Durable storage backend support (NATS JetStream / Redis / Postgres)?

[joeblew999]

Hi team 👋

I’ve been evaluating a2a-go for use in a multi-VM environment (Fly.io single-container VMs).

One limitation I’ve run into is that the current task manager and event queue appear to be in-memory only, meaning:

• tasks and events are not durable
• state is lost on process/VM restart
• multi-node / cluster deployments can’t coordinate work

Before building my own backend layer, I wanted to ask:
Is there any existing or planned support for a durable backend?

(E.g. NATS JetStream, Redis, or Postgres)
Specifically, I’m looking for a way to persist:
task queue
task execution state
event log
task results

The Python ecosystem has a2a-redis, but I couldn’t find an equivalent for Go.
Is durable storage on the roadmap for a2a-go, or should implementers provide their own backend via the extension points being discussed in issue #115?

Thanks!

[yarolegovich]

Hi @joeblew999, it's great to hear there's interest in these things.

We're certainly planning to provide some generic integrations with popular backends as well as a multi-process deployment example, but it's not the near-term focus. Spec 1.0.0 work is ongoing and expected to land soon, so the effort is going to be shifted to supporting it as soon as possible.

Then #115 in my mind is a pre-requisite for building proper storage backend and messaging backend integrations. As part ofi t TaskStore interface will likely be extended to support optimistic concurrency updates like SaveWhere(ctx, old, new).

I'll be happy to assist with reviews or design if you decide to contribute.
I would probably start with providing database/sql-based implementations for everything: a2asrv.TaskStore, workqueue.Queue and eventqueue.Queue. An appropriate driver will be selected by users.

[yarolegovich]

You might find #133 interesting. We'll likely be providing implementations similar to what's in there once interfaces get updated.