Require a moved OpenClaw service to be running again

Upstream's installer rewrites a loaded gateway onto the new copy but only warns when it will not start, which let the install finish with no gateway running. A service that was running before the move now has to be running after it, or the install fails and says the service is stopped; a failed seed says so too.

Co-Authored-By: Codex XHigh <noreply@openai.com>

bin/omarchy-install-openclaw-cli

@@ -79,22 +79,18 @@ unit_runs() {
 # rewrites a loaded gateway service to the copy it has just installed, so
 # seeding beside one would take it over.
 foreign_gateway() {
-  local unit="$HOME/.config/systemd/user/openclaw-gateway.service"
+  local unit
+  unit=$(unit_path gateway)
   [[ -f $unit ]] && ! unit_runs "$unit" "$prefix/" && ! unit_runs "$unit" "/usr/lib/node_modules/openclaw/"
 }
 
 # A service the openclaw package's own copy installed runs from
 # /usr/lib/node_modules/openclaw, which that package no longer ships, and is
 # moved to the runtime, where `openclaw update` can replace its code. A service
-# running any other OpenClaw is the user's arrangement and stays.
-legacy_units() {
-  local role unit
-  for role in gateway node; do
-    unit="$HOME/.config/systemd/user/openclaw-$role.service"
-    if [[ -f $unit ]] && unit_runs "$unit" "/usr/lib/node_modules/openclaw/"; then
-      echo "$role"
-    fi
-  done
+# running any other OpenClaw is the user's arrangement and stays. The --now
+# steps below list these, and which were running, before anything changes.
+unit_path() {
+  echo "$HOME/.config/systemd/user/openclaw-$1.service"
 }
 
 # Stopped before anything is installed: still running the old code from deleted
@@ -102,7 +98,7 @@ legacy_units() {
 # migrate, which upstream's installer does the moment it finds one loaded.
 stop_legacy_services() {
   local role
-  for role in $(legacy_units); do
+  for role in "${moving[@]}"; do
     if ! systemctl --user stop "openclaw-$role.service"; then
       echo "Could not stop the OpenClaw $role service, so OpenClaw was not moved to $prefix." >&2
       return 1
@@ -110,19 +106,36 @@ stop_legacy_services() {
   done
 }
 
-# Installed again from the runtime, upstream rewrites the unit and starts it,
-# unless its installer already did.
+# Not moved yet: still on the old copy, or running before and not now.
+unsettled() {
+  unit_runs "$(unit_path "$1")" "/usr/lib/node_modules/openclaw/" ||
+    { [[ " ${running[*]} " == *" $1 "* ]] && ! systemctl --user is-active --quiet "openclaw-$1.service"; }
+}
+
+# Upstream's installer rewrites a loaded gateway itself, but only warns when
+# the restart fails, so whatever is still unsettled is installed again from the
+# runtime and then has to have settled.
 rehome_services() {
   local role
-  for role in $(legacy_units); do
-    echo "Moving the OpenClaw $role service to $prefix..."
-    if ! "$runtime_command" "$role" install --force; then
-      echo "Could not move the OpenClaw $role service. Finish with: openclaw $role install --force" >&2
+  for role in "${moving[@]}"; do
+    if unsettled "$role"; then
+      echo "Moving the OpenClaw $role service to $prefix..."
+      "$runtime_command" "$role" install --force || true
+    fi
+    if unsettled "$role"; then
+      echo "Could not move the OpenClaw $role service to $prefix. Finish with: openclaw $role install --force" >&2
       return 1
     fi
   done
 }
 
+# Said whenever a failure leaves a service stopped by this run.
+stopped_note() {
+  if (( ${#running[@]} )); then
+    echo "The OpenClaw ${running[*]} service was stopped for this and is not running now." >&2
+  fi
+}
+
 case "$mode" in
 --check)
   if installed; then exit 0; else exit 1; fi
@@ -167,15 +180,29 @@ if [[ $seeding == "true" && ( ! -r $seed/install-cli.sh || ! -r $seed/openclaw.t
   exit 1
 fi
 
