Stores

Persist batches watched by BatchPoller.

BatchPoller depends on the async BatchStore protocol: get(batch_id), set(record), delete(batch_id), and list(delivered=None). Custom stores can implement that protocol without inheriting a Batchwork class.

Memory

MemoryBatchStore is process-local and guarded by an async lock. It is suitable for development, tests, and one-process services; restarting the process loses tracked batches.

from batchwork import BatchPoller, MemoryBatchStore

poller = BatchPoller(MemoryBatchStore())

Upstash Redis

Redis support is optional and imported only when a Redis store is created:

uv add "batchwork[redis]"

Without the extra, creating the default Redis client raises MissingDependencyError. Configure Upstash’s standard variables:

UPSTASH_REDIS_REST_URL=https://...
UPSTASH_REDIS_REST_TOKEN=...

With no injected client, create_redis_store() creates an async Upstash client from those variables:

from batchwork import BatchPoller, create_redis_store

store = create_redis_store(prefix="my-service")
poller = BatchPoller(store)

The prefix defaults to batchwork and must not be empty. Use a distinct prefix when applications share a database.

RedisBatchStore also accepts an injected sync or async Upstash-compatible client. In that case no client is created from the environment:

from batchwork import BatchPoller, create_redis_store

store = create_redis_store(redis=my_redis_client, prefix="my-service")
poller = BatchPoller(store)

The base batchwork install does not import or require upstash-redis.