--parallel-partitions¶
Why we added this flag: when manifests contain many partitions, running them sequentially can take hours; limited parallelism keeps throughput high while respecting API quotas.
What it does¶
- Controls how many partitions idxr indexes concurrently (default:
1). - Only applies to manifest-driven runs.
- Each worker maintains its own resume state to ensure idempotency.
Typical usage¶
idxr vectorize index \
--model "$IDXR_MODEL" \
--partition-manifest workdir/partitions/manifest.json \
--partition-out-dir workdir/chroma_partitions \
--parallel-partitions 4
Tips¶
- Match the value to your embedding API rate limits. Start with
2and scale up while monitoring throttling. - Ensure the machine hosting idxr has enough CPU and memory—each partition spawns its own worker thread.