelta(minutes=90),
     },

scripts/code-style.sh

@@ -22,14 +22,37 @@ FOLDERS_TO_FORMAT=("dags" "xlml")
 echo "[code-style] Running Semgrep static analysis..."
 semgrep --config scripts/semgrep-rules.yaml --error
 
-for folder in "${FOLDERS_TO_FORMAT[@]}"
-do
-  pyink "$folder" --pyink-indentation=2 --pyink-use-majority-quotes --line-length=80 --check --diff
-done
-
-for folder in "${FOLDERS_TO_FORMAT[@]}"
-do
-  pylint "./$folder" --fail-under=9.6
-done
+HEAD_SHA="$(git rev-parse HEAD)"
+BASE_BRANCH="dev"
+
+if ! git rev-parse --verify "$BASE_BRANCH" >/dev/null 2>&1; then
+  git fetch origin "$BASE_BRANCH":"$BASE_BRANCH" || {
+    echo "[code-style] base branch '$BASE_BRANCH' not found, skip diff-based check."
+    exit 0
+  }
+fi
+
+CHANGED_PY_FILES="$(
+  git diff --name-only --diff-filter=ACM "${BASE_BRANCH}" "${HEAD_SHA}" \
+    | grep '\.py$' \
+    | while read -r f; do
+        for folder in "${FOLDERS_TO_FORMAT[@]}"; do
+          if [[ "$f" == "$folder/"* ]]; then
+            echo "$f"
+            break
+          fi
+        done
+      done \
+    | sort -u
+)"
+
+if [[ -z "${CHANGED_PY_FILES}" ]]; then
+  echo "[pre-push hook] no changed files detected between ${HEAD_SHA} and ${BASE_BRANCH}"
+  exit 1
+fi
+
+pyink ${CHANGED_PY_FILES} --pyink-indentation=2 --pyink-use-majority-quotes --line-length=80 --check --diff
+
+pylint ${CHANGED_PY_FILES} --fail-under=9.6 --disable=E1123
 
 echo "Successfully clean up all codes."