Retries And Recovery
External retries are normal. The external orchestrator decides whether to start another command after a failure. Firecube checks ChunkManager before writing so a retry does not blindly overwrite or overlap product state.
Retry Flow
- The external orchestrator starts
firecube ingest. - If the command succeeds, the external orchestrator can continue to the next step.
- If the command exits non-zero, the external orchestrator may retry according to its own policy.
- On retry, Firecube checks ChunkManager before writing.
- If a previous run is still active, a claim is stale, or the requested work conflicts with existing records, Firecube fails closed.
At that point, inspect the product before retrying again:
uv run firecube chunks runs list --product file:///data/products/MY_PRODUCT.zarr
uv run firecube chunks claims list --product file:///data/products/MY_PRODUCT.zarr
Use Recover Runs And Claims when a failed command leaves a stuck run or stale claim.
Make Retry Intent Explicit
Use resume_existing=true when the retry should continue work that has already
started:
uv run firecube ingest <plugin> \
--source /data/source \
--target file:///data/products/MY_PRODUCT.zarr \
--product-name MY_PRODUCT \
--storage-type local \
--storage-driver fsspec \
--output-format zarr \
--write-mode staged \
--option resume_existing=true
Use force or deletion workflows only when the operation page tells you to. They are recovery actions, not generic retry flags.
Correlate Runs
Pass a stable --option run_id=... when you need Firecube run records to line up
with the external orchestrator job:
uv run firecube ingest <plugin> \
--source /data/source \
--target file:///data/products/MY_PRODUCT.zarr \
--product-name MY_PRODUCT \
--storage-type local \
--storage-driver fsspec \
--output-format zarr \
--write-mode staged \
--option run_id="$RUN_ID"
Metrics, logs, and traces have their own configuration. See Observability for correlation details.
Next Steps
- Recover Runs And Claims — recover a stuck run or stale claim
- Inspect ChunkManager State — inspect runs, claims, spans, or snapshots
- Observability — connect metrics, logs, and traces to external orchestrator jobs
- Production Guide — choose reliable production defaults