feat(ingestion): async background processing with progressive UI

.github/workflows/deploy-mcp.yml

@@ -75,6 +75,8 @@ jobs:
             --cpu 1 \
             --min-instances 0 \
             --max-instances 10 \
+            --no-cpu-throttling \
+            --cpu-boost \
             --timeout 300 \
             --port 8080 \
             --set-env-vars "ENVIRONMENT=${{ github.event.inputs.environment }},MCP_SERVICE_NAME=${{ env.SERVICE_NAME }}" \

000-docs/ingestion-run-button-debug.md

@@ -0,0 +1,149 @@
+# Ingestion Run Button - Debug Runbook
+
+## System Map
+
+```text
+Dashboard (IngestionButton.tsx)
+    │
+    ├── POST /trigger/ingestion → 202 Accepted {run_id, poll_url}
+    │   └── Background: run_ingestion_pipeline()
+    │       ├── Phase: loading_sources (rss_sources.yaml)
+    │       ├── Phase: fetching_feeds (128x /mcp/tools/fetch_rss_feed)
+    │       ├── Phase: storing_articles (batches → /mcp/tools/store_articles)
+    │       └── Phase: upserting_authors (/mcp/tools/upsert_author)
+    │
+    ├── Poll: GET /trigger/ingestion/{run_id} (every 3s)
+    │   └── Reads Firestore ingestion_runs/{run_id}
+    │
+    └── On completion: CustomEvent('ingestion-complete')
+        └── SystemActivityCard auto-refreshes
+```
+
+## Failure Taxonomy
+
+### F1: No toast appears after clicking Run Ingestion
+- **Cause**: POST failed (network, CORS, Cloud Run cold start timeout)
+- **Debug**:
+  ```bash
+  curl -X POST <SERVICE_URL>/trigger/ingestion \
+    -H "Content-Type: application/json" \
+    -d '{"trigger":"manual"}'
+  ```
+- **Fix**: Check Cloud Run logs, verify service is deployed, check CORS config
+
+### F2: "Already in progress" when nothing is running
+- **Cause**: Previous run stuck in `accepted` or `running` status
+- **Debug**:
+  ```bash
+  # Check Firestore for stuck runs
+  gcloud firestore documents list \
+    --project=perception-with-intent \
+    --database=perception-db \
+    --collection=ingestion_runs \
+    --filter='status="running"'
+  ```
+- **Fix**: Stale runs (>10 min) are auto-cleaned. Manual fix: update status to `failed` in Firestore console
+
+### F3: Stuck forever (spinner never stops)
+- **Cause**: Background task crashed, Firestore never updated to terminal state
+- **Debug**: Check Cloud Run logs for unhandled exceptions
+  ```bash
+  gcloud logging read 'resource.type="cloud_run_revision" AND resource.labels.service_name="perception-mcp" AND severity>=ERROR' \
+    --project=perception-with-intent --limit=20 --format=json
+  ```
+- **Fix**: Dashboard has 5-minute stuck detection. Add catch-all in `run_ingestion_pipeline`
+
+### F4: No data in SystemActivityCard
+- **Cause**: Firestore docs not being written, or Firebase client SDK can't read them
+- **Debug**:
+  ```bash
+  # Check if docs exist
+  gcloud firestore documents list \
+    --project=perception-with-intent \
+    --database=perception-db \
+    --collection=ingestion_runs --limit=5
+  ```
+- **Fix**: Verify `ingestion_runs` collection exists, check Firestore security rules
+
+### F5: High error count in completed run
+- **Cause**: RSS feeds returning errors (timeouts, 403s, DNS failures)
+- **Debug**: Check `errors` array in Firestore doc
+- **Fix**: Review error patterns, update/remove broken feeds from `rss_sources.yaml`
+
+### F6: 0 articles stored
+- **Cause**: All feeds failed, or `store_articles` endpoint broken
+- **Debug**:
+  ```bash
+  curl -X POST <SERVICE_URL>/mcp/tools/store_articles \
+    -H "Content-Type: application/json" \
+    -d '{"run_id":"test","articles":[{"title":"t","url":"https://x.com","source_id":"s","published_at":"2025-01-01"}]}'
+  ```
+- **Fix**: Check Firestore permissions for MCP service account
+
+### F7: Cold start timeout on POST
+- **Cause**: Cloud Run container starting up, POST times out before ready
+- **Debug**: Check Cloud Run startup latency in metrics
+- **Fix**: `--cpu-boost` flag enables faster cold starts; consider `--min-instances 1`
+
+## Key Diagnostic Queries
+
+### Cloud Logging
+```bash
+# All ingestion run logs
+gcloud logging read 'resource.type="cloud_run_revision" AND jsonPayload.router="trigger"' \
+  --project=perception-with-intent --limit=50
+
+# Specific run
+gcloud logging read 'resource.type="cloud_run_revision" AND jsonPayload.run_id="run-XXXXX"' \
+  --project=perception-with-intent --limit=50
+
+# Errors only
+gcloud logging read 'resource.type="cloud_run_revision" AND severity>=ERROR AND resource.labels.service_name="perception-mcp"' \
+  --project=perception-with-intent --limit=20
+```
+
+### Firestore
+```bash
+# List recent runs
+gcloud firestore documents list \
+  --project=perception-with-intent \
+  --database=perception-db \
+  --collection=ingestion_runs --limit=10
+
+# Get specific run
+gcloud firestore documents get \
+  --project=perception-with-intent \
+  --database=perception-db \
+  ingestion_runs/run-XXXXX
+```
+
+### curl Commands
+```bash
+# Health check
+curl <SERVICE_URL>/health
+
+# Trigger ingestion
+curl -X POST <SERVICE_URL>/trigger/ingestion \
+  -H "Content-Type: application/json" \
+  -d '{"trigger":"manual","time_window_hours":24,"max_items_per_source":50}'
+
+# Check run status
+curl <SERVICE_URL>/trigger/ingestion/run-XXXXX
+```
+
+## SLO Targets
+
+| Metric | Target | Alert Threshold |
+|--------|--------|-----------------|
+| POST response time | < 1s | > 3s |
+| End-to-end success rate | > 95% | < 90% |
+| Source error rate | < 10% | > 25% |
+| Total duration | < 300s | > 600s |
+| Articles stored per run | > 100 | < 10 |
+
+## Architecture Notes
+
+- Background task uses `asyncio.create_task()` - requires `--no-cpu-throttling` on Cloud Run
+- `--cpu-boost` gives extra CPU during cold starts
+- Idempotency guard prevents concurrent runs (10 min stale timeout)
+- Firestore composite index on `[status ASC, startedAt DESC]` for active run query