Patterns for serverless handlers, read-only filesystems, cloud sync, and backend rate limits.
Cloud sync is asynchronous. Always flush before short-lived handlers exit. Provider-call retry lives under Reliability.
Pick one deployment topic below — each page covers a single concern.
flush() and flushAndWait() before Lambda or Edge handlers return.
Read-only FS fallbacks, AUREX_LEDGER_PATH, per_process_ledger.
HTTP retries, 429 handling, queue drops — not provider-call retry.
Per-key ingest limits and escalation auth.
| Environment | Recommended |
|---|---|
| AWS Lambda | AUREX_LEDGER_PATH=/tmp/..., flush at handler end |
| Kubernetes | per_process_ledger, cloud_sync, monitor queue_size |
| Vercel / Edge | cloud_sync + memory storage; explicit flushAndWait |