+# What the old package installed and what of it was running, read before
+# anything changes.
+moving=()
+running=()
+for role in gateway node; do
+  if [[ -f $(unit_path "$role") ]] && unit_runs "$(unit_path "$role")" "/usr/lib/node_modules/openclaw/"; then
+    moving+=("$role")
+    if systemctl --user is-active --quiet "openclaw-$role.service"; then
+      running+=("$role")
+    fi
+  fi
+done
+
 stop_legacy_services
 
 if [[ $seeding == "true" ]]; then
   # Every choice the installer reads from the environment is named, so an
   # OPENCLAW_INSTALL_METHOD or OPENCLAW_PREFIX left in a shell cannot move it.
   echo "Setting up OpenClaw in $prefix..."
-  bash "$seed/install-cli.sh" --install-method npm --prefix "$prefix" --version "$seed/openclaw.tgz" --no-onboard
+  bash "$seed/install-cli.sh" --install-method npm --prefix "$prefix" --version "$seed/openclaw.tgz" --no-onboard || true
   if ! runtime_runs; then
     echo "OpenClaw setup did not complete. Re-run this command after resolving the installer error." >&2
+    stopped_note
     exit 1
   fi
 fi
@@ -183,7 +210,10 @@ fi
 mkdir -p "${command_path%/*}"
 ln -sfn "$runtime_command" "$command_path"
 
-rehome_services
+if ! rehome_services; then
+  stopped_note
+  exit 1
+fi
 
 if ! on_path; then
   echo "$runtime_command is ready, but 'openclaw' on PATH is $(type -P openclaw || echo missing), which is what Omarchy runs." >&2

test/shell.d/openclaw-cli-test.sh

@@ -21,16 +21,27 @@ cat >"$mock_bin/omarchy-pkg-add" <<'SH'
 printf 'pkg-add %s\n' "$*" >>"$OMARCHY_TEST_ROOT/events"
 touch "$OMARCHY_TEST_ROOT/package-installed"
 SH
+# The user manager, as far as these tests need one: a service is active while
+# a marker says so. Stopping clears it; OMARCHY_TEST_STOP_FAIL refuses.
 cat >"$mock_bin/systemctl" <<'SH'
 #!/bin/bash
-printf 'systemctl %s\n' "$*" >>"$OMARCHY_TEST_ROOT/events"
-[[ -z ${OMARCHY_TEST_STOP_FAIL:-} ]]
+case "$2" in
+  stop)
+    printf 'systemctl %s\n' "$*" >>"$OMARCHY_TEST_ROOT/events"
+    [[ -z ${OMARCHY_TEST_STOP_FAIL:-} ]] || exit 1
+    rm -f "$HOME/active-$3"
+    ;;
+  is-active) [[ -e $HOME/active-$4 ]] ;;
+esac
 SH
 chmod +x "$mock_bin/"*
 
