Service cache not invalidated after owner ack, causing prolonged request failures

[Ravenyjh]

Problem

When a provider's TEE signer has not been acknowledged by the owner, incoming requests correctly fail with "service not acknowledge the tee signer". However, the service state (TeeSignerAcknowledged: false) gets cached for 15 minutes in serviceCache. Even after the owner acknowledges the TEE signer on-chain, all subsequent requests continue to fail until the cache expires.

Root Cause

ValidateRequestWithEstimatedFee in request.go:298-324 uses serviceCache (15-min TTL) to check TeeSignerAcknowledged. There is no mechanism to invalidate this cache when the owner performs an ack on-chain. The only cache clearing happens in SyncService() (service.go:65), which is called when the provider syncs its own config — not triggered by ack events.

Impact

1. Request layer: All user requests fail for up to 15 minutes after the owner acks, even though the on-chain state is already correct.
2. Settlement layer: ensureTeeSignerReady() in settlement_processor.go provides a guard, but if settlement somehow proceeds with NO_TEE_SIGNER status, it's treated as a permanent failure (settlement_tee.go:579-592), deleting all pending requests for the affected user — resulting in lost fees.

Affected Code

• api/inference/internal/ctrl/request.go:298-328 — cached service check
• api/inference/internal/ctrl/service.go:24-42 — GetCachedService() with no ack-aware invalidation
• api/inference/internal/ctrl/settlement_tee.go:579-592 — permanent failure handling deletes all pending requests

Possible Solutions

• Watch for ProviderTEESignerAcknowledged on-chain events and invalidate serviceCache upon receipt
• Reduce cache TTL for unacknowledged services (e.g., use short TTL when TeeSignerAcknowledged == false)
• Bypass cache when service is in unacknowledged state, always fetching fresh from contract