dpla-ingest-debug

name: dpla-ingest-debug description: Debug and fix DPLA hub ingestion failures (harvest/mapping/enrichment/jsonl/s3-sync/anomaly). Use when user asks why a hub failed, to debug an ingest failure, check an escalation report, or retry a failed hub/stage.

DPLA Ingest Debugging

Goal

Quickly identify what stage failed, find the relevant logs/escalation report, apply a targeted fix, and re-run only the necessary steps.

Environment

For any commands that depend on project env (especially the orchestrator), run source .env first so JAVA_HOME, DPLA_DATA, I3_CONF, and SLACK_WEBHOOK are available.

Gather Context (fast path)

1. If this was an orchestrator run, start with the escalation report (if present):

ls -lt data/escalations/ | head

2. Check current per-hub status (orchestrator writes these continuously):

./scripts/status/ingest-status.sh --watch
# or for one hub:
./scripts/status/ingest-status.sh <hub>

3. Find recent logs:

ls -lt logs/ | head

Classify the Failure Stage

Look for one of: harvest, mapping, enrichment, jsonl, sync, anomaly.

If you have only a hub name and "it failed", the quickest approach is:

• check logs/status/<hub>.status (JSON), and
• check the newest matching log in logs/.

Before Re-running: Resolve harvest type and runbook

Before re-running any scripts, determine the correct runbook for the hub:

1. Read harvest type from i3.conf: grep "^<hub>\.harvest\.type" "$I3_CONF" (or use the dpla-hub-info skill).
2. Choose the matching runbook for that harvest type: file, api, localoai, or special-case runbooks (Smithsonian preprocessing, NARA delta workflow). See runbooks/README.md.
3. Only then run harvest.sh, remap.sh, s3-sync.sh, or orchestrator retries — do not apply a generic rerun that's inappropriate for the hub type.

Re-run the Minimal Steps

All commands below assume you're running from repo root and that you've resolved harvest type and runbook (see above).

Harvest failed:

./scripts/harvest.sh <hub>

Mapping/enrichment/jsonl failed (re-run remap):

./scripts/remap.sh <hub>

S3 sync failed (or you want to re-sync after a successful run):

./scripts/s3-sync.sh <hub>

Orchestrator retry (failed hubs from last run):

./venv/bin/python -m scheduler.orchestrator.main --retry-failed

Orchestrator retry (one hub):

./venv/bin/python -m scheduler.orchestrator.main --hub=<hub>

Common Issues / Fix Heuristics

• Timeout / unreachable feed: retry harvest; if repeated, confirm the endpoint in i3.conf and capture the exact error line for escalation.
• Missing input (mapping complains about harvest input): harvest didn't produce output → re-run harvest, then remap.
• sbt contention / orphan processes: ensure the fat JAR is built (sbt assembly) before running pipeline scripts. If you must intervene, inspect before killing:
  • pgrep -fl 'sbt|java.*ingestion3'
  • then kill <pid> (avoid broad pkill patterns unless you're sure).
• Smithsonian / NARA special workflows: do not improvise; follow the dedicated runbooks (Smithsonian preprocessing, NARA delta workflow).

Verify Output

After re-running, verify _SUCCESS markers and counts for all pipeline steps (harvest, mapping, enrichment, jsonl):

ls "$DPLA_DATA/<hub>/harvest"/*/_SUCCESS
ls "$DPLA_DATA/<hub>/mapping"/*/_SUCCESS
ls "$DPLA_DATA/<hub>/enrichment"/*/_SUCCESS
ls "$DPLA_DATA/<hub>/jsonl"/*/_SUCCESS

cat "$DPLA_DATA/<hub>/mapping"/*/_SUMMARY | head
cat "$DPLA_DATA/<hub>/enrichment"/*/_SUMMARY | head

The pipeline is not successful if enrichment failed; ensure all four stages have _SUCCESS markers.