@fastybird/smart-panel-backend 1.1.0-alpha.15 → 1.1.0-alpha.17

package/dist/modules/system/scripts/update-worker.sh

@@ -7,6 +7,328 @@ INSTALL_TYPE="${INSTALL_TYPE:-npm}"
 IMAGE_BASE_DIR="${IMAGE_BASE_DIR:-/opt/smart-panel}"
 DOWNLOAD_URL="${DOWNLOAD_URL:-}"
 
+# The attempt record and lock live beside the public status file, never inside a versioned image
+# directory.  This lets the record survive a backend restart, target-directory cleanup or a
+# failed migration, while keeping it outside the release tarball and private to the service user.
+ATTEMPT_DIR="${ATTEMPT_DIR:-$(dirname "$STATUS_FILE")/update-attempt}"
+ATTEMPT_FILE="${ATTEMPT_DIR}/attempt.json"
+ATTEMPT_LOCK="${ATTEMPT_DIR}/lock"
+ATTEMPT_ID=""
+MIGRATION_ENTERED="false"
+FINALIZED="false"
+MIGRATION_LOG=""
+CURRENT_PHASE="preparing"
+SERVICE_NAME="smart-panel"
+QUIESCENCE_TIMEOUT_SECONDS="${QUIESCENCE_TIMEOUT_SECONDS:-30}"
+START_TIMEOUT_SECONDS="${START_TIMEOUT_SECONDS:-60}"
+HEALTH_URL="${HEALTH_URL:-}"
+CAPTURED_SERVICE_CGROUP_PROCS=""
+CAPTURED_SERVICE_MEMBER_IDENTITIES=""
+
+json_escape() {
+	printf '%s' "$1" | sed 's/\\/\\\\/g; s/"/\\"/g; s/[[:cntrl:]]//g'
+}
+
+write_attempt() {
+	local state="$1"
+	local phase="$2"
+	local error="${3:-}"
+	local recovery_required="${4:-false}"
+	local migration_entered="${5:-$MIGRATION_ENTERED}"
+	local tmp_file="${ATTEMPT_FILE}.tmp.$$"
+	CURRENT_PHASE="$phase"
+
+	mkdir -p "$ATTEMPT_DIR"
+	chmod 700 "$ATTEMPT_DIR"
+
+	{
+		printf '{\n'
+		printf '\t"attemptId": "%s",\n' "$(json_escape "$ATTEMPT_ID")"
+		printf '\t"ownerPid": %s,\n' "$$"
+		printf '\t"targetVersion": "%s",\n' "$(json_escape "$VERSION")"
+		printf '\t"state": "%s",\n' "$(json_escape "$state")"
+		printf '\t"phase": "%s",\n' "$(json_escape "$phase")"
+		printf '\t"migrationEntered": %s,\n' "$migration_entered"
+		printf '\t"recoveryRequired": %s,\n' "$recovery_required"
+		printf '\t"migrationLog": "%s",\n' "$(json_escape "$MIGRATION_LOG")"
+		printf '\t"updatedAt": "%s",\n' "$(date -Iseconds)"
+		printf '\t"error": "%s"\n' "$(json_escape "$error")"
+		printf '}\n'
+	} > "$tmp_file"
+
+	chmod 600 "$tmp_file"
+	mv -f "$tmp_file" "$ATTEMPT_FILE"
+}
+
+attempt_is_unresolved() {
+	if [ -f "$ATTEMPT_FILE" ] && grep -q '"state": "complete"' "$ATTEMPT_FILE"; then
+		# A crash after the durable completion record but before rmdir must not strand a clean
+		# attempt forever; completion is the evidence that makes this lock removable.
+		rmdir "$ATTEMPT_LOCK" 2>/dev/null || true
+
+		return 1
+	fi
+
+	if [ -d "$ATTEMPT_LOCK" ]; then
+		return 0
+	fi
+
+	if [ ! -f "$ATTEMPT_FILE" ]; then
+		return 1
+	fi
+
+	grep -q '"state": "active"' "$ATTEMPT_FILE" || grep -q '"recoveryRequired": true' "$ATTEMPT_FILE"
+}
+
+acquire_attempt() {
+	mkdir -p "$ATTEMPT_DIR"
+	chmod 700 "$ATTEMPT_DIR"
+
+	if attempt_is_unresolved; then
+		return 1
+	fi
+
+	# mkdir is the portable atomic OS-level lock used by the legacy launcher.  An orphaned lock is
+	# deliberately retained; an expired timestamp cannot prove that a migration did not commit.
+	if ! mkdir "$ATTEMPT_LOCK" 2>/dev/null; then
+		return 1
+	fi
+	chmod 700 "$ATTEMPT_LOCK"
+
+	ATTEMPT_ID="$(date +%s)-$$"
+	write_attempt "active" "preparing" "" "false" "false"
+}
+
+release_attempt() {
+	if [ -n "$ATTEMPT_ID" ]; then
+		write_attempt "complete" "complete" "" "false" "$MIGRATION_ENTERED"
+	fi
+
+	rmdir "$ATTEMPT_LOCK" 2>/dev/null || true
+}
+
+hold_attempt() {
+	local phase="$1"
+	local error="$2"
