claude-multiacc 2.0.25 → 2.0.26

package/bin/claude-accounts

@@ -1434,7 +1434,21 @@ limits_distribute() {
   sync_is_replica && return 0
   command -v rsync >/dev/null 2>&1 || return 0
   [ "${CLAUDE_MULTIACC_NO_DISTRIBUTE:-0}" = "1" ] && return 0
-  ( limits_distribute_now >/dev/null 2>&1 & ) >/dev/null 2>&1
+  # In its OWN session, not merely backgrounded. launchd tears down the job's
+  # whole process group the moment `limits` exits, and a `( … & )` subshell is
+  # still in that group: every scheduled pass spawned its push and launchd killed
+  # it before one rsync had finished, with nothing logged — while this machine
+  # refreshed every 15 minutes, the seven runner Macs ranked on readings last
+  # pushed at 21:13Z until 01:10Z on 2026-09-09 (a foreground pass delivered them
+  # in seconds). setsid puts the push outside the group; install.sh also marks
+  # the agent AbandonProcessGroup, and either one alone is enough.
+  "$PYBIN" -c '
+import os, sys
+if os.fork():
+    sys.exit(0)
+os.setsid()
+os.execv(sys.argv[1], [sys.argv[1], "limits-distribute-now"])
+' "$BIN_DIR/claude-accounts" >/dev/null 2>&1 </dev/null &
   return 0
 }
 
@@ -1472,21 +1486,26 @@ limits_distribute_now() {
   done
   if [ ! -s "$list" ]; then rm -f "$list"; return 0; fi
 
+  local pushed=0 failed=0
   server="$(sync_target)"
   sroot="$(sync_target_root)"
   if [ -n "$server" ] && ! sync_target_is_local "$server" \
      && valid_ssh_target "$server" && valid_remote_path "$sroot"; then
-    limits_push_to "$server" "$sroot" "$list"
+    if limits_push_to "$server" "$sroot" "$list"; then pushed=$((pushed + 1)); else failed=$((failed + 1)); fi
   fi
   # Peers are the OTHER Macs — the ones actually running tasks, and so the ones
-  # whose selection goes blind without this.
-  manifest_peers | while IFS="$(printf '\t')" read -r pt pr pp; do
+  # whose selection goes blind without this. Read through process substitution,
+  # not a pipe: the counters below must survive the loop.
+  while IFS="$(printf '\t')" read -r pt pr pp; do
     [ "$pt" = "MALFORMED" ] && continue
     [ -n "$pt" ] && [ -n "$pr" ] || continue
     valid_ssh_target "$pt" || continue
     valid_remote_path "$pr" || continue
-    limits_push_to "$pt" "$pr" "$list"
-  done
+    if limits_push_to "$pt" "$pr" "$list"; then pushed=$((pushed + 1)); else failed=$((failed + 1)); fi
+  done < <(manifest_peers)
+  # Every pass writes its outcome: a silent push is indistinguishable from a
+  # killed one, which is exactly how the 2026-09-08 outage went unnoticed.
+  log_to sync.log "limits distributed to $pushed target(s), $failed failed"
   rm -f "$list"
   rmdir "$lock" 2>/dev/null || true
   trap - EXIT
@@ -1497,8 +1516,7 @@ limits_push_to() { # $1 target, $2 remote root, $3 file list
   rsync -az --timeout=20 \
     -e 'ssh -o BatchMode=yes -o ConnectTimeout=10 -o StrictHostKeyChecking=accept-new' \
     --files-from="$3" "$ACC_ROOT/" "$1:$2/" >>"$ACC_ROOT/sync.log" 2>&1 \
-    || log_to sync.log "limits push to $1 failed (telemetry there will age)"
-  return 0
+    || { log_to sync.log "limits push to $1 failed (telemetry there will age)"; return 1; }
 }
 
