fix(maestro): stop retrying a WS attach that never opened

Integration review (#168): a rejected attach handshake — a dormant
session's 409 not_attachable, most likely — looks identical to a
dropped stream from the WebSocket API (close code 1006, no reason).
shouldRetryClose kept retrying it on the catalog's Running/Dead check
alone, which loops for as long as the catalog hasn't caught up with
the session going dormant.

Track whether the socket ever reached `open`; a handshake that never
did is a rejection, not a blip, so it is never retried regardless of
catalog status. The caller (wired up in the next commit) gets a
distinct onNeverAttached callback instead of onClosed, so it can offer
the resume affordance rather than assuming the session is gone.
This commit is contained in:
Miguel Palhas
2026-09-04 09:19:09 +01:00
parent 8269bcabcc
commit 938dd62648
4 changed files with 46 additions and 5 deletions
+8
View File
@@ -26,7 +26,15 @@ export function shouldRetryClose(
sessionId: Uuid,
code: number,
sessions: ReadonlyMap<Uuid, Session>,
/** Did this socket ever reach `open`? A handshake that never completed is
* the daemon rejecting the attach (dormant/unknown) rather than a dropped
* stream — retrying it spins in place until the catalog updates, which for
* a dormant session it never will (#168 review). Defaults to `true` so
* existing callers that only track drops of an established link are
* unaffected. */
everOpened = true,
): boolean {
if (code >= APPLICATION_CLOSE || code === CLEAN_CLOSE) return false;
if (!everOpened) return false;
return sessions.get(sessionId)?.status === "Running";
}