The type of the final result
Create an error result when the server rejects the request or when a non-exception error occurs.
The GraphQL data provider instance
OptionaldeferHold PubSub messages that arrive before the acknowledgement, then route them, in arrival order, once validateAck has accepted it. For an operation whose identity comes FROM the acknowledgement (the server names the record it created), so a message that beats the ack over the WebSocket can still be attributed rather than lost. Off by default: messages are routed as they arrive, exactly as before.
Extract the final result from a matching completion event. Called when isCompletionEvent returns true.
OptionalextractOptional SYNCHRONOUS-fallback extractor for headless / non-browser clients.
The fire-and-forget pattern delivers its result over a WebSocket PubSub channel
(GraphQLDataProvider.PushStatusUpdates). A headless client — the integration
GraphQL client, a Node/MCP consumer, any provider without a subscription surface —
does not implement PushStatusUpdates, so it can never hear the completion event.
Historically this crashed with PushStatusUpdates is not a function.
When this extractor is supplied AND the data provider has no PushStatusUpdates
method, FireAndForgetHelper.Execute instead runs the mutation in SYNCHRONOUS
mode (it forces fireAndForget: false), so the resolver awaits the run and returns the
full result inline in the mutation's own return payload. This function receives that
ack payload (result[mutationFieldName]) and produces the final TResult.
When omitted, a provider lacking PushStatusUpdates degrades to a clear error result
via createErrorResult rather than crashing.
Check if a parsed PubSub message is the matching completion event. Called for every PubSub message received on the session. Return true only for the completion event that matches this specific request.
OptionalisOptional predicate deciding whether a PubSub message belongs to THIS operation.
The push-status subscription is per-SESSION, not per-operation, so without this every message for every concurrent run resets every pending operation's idle timer. With several agents in flight, one chatty run holds a dead run's timer open indefinitely — the "idle" window silently means "the whole session went quiet", which is not what any caller intends (MJ #4222).
MUST FAIL OPEN: return true for anything not positively identifiable as another
operation's. Wrongly ignoring a message that IS ours shortens the window and causes a
spurious reconcile; wrongly accepting one that is not merely preserves the old
behaviour. When omitted, every message counts as activity, exactly as before.
OptionalmaxMax consecutive 'continue' reconciliations before giving up (default: 6)
The GraphQL mutation document (gql tagged template)
The mutation field name to extract the acknowledgement from the result e.g., 'RunTest', 'RunAIAgent', 'RunAIAgentFromConversationDetail'
OptionalonOptional handler for all PubSub messages (progress, streaming, etc.). Called for every message, not just completion events. Use this to forward progress updates to UI callbacks.
OptionalonOptional reconciliation hook invoked when the idle timer expires (no message received within the idle window) or the PubSub stream drops. Should query the authoritative server-side run record and return a StallDecision. When omitted, an idle expiry rejects with a timeout message (legacy behavior).
OptionaloperationLabel for logging (e.g., 'RunTest', 'RunAIAgent')
OptionaltimeoutCustom timeout error message
OptionaltimeoutIdle timeout in milliseconds — resets on activity (default: 12 minutes)
Check if the server accepted the fire-and-forget request.
The ack result comes from result[mutationFieldName].
Return true if accepted, false if rejected.
Variables for the mutation (should include fireAndForget: true)
Configuration for a fire-and-forget GraphQL mutation execution.
Fire-and-forget avoids Azure's ~230s HTTP proxy timeout by:
fireAndForget: true