mirror of
https://github.com/stablyai/orca.git
synced 2026-11-01 09:37:33 +00:00
Brennan Benson
bfc633af8f
* Guard managed server restarts with host work reservations * Protect legacy server update requests with host admission * Reserve chat-safe restarts in SSH and WSL server updates * Keep activation host fixtures within the module line limit * fix(orcad): preserve session queue timing during admission * Explain chat-safe update deferrals and superseded restarts * Clarify managed server Force upgrades and retry instructions * Separate chat entry delegates from conversation creation * Report child work through main's child evidence in the restart drain test * Drive main's snapshot-room tests through the restart reservation Main's room tests deployed over a serving incumbent with a census that has no chat work count and no restart admission, so the update deferred before the room probe ran. Give them the same idle census and slot admission as the other activation tests, and pin the probe before the reservation instead of before the recount it replaced. * fix(native-chat): an update keeps waiting on a chat start after its conversation is closed The restart check counted a start only while its conversation stayed open, so an idle close during a stalled start let a managed update through. Every tracked attach, create, start and open now counts as restart work in the same tracker that counts the session queue. * fix(native-chat): keep the structured session view within its line limit once formatted Main's copy sat at 400 lines only because one line was left unformatted; the merge's formatter pushed it to 401. Inlining the one-use question list keeps it at 400 with identical behavior. * fix(orcad): a recovery stop request stops the server; only an older client's empty request waits for idle Stopping a slot and restarting it for an older client's update were both an empty .orcad-stop-request file, so routing that file through restart admission let a crash recovery, stranded-setup release or rejected-candidate rollback leave the server running. New clients now write "stop-now" into the request (renamed in whole), which stops at once as before; an empty or unrecognized request takes the chat-safe restart, and a deferred restart logs why. Also gives the restart admission activation fixtures the now-required recount. * docs(orcad): say a directory at the stop-request path is refused, not restarted * docs(orcad): the plain stop request's comment states both of its meanings * style: format the restart admission files for the format check * test(wsl): count the incumbent's recount bridge after the reservation's * fix(orcad): a Windows stop request retries its rename while antivirus holds the new file Renaming the just-written stop-now request can briefly fail with EPERM, EACCES or EBUSY on Windows; the stop now retries those with the same bounded delays as record-publish instead of answering SIGNAL_FAILED and leaving the server up. * test(orcad): a real Windows host stops on the slot's stop-now request and relaunches * test(orcad): recovery stops a real server whose restart admission refuses * test(orcad): the recovery-stop cell tunnels over the cell's SSH connection * test(orcad): the recovery-stop cell names itself in the CI log * fix(settings): a deferred WSL update shows its full reason below the row * test(orcad): fixtures pass releaseIncumbentAdmission; the census test calls its own method * test(orcad): the capable-server census in the fence test reports idle chat work * Record server chat restart offers and desktop creation history * Offer interrupted chat recovery per managed runtime * Ship the remote chat resume action labels in the runtime catalog * Update removal contracts for desktop chat history cleanup * Scope remote chat recovery to each server occurrence * Preserve recovery choices and uncertain continuation outcomes * fix: preserve uncertain restart action outcomes * Fix restart recovery CI imports and fixtures * test: align resume assertions with unconfirmed outcomes * test: bound restart recovery work across scaled populations * fix(native-chat): title the remote resume dialog with the restart The dialog can appear out of nowhere when another desktop updates a shared server, so its title now says what happened: "<machine> restarted".