 cmd_limits() {
@@ -2956,6 +2974,8 @@ case "${1:-help}" in
   sync) shift; cmd_sync "$@" ;;
   verify) shift; cmd_verify "$@" ;;
   limits) shift; cmd_limits "$@" ;;
+  # Internal: the detached telemetry push `limits` starts in its own session.
+  limits-distribute-now) shift; limits_distribute_now ;;
   post-sync) shift; cmd_post_sync "$@" ;;
   health) shift; cmd_health "$@" ;;
   self-update) shift; cmd_self_update "$@" ;;

package/install.sh

@@ -183,6 +183,9 @@ mac_schedule_install() {
   </array>
 $(plist_env_block)  <key>StartInterval</key><integer>900</integer>
   <key>RunAtLoad</key><true/>
+  <!-- The pass hands its readings to the peers from a detached push; launchd
+       must not kill that push with the job's process group. -->
+  <key>AbandonProcessGroup</key><true/>
   <key>StandardOutPath</key><string>/dev/null</string>
   <key>StandardErrorPath</key><string>/dev/null</string>
 </dict></plist>
@@ -238,6 +241,7 @@ EOF
   </array>
 $(plist_env_block)  <key>StartInterval</key><integer>300</integer>
   <key>RunAtLoad</key><true/>
+  <key>AbandonProcessGroup</key><true/>
   <key>StandardOutPath</key><string>/dev/null</string>
   <key>StandardErrorPath</key><string>/dev/null</string>
 </dict></plist>

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "claude-multiacc",
-  "version": "2.0.25",
+  "version": "2.0.26",
   "description": "Unified Claude Code and OpenAI Codex subscription pooling with quota-aware selection.",
   "type": "module",
   "bin": {

package/tests/run-tests.sh

@@ -3759,6 +3759,7 @@ mkdir -p "$RSFAKE"
 cat > "$RSFAKE/rsync" <<'EOF'
 #!/usr/bin/env bash
 printf 'RSYNC %s\n' "$*" >> "${RSLOG:?}"
+printf 'PGID %s\n' "$(ps -o pgid= -p $$ | tr -d ' ')" >> "$RSLOG"
 for a in "$@"; do
   case "$a" in
     --files-from=*)
@@ -3795,6 +3796,23 @@ for peer in gas@mini-3 gas@mini-4; do
     && t_ok "telemetry is pushed to manifest peer $peer" \
     || t_fail "limits distribute peer" "$peer never received a push: $(grep '^RSYNC' "$RSLOG")"
 done
+# The push must run OUTSIDE the calling job's process group: launchd kills that group
+# the moment `limits` exits, and a `( … & )` subshell is still in it — on 2026-09-08 the
+# source Mac refreshed every 15 minutes while every scheduled push died before its first
+# rsync, and the seven runner Macs ranked on 21:13Z readings until 01:10Z the next day.
+own_pgid="$(ps -o pgid= -p $$ | tr -d ' ')"
+push_pgids="$(grep '^PGID ' "$RSLOG" | sort -u | sed 's/^PGID //' | tr '\n' ' ')"
+case " $push_pgids " in
+  *" $own_pgid "*) t_fail "detached push session" "a push ran inside the caller's process group $own_pgid (launchd would kill it)" ;;
+  *) [ -n "$push_pgids" ] && t_ok "the telemetry push runs in its own session, outside the job's process group" \
+       || t_fail "detached push session" "no push recorded a process group" ;;
+esac
+grep -q "limits distributed to 3 target(s), 0 failed" "$DPOOL/sync.log" \
+  && t_ok "every distribute pass records its outcome in sync.log" \
+  || t_fail "distribute summary" "sync.log: $(tail -3 "$DPOOL/sync.log" 2>/dev/null | tr '\n' '|')"
+[ "$(grep -c '<key>AbandonProcessGroup</key><true/>' "$REPO_DIR/install.sh")" = "2" ] \
+  && t_ok "install.sh abandons the process group of both limits agents" \
+  || t_fail "AbandonProcessGroup" "expected both limits plists to carry AbandonProcessGroup"
 # ...and every target gets the SAME list: each account's reading, plus any marker.
 for f in "acct-01/limits.json" "acct-02/limits.json" "acct-02/.limited"; do
   [ "$(grep -c "^FILE $f\$" "$RSLOG")" = "3" ] \