reconnectSecondary fail-stopped only on "connection removed" and "profile
deleted"; every other dial failure fell through to scheduleReconnect with a
15s-capped jitter and no budget. For a tile-pinned owner scope that is
forever wantOpen, a pool-slot wait timeout therefore re-issued a background
spawn on every backoff tick, for every such tile, indefinitely -- the
"waiting for a free local slot ... timed out ... will retry" cycle in
#103375's desktop.log, in the same profile order each pass.
Count only STALLED dials (a slot-wait timeout or the 20s dial timeout):
those are the ones that hold a queue position and re-queue a spawn. After
SECONDARY_STALLED_DIAL_BUDGET consecutive stalls the entry parks
(wantOpen=false) with a console warning. Fast failures (ECONNREFUSED while
a gateway restarts) keep the ordinary unbounded backoff, so a socket that
merely dropped still comes back on its own.
Parked is not evicted: every explicit open of the scope (requestGatewayForAgent,
openGatewayForAgent, ensureGatewayForAgent, and ensureActiveGatewayOpen for
the viewed scope -- the Reconnect action and request retries) goes through
rearmSecondary, which sets wantOpen and resets the budget.
Not changed: resumeTile's non-passive transcript prefetch. It shares the
main-process dial claim with the session.resume dial for the same scope, so
it never adds a spawn, and a cold tile needs it to paint (resume runs with
omit_messages).
kshitijk4poor
dc90a75ab4