Advanced usage¶
Choose the container that matches the access pattern. Examples use the illustrative
example namespace and DEMO source from the quick start.
| Need | Interface | Read more |
|---|---|---|
| Structured records and primary-key lookup | table(), upsert(), loc |
Tables |
| Timestamped numeric matrices | timeseries(), write_timeseries() |
Time series |
| Document queries | collection() |
API reference |
| Latest entity state | cache() |
Redis caches |
| Streaming messages | stream() |
API reference |
| Source selection by symbol | dataframe(), semantic HTTP API |
Data by symbol |
Process boundaries¶
Local consumers map the same files through the page cache. Coordinate indexed writes through the provided process lock. Retain refreshed views after structural operations rather than assuming an old NumPy or pandas view remains valid.
Remote consumers use the authenticated HTTP API. Remote reads serialize data and do not share the source host's mapped pages. The semantic API resolves sources and copies rows; keep direct container access in performance-critical producers.
Execution¶
Use tracked worker commands for routines and batch jobs. A queued command is not proof of successful execution: check worker state, job results and logs. Read schedules and batch jobs before building dependency chains.
Durability¶
Memory visibility, filesystem persistence and remote backup are separate concerns. Choose a backup policy for your workload, verify recovery on disposable data, and read storage guarantees before relying on crash recovery.