🤖 ci: improve extract_pr_logs to show in-progress job steps

Author: ammar-agent

scripts/extract_pr_logs.sh

@@ -1,54 +1,67 @@
 #!/usr/bin/env bash
-# Extract logs from failed GitHub Actions runs for a PR
-# Usage: ./scripts/extract_pr_logs.sh <pr_number_or_run_id> [job_name_pattern] [--wait]
+# Extract logs from GitHub Actions runs for a PR (including in-progress jobs)
+# Usage: ./scripts/extract_pr_logs.sh <pr_number_or_run_id> [job_name_pattern] [--all]
 #
 # Examples:
 #   ./scripts/extract_pr_logs.sh 329              # Latest failed run for PR #329
 #   ./scripts/extract_pr_logs.sh 329 Integration  # Only Integration Test jobs
-#   ./scripts/extract_pr_logs.sh 329 --wait       # Wait for logs to be available
+#   ./scripts/extract_pr_logs.sh 329 --all        # Show all jobs (not just failed)
 #   ./scripts/extract_pr_logs.sh 18640062283      # Specific run ID
 
 set -euo pipefail
 
 INPUT="${1:-}"
 JOB_PATTERN="${2:-}"
-WAIT_FOR_LOGS=false
+SHOW_ALL_JOBS=false
 
 # Parse flags
-if [[ "$JOB_PATTERN" == "--wait" ]]; then
-  WAIT_FOR_LOGS=true
+for arg in "$@"; do
+  if [[ "$arg" == "--all" ]]; then
+    SHOW_ALL_JOBS=true
+  fi
+done
+
+# Remove flags from JOB_PATTERN if they were set as second arg
+if [[ "$JOB_PATTERN" == "--all" ]]; then
   JOB_PATTERN=""
-elif [[ "${3:-}" == "--wait" ]]; then
-  WAIT_FOR_LOGS=true
 fi
 
 if [[ -z "$INPUT" ]]; then
-  echo "❌ Usage: $0 <pr_number_or_run_id> [job_name_pattern]" >&2
+  echo "❌ Usage: $0 <pr_number_or_run_id> [job_name_pattern] [--all]" >&2
   echo "" >&2
   echo "Examples:" >&2
-  echo "  $0 329              # Latest failed run for PR #329 (RECOMMENDED)" >&2
+  echo "  $0 329              # Latest failed run for PR #329" >&2
   echo "  $0 329 Integration  # Only Integration Test jobs from PR #329" >&2
+  echo "  $0 329 --all        # Show all jobs (not just failed/in-progress)" >&2
   echo "  $0 18640062283      # Specific run ID" >&2
   exit 1
 fi
 
 # Detect if input is PR number or run ID (run IDs are much longer)
 if [[ "$INPUT" =~ ^[0-9]{1,5}$ ]]; then
   PR_NUMBER="$INPUT"
-  echo "🔍 Finding latest failed run for PR #$PR_NUMBER..." >&2
-
-  # Get the latest failed run for this PR
-  RUN_ID=$(gh pr checks "$PR_NUMBER" --json name,link,state --jq '.[] | select(.state == "FAILURE") | .link' | head -1 | sed -E 's|.*/runs/([0-9]+).*|\1|' || echo "")
+  
+  # If --all flag is set or no failures, get latest run regardless of status
+  if [[ "$SHOW_ALL_JOBS" == true ]]; then
+    echo "🔍 Finding latest run for PR #$PR_NUMBER..." >&2
+    RUN_ID=$(gh pr checks "$PR_NUMBER" --json name,link,state --jq '.[] | select(.link | contains("/runs/")) | .link' | head -1 | sed -E 's|.*/runs/([0-9]+).*|\1|' || echo "")
+  else
+    echo "🔍 Finding latest failed run for PR #$PR_NUMBER..." >&2
+    # Get the latest failed run for this PR
+    RUN_ID=$(gh pr checks "$PR_NUMBER" --json name,link,state --jq '.[] | select(.state == "FAILURE") | select(.link | contains("/runs/")) | .link' | head -1 | sed -E 's|.*/runs/([0-9]+).*|\1|' || echo "")
+  fi
 
   if [[ -z "$RUN_ID" ]]; then
     echo "❌ No failed runs found for PR #$PR_NUMBER" >&2
     echo "" >&2
     echo "Current check status:" >&2
     gh pr checks "$PR_NUMBER" 2>&1 || true
+    echo "" >&2
+    echo "💡 Tip: Use --all flag to see logs from any run (not just failed)" >&2
     exit 1
   fi
 
-  echo "📋 Found failed run: $RUN_ID" >&2
+  echo "📋 Found run: $RUN_ID" >&2
 else
   RUN_ID="$INPUT"
   echo "📋 Fetching logs for run $RUN_ID..." >&2
@@ -65,12 +78,18 @@ if [[ -z "$JOBS" ]]; then
   exit 1
 fi
 
