Appearance

Troubleshooting ​

Start here when a command exits non-zero or the destination does not have the rows you expected. Re-run the same command after you fix the cause. Skipprd retries committed batches from the write-ahead log; you do not rewind the source by hand.

Pipeline '<name>' not found ​

Cause: Skipprd has no schema for that pipeline yet.

Fix: Discover first, then sync:

bash
skipprd discover --pipeline payments --log
skipprd sync --pipeline payments --once --log

Check the name matches pipelines: in skippr.yml.

TABLE_NOT_FOUND ​

Cause: The destination table does not exist yet. Common on the first sync, or when discover never finished.

Fix: Confirm discover completed and the destination is paired with its schema sink (Athena needs Glue). Then run sync again so tables can be created.

Destination credential errors ​

Cause: The warehouse or object store rejected the caller.

Fix: Export the variables your connector page names (for AWS: AWS_ACCESS_KEY_ID, AWS_SECRET_ACCESS_KEY, AWS_DEFAULT_REGION, or an instance role). Run the command in the same shell. skipprd doctor reports missing engine settings.

Skipprd could not download a connector ​

The log line starts Skippr could not find a published runtime.

Cause: Skipprd could not download that source or destination. The name is wrong, the pinned version does not exist, or the machine cannot reach install.skippr.io.

Fix:

  1. Check the connector name and casing in skippr.yml against the connector catalog.
  2. Remove version: if you meant “latest”.
  3. Confirm outbound HTTPS to install.skippr.io.
  4. On a first run, wait — Skipprd caches the download under ~/.skippr/runtime_plugins.

The run did not finish cleanly ​

Look for Finalising: compactor drain/stop did not complete cleanly or integrity check mismatch.

Cause: Skipprd exited before in-flight writes settled, or uploaded counts do not match.

Fix: Treat that run's counts as untrusted. Re-run:

bash
skipprd sync --pipeline payments --once --log

The next run recovers from the WAL. Confirm Pipeline sync complete and that destination counts match the source. If a summary reports quarantined_parts greater than 0, a slice failed to read — check source files or objects for corruption, then re-run.

Deadletters ​

Discarded <n> deadletter records because no deadletter sink is configured means rows failed validation and were dropped. Add a deadletter_sink if you want to inspect them. See Deadletters.

Deadletter id=… ns=… err=… is one failed record. Query the deadletter destination and fix the payload or the schema.

Recovery after a crash ​

You do not run a special recover command.

  1. The WAL still holds committed batches.
  2. Resume positions say how far the source got.
  3. The next skipprd sync writes remaining batches and continues the source.

Use the same DATA_DIR (or the same S3 WAL and state store). If you deleted them, Skipprd re-reads the source and append-only destinations can duplicate rows. See WAL and buffering.

Next steps ​