Logging
Add --log to any command. The default level is info. Use --log debug only when you are chasing a specific failure — it is noisy.
skipprd sync --pipeline payments --once --logRead logs in the terminal or the collector you already use. Skipprd writes structured lines you can grep.
A successful run
Treat a sync --once as good when all of these are true:
- The process exits 0.
- The log contains
Pipeline sync complete. - Uploaded row counts match what Skipprd expected, and no partitions were quarantined.
- A query against the destination returns the rows you expected.
Typical lines on a healthy pass, in order:
| Line (search for) | Meaning |
|---|---|
Syncing pipeline: | The named pipeline started |
Discovered new namespace: / Discovered new field: | First-seen table or column — expected on the first run and after a schema change |
Uploaded + parquet + rows= | A file landed in the destination |
Pipeline sync complete | The run finished |
On shutdown you may also see a drain line such as Finalising: compactor drained and stopped. That is the process finishing in-flight writes. The success marker you operate on is still Pipeline sync complete plus exit 0.
When to treat the run as failed
| Line (search for) | Meaning | What you do |
|---|---|---|
Finalising: compactor drain/stop did not complete cleanly | In-flight writes did not finish | Treat counts as untrusted. Re-run the same command. |
integrity check mismatch or quarantined_parts greater than 0 | Uploaded rows do not match, or a slice could not be read | Compare destination counts to the source. See Troubleshooting. |
Pipeline '<name>' not found | No schema for that pipeline | Run skipprd discover --pipeline <name> first. |
TABLE_NOT_FOUND | The warehouse table is missing | Discover succeeded? Then re-run sync so the destination can create the table. |
| Non-zero exit | The run said it failed | Re-run. Committed WAL batches are retried. |
Throughput lines (Messages per Min, Uploads total) are trends, not a pass/fail signal.
Chaos testing
SKIPPR_CHAOS_MODE=yes kills the process on purpose. You should see a SIGKILL, then a later run that recovers. Leave it off in production. See Advanced settings.