-# Filter to failed jobs only (unless specific pattern requested)
-if [[ -z "$JOB_PATTERN" ]]; then
-  FAILED_JOBS=$(echo "$JOBS" | jq -r 'select(.conclusion == "FAILURE" or .conclusion == "TIMED_OUT" or .conclusion == "CANCELLED")')
+# Filter jobs based on flags and pattern
+if [[ -z "$JOB_PATTERN" ]] && [[ "$SHOW_ALL_JOBS" == false ]]; then
+  # Show failed/timed out/cancelled jobs, OR in-progress/pending jobs if no failures exist
+  FAILED_JOBS=$(echo "$JOBS" | jq -r 'select(.conclusion == "failure" or .conclusion == "timed_out" or .conclusion == "cancelled")')
+  IN_PROGRESS_JOBS=$(echo "$JOBS" | jq -r 'select(.status == "in_progress" or .status == "queued" or .status == "pending")')
+  
   if [[ -n "$FAILED_JOBS" ]]; then
-    echo "🎯 Showing only failed jobs (use job_pattern to see others)" >&2
+    echo "🎯 Showing only failed jobs (use --all to see all jobs)" >&2
     JOBS="$FAILED_JOBS"
+  elif [[ -n "$IN_PROGRESS_JOBS" ]]; then
+    echo "⏳ No failures yet - showing in-progress/pending jobs (use --all to see all)" >&2
+    JOBS="$IN_PROGRESS_JOBS"
   fi
 fi
 
@@ -111,43 +130,57 @@ suggest_local_command() {
   esac
 }
 
+# Show step-by-step progress for in-progress/pending jobs
+show_job_steps() {
+  local job_id="$1"
+  local job_status="$2"
+  
+  if [[ "$job_status" == "in_progress" ]] || [[ "$job_status" == "queued" ]] || [[ "$job_status" == "pending" ]]; then
+    echo "" >&2
+    echo "📊 Step-by-step status:" >&2
+    gh api "/repos/coder/cmux/actions/jobs/$job_id" | jq -r '.steps[] | "  [\(.status | ascii_upcase)] \(.name)\(if .conclusion then " (\(.conclusion))" else "" end)"' >&2
+    echo "" >&2
+  fi
+}
+
 # Extract and display logs for each job
 for JOB_ID in $JOB_IDS; do
   JOB_INFO=$(echo "$JOBS" | jq -r "select(.databaseId == $JOB_ID)")
   JOB_NAME=$(echo "$JOB_INFO" | jq -r '.name')
-  JOB_STATUS=$(echo "$JOB_INFO" | jq -r '.conclusion // .status')
+  JOB_STATUS=$(echo "$JOB_INFO" | jq -r '.status')
+  JOB_CONCLUSION=$(echo "$JOB_INFO" | jq -r '.conclusion // "N/A"')
+  
+  # Display status: show conclusion if completed, otherwise show status
+  if [[ "$JOB_STATUS" == "completed" ]]; then
+    DISPLAY_STATUS="$JOB_CONCLUSION"
+  else
+    DISPLAY_STATUS="$JOB_STATUS"
+  fi
 
   echo "" >&2
   echo "════════════════════════════════════════════════════════════" >&2
-  echo "Job: $JOB_NAME (ID: $JOB_ID) - $JOB_STATUS" >&2
+  echo "Job: $JOB_NAME (ID: $JOB_ID)" >&2
+  echo "Status: $DISPLAY_STATUS" >&2
   echo "════════════════════════════════════════════════════════════" >&2
 
   # Suggest local reproduction command
   suggest_local_command "$JOB_NAME" >&2
-  echo "" >&2
-
-  # Fetch logs with retry logic if --wait flag is set
-  MAX_RETRIES=3
-  RETRY_COUNT=0
+  
+  # Show step-by-step status for in-progress/pending jobs
+  show_job_steps "$JOB_ID" "$JOB_STATUS"
 
-  while [ $RETRY_COUNT -lt $MAX_RETRIES ]; do
-    # Use gh api to fetch logs (works for individual completed jobs even if run is in progress)
+  # Try to fetch logs
+  if [[ "$JOB_STATUS" == "completed" ]]; then
+    # Completed job - logs should be available
     if gh api "/repos/coder/cmux/actions/jobs/$JOB_ID/logs" 2>/dev/null; then
-      break
+      echo "" >&2
     else
-      RETRY_COUNT=$((RETRY_COUNT + 1))
-      if [ $RETRY_COUNT -lt $MAX_RETRIES ] && [ "$WAIT_FOR_LOGS" = true ]; then
-        echo "⏳ Logs not ready yet, waiting 5 seconds... (attempt $RETRY_COUNT/$MAX_RETRIES)" >&2
-        sleep 5
-      else
-        echo "⚠️  Could not fetch logs for job $JOB_ID" >&2
-        if [ "$WAIT_FOR_LOGS" = false ]; then
-          echo "   Tip: Use --wait flag to retry if logs are still processing" >&2
-        else
-          echo "   (logs may have expired or are still processing)" >&2
-        fi
-        break
-      fi
+      echo "⚠️  Could not fetch logs for completed job $JOB_ID (logs may have expired)" >&2
+      echo "" >&2
     fi
-  done
+  else
+    # In-progress/pending/queued job - logs not yet available
+    echo "ℹ️  Job is $JOB_STATUS - logs will be available when job completes" >&2
+    echo "" >&2
+  fi
 done