New automatic strategy for `ServiceSwitcher` that works on service failures

[2010b9]

Problem Statement

Currently, there's only one service switcher strategy, which is the ServiceSwitcherStrategyManual. As the name implies, it only works if service is switched explicitly. I believe this could be improved by adding a new strategy that automatically switches to another service when the primary one is down, keeps track of the services' state in the background and switches to the main one once it is healthy again.

Proposed Solution

I think something similar to what Livekit does with their FallbackAdapter could be done for pipecat as well.

When in use, FallbackAdapter handles the following:

• Automatically resubmits the failed request to backup providers when the primary provider fails.
• Marks the failed provider as unhealthy and stops sending requests to it.
• Continues to use the backup providers until the primary provider recovers.
• Periodically checks the primary provider's status in the background.

Alternative Solutions

No response

Additional Context

No response

Would you be willing to help implement this feature?

• Yes, I'd like to contribute
• No, I'm just suggesting

[AndreTeixeira1998]

Second this. Would be useful to have some automatic mechanism to switch services if the primary fails

[kollaikal-rupesh]

I'd like to take a stab at this.

Looking at the existing ServiceSwitcher and ServiceSwitcherStrategyManual architecture, I think the approach would be:

1. Add a handle_error() hook to the base ServiceSwitcherStrategy (default no-op) so strategies can react to service errors without new frame types
2. Have ServiceSwitcher.push_frame() call strategy.handle_error() when it sees a non-fatal ErrorFrame from the active service
3. Implement ServiceSwitcherStrategyAutomatic that:
   • On error: marks the active service as unhealthy and switches to the next healthy backup
   • Tracks a recovery timeout per service (configurable, default 30s)
   • When the primary service's cooldown expires and prefer_primary is set, switches back automatically
   • Still supports ManuallySwitchServiceFrame for explicit overrides
4. Fires the existing on_service_switched event on every automatic switch so downstream observers stay informed

This reuses ServiceSwitcherFrame, ServiceSwitcherRequestMetadataFrame, and the filter gating pattern already in place. No changes to how services report errors (push_error with fatal=False).

@markbackman would appreciate your thoughts on this approach before I go too far.