 # Stands in for upstream's install-cli.sh: it writes the command the way the
 # real one does, execing into the prefix's tools, and that command logs what
 # it is asked. OMARCHY_TEST_INSTALL_BROKEN leaves a command that cannot run.
+# Like upstream, `<role> install --force` rewrites the unit onto the runtime
+# and starts it, OMARCHY_TEST_START_FAIL making the start fail, and the
+# installer does that itself for a gateway it finds loaded.
 cat >"$seed/install-cli.sh" <<'SH'
 printf 'install-cli %s%s\n' "$*" "${OPENCLAW_PROFILE:+ profile=$OPENCLAW_PROFILE}" >>"$OMARCHY_TEST_ROOT/events"
 prefix=$HOME/.openclaw
@@ -38,11 +49,19 @@ mkdir -p "$prefix/bin" "$prefix/tools/node-v24.19.0"
 cat >"$prefix/bin/openclaw" <<EOF
 #!/usr/bin/env bash
 set -euo pipefail
-[[ -z "${OMARCHY_TEST_INSTALL_BROKEN:-}" ]] || exit 1
+[[ -z "\${OMARCHY_TEST_INSTALL_BROKEN:-}" ]] || exit 1
 printf 'runtime %s%s\n' "\$*" "\${OPENCLAW_PROFILE:+ profile=\$OPENCLAW_PROFILE}" >>"$OMARCHY_TEST_ROOT/events"
+if [[ \${2:-} == "install" ]]; then
+  printf 'ExecStart=$prefix/tools/node-v24.19.0/bin/node $prefix/tools/node-v24.19.0/lib/node_modules/openclaw/dist/index.js %s\n' "\$1" >"\$HOME/.config/systemd/user/openclaw-\$1.service"
+  [[ -z "\${OMARCHY_TEST_START_FAIL:-}" ]] || exit 1
+  touch "\$HOME/active-openclaw-\$1.service"
+fi
 exec true "$prefix/tools/node-v24.19.0/lib/node_modules/openclaw/dist/entry.js" "\$@"
 EOF
 chmod 755 "$prefix/bin/openclaw"
+if [[ -f $HOME/.config/systemd/user/openclaw-gateway.service ]]; then
+  "$prefix/bin/openclaw" gateway install --force || true
+fi
 SH
 touch "$seed/openclaw.tgz"
 
@@ -180,6 +199,29 @@ grep -q "Could not stop the OpenClaw gateway service" "$test_tmp/output" || fail
   fail "a gateway that will not stop leaves the runtime unseeded" "$(cat "$events")"
 pass "a gateway that will not stop stops the install before anything is seeded"
 
+# Upstream's installer rewrites the gateway itself and only warns when it will
+# not start again, so a gateway that was running has to be running afterwards.
+new_home start-fails
+mkdir -p "$test_home/.config/systemd/user"
+printf 'ExecStart=/usr/bin/node /usr/lib/node_modules/openclaw/dist/index.js gateway --port 18789\n' >"$test_home/.config/systemd/user/openclaw-gateway.service"
+touch "$test_home/active-openclaw-gateway.service"
+OMARCHY_TEST_START_FAIL=1 run omarchy-install-openclaw-cli --now && fail "a moved gateway that does not start fails the install"
+grep -q "Could not move the OpenClaw gateway service" "$test_tmp/output" && grep -q "is not running now" "$test_tmp/output" ||
+  fail "a moved gateway that does not start is named, and so is its being stopped" "$(cat "$test_tmp/output")"
+pass "a gateway that was running is running again from the runtime, or the install fails saying it is stopped"
+
+# With the runtime already in place nothing is seeded, so the move is Omarchy's.
+new_home runtime-first
+run omarchy-install-openclaw-cli --now || fail "--now sets OpenClaw up" "$(cat "$test_tmp/output")"
+mkdir -p "$test_home/.config/systemd/user"
+printf 'ExecStart=/usr/bin/node /usr/lib/node_modules/openclaw/dist/index.js gateway --port 18789\n' >"$test_home/.config/systemd/user/openclaw-gateway.service"
+touch "$test_home/active-openclaw-gateway.service"
+: >"$events"
+run omarchy-install-openclaw-cli --now || fail "--now moves a gateway beside a runtime that already runs" "$(cat "$test_tmp/output")"
+! grep -q '^install-cli' "$events" && grep -Fxq "runtime gateway install --force" "$events" && [[ -e $test_home/active-openclaw-gateway.service ]] ||
+  fail "--now moves a gateway beside a runtime that already runs" "$(cat "$events")"
+pass "a gateway beside a runtime that already runs is moved without seeding"
+
 # The migration moves only machines that have the package, and waits for the
 # package that seeds.
 new_home migration-none