A session whose tree a pod cannot put back is parked read-only, once, rather than left active for every client that opens it to fail on¶
Decision 661 · accepted · 2026-09-20
Treating every refused
activation the same way — dormant, and try again next time — is a loop when the
directory is gone, with a raw error for whoever clicked each time. So the worker
names the class: Restore.start failing with {:not_a_directory, path} answers
the plane's session.activate push as not_found with data.reason:
"workspace_gone" (Commands.activation_error/1), and — because the client-driven
path, a pod activating lazily when somebody attaches, never answers a push — the
manager also reports it as a session.unrestorable notification
(Manager.unrestorable_report/2, only for that class; a stale epoch or storage
that did not answer is still the plane's to retry or refuse). On the plane both
roads end in Sessions.unrestorable/2: the row goes read_only, off its worker,
fenced on the epoch like a status report so a pod on an older epoch cannot park a
session a newer one serves, and saying it about a session already parked, erased or
unknown changes nothing. The slot and the budget slice go back as they do at
dormancy. read_only is the honest state — history readable, nothing activates it
again — and it already refuses activation with forbidden before any pod is asked,
so the failure happens once and then never.
Governs: apps/troupe_plane/lib/troupe/plane/control/connection.ex, apps/troupe_plane/lib/troupe/plane/harness.ex, apps/troupe_plane/lib/troupe/plane/sessions.ex, apps/troupe_worker/lib/troupe/worker/plane/commands.ex, apps/troupe_worker/lib/troupe/worker/session/manager.ex, apps/troupe_worker/lib/troupe/worker/session/restore.ex