Hands-on Examples

English · 中文版

Install from the repository root:

uv sync

The examples call the Direct API and create only temporary state.

1. AOF crash recovery

uv run python examples/aof_crash_recovery.py

Expected landmarks:

1. SET before crash: Ok(message=b'OK')
2. Simulating a crash (no graceful AOF drain)...
3. GET after restart: Bytes(value=b'durable')
4. Recovery verified: True

Watch the AofPolicy.ALWAYS boundary: the successful SET crosses the AOF durability barrier before the reply is released. simulate_crash() skips a graceful drain; reopening and replay still recover the value.

2. Deterministic LFU eviction

uv run python examples/lfu_eviction.py

Expected landmarks:

cold (least frequent, expected missing): Bytes(value=None)
hot  (expected retained): Bytes(value=b'x')
new  (expected retained): Bytes(value=b'xxx...')

Watch public replies rather than internal counters. Four reads make hot more frequent than cold; inserting new exceeds the configured logical-memory budget, so deterministic allkeys-LFU removes cold.

3. Partial resynchronization and full-sync fallback

uv run python examples/replication_resync.py

Expected landmarks:

Initial attachment: full
Short disconnect resumed with: partial
Cursor older than backlog resumed with: full

The first disconnect misses only one retained batch, so the replication cursor can resume from the backlog. The second misses more history than the two-batch backlog retains, forcing a complete state transfer. This is an in-process logical model of the PSYNC decision, not Redis wire replication.

Continue with executable evidence

Use the behavior matrix to find the focused pytest node for a mechanism, or run the entire suite:

uv run pytest -q