+
+	write_attempt "recovery_required" "$phase" "$error" "true" "$MIGRATION_ENTERED"
+}
+
+finish_before_migration() {
+	local phase="$1"
+	local error="$2"
+
+	if [ -n "${TMP_TARBALL:-}" ]; then
+		rm -f "$TMP_TARBALL" 2>/dev/null || true
+	fi
+
+	write_attempt "failed_preflight" "$phase" "$error" "false" "false"
+	rmdir "$ATTEMPT_LOCK" 2>/dev/null || true
+}
+
+phase_requires_hold() {
+	case "$CURRENT_PHASE" in
+		stopping|quiesced|switching) return 0 ;;
+		*) return 1 ;;
+	esac
+}
+
+fail_before_migration() {
+	local phase="$1"
+	local error="$2"
+
+	if [ -n "$ATTEMPT_ID" ]; then
+		finish_before_migration "$phase" "$error"
+	fi
+
+	update_status "failed" "failed" "$error"
+	FINALIZED="true"
+	exit 1
+}
+
+restore_before_migration() {
+	local previous_target="$1"
+	local current_link="$2"
+	local new_version_dir="$3"
+	local resolved_previous resolved_current
+
+	[ -n "$previous_target" ] || return 1
+	sudo -n ln -sfn "$previous_target" "$current_link" || return 1
+
+	resolved_previous="$(realpath "$previous_target" 2>/dev/null || true)"
+	resolved_current="$(realpath "$current_link" 2>/dev/null || true)"
+	[ -n "$resolved_previous" ] && [ "$resolved_previous" = "$resolved_current" ] || return 1
+
+	rm -rf "$new_version_dir" || return 1
+	sudo -n systemctl start "$SERVICE_NAME" 2>/dev/null || return 1
+	wait_for_started_service
+}
+
+service_property() {
+	local property="$1"
+
+	systemctl show "$SERVICE_NAME" --property="$property" --value
+}
+
+process_identity() {
+	local pid="$1"
+	local proc_root="${SERVICE_PROCESS_ROOT:-/proc}"
+	local proc_dir="${proc_root}/${pid}"
+	local stat_line start_time executable command
+
+	[ "$pid" != "0" ] && [ -n "$pid" ] || return 1
+
+	# SERVICE_PROCESS_ROOT is a private fixture hook; the production default is /proc.
+	if [ -r "${proc_dir}/identity" ]; then
+		printf '%s|%s\n' "$pid" "$(cat "${proc_dir}/identity")"
+		return 0
+	fi
+
+	if [ -r "${proc_dir}/stat" ]; then
+		stat_line="$(cat "${proc_dir}/stat")" || return 2
+		start_time="$(printf '%s\n' "$stat_line" | awk '{ sub(/^.*\\) /, ""); print $20 }')" || return 2
+		[ -n "$start_time" ] || return 2
+		executable="$(readlink "${proc_dir}/exe" 2>/dev/null)" || return 2
+		[ -n "$executable" ] || return 2
+		printf '%s|%s|%s\n' "$pid" "$start_time" "$executable"
+		return 0
+	fi
+
+	if ! kill -0 "$pid" 2>/dev/null; then
+		return 1
+	fi
+
+	command="$(ps -p "$pid" -o command= 2>/dev/null)" || return 2
+	[ -n "$command" ] || return 2
+	printf '%s|%s\n' "$pid" "$command"
+}
+
+capture_service_identity() {
+	local active_state main_pid control_pid control_group cgroup_procs members pid identity
+
+	active_state="$(service_property ActiveState 2>/dev/null)" || return 2
+	main_pid="$(service_property MainPID 2>/dev/null)" || return 2
+	control_pid="$(service_property ControlPID 2>/dev/null)" || return 2
+	control_group="$(service_property ControlGroup 2>/dev/null)" || return 2
+
+	case "$active_state" in
+		active|activating) ;;
+		*) return 1 ;;
+	esac
+
+	[ "$main_pid" -gt 0 ] 2>/dev/null || return 1
+	[ "$control_pid" = "0" ] || [ "$control_pid" -gt 0 ] 2>/dev/null || return 2
+	[ -n "$control_group" ] || return 2
+
+	cgroup_procs="${SERVICE_CGROUP_PROCS_FILE:-/sys/fs/cgroup${control_group}/cgroup.procs}"
+	[ -r "$cgroup_procs" ] || return 2
+	members="$(grep -E '^[0-9]+$' "$cgroup_procs" 2>/dev/null)" || true
+	[ -n "$members" ] || return 1
+
+	CAPTURED_SERVICE_MEMBER_IDENTITIES=""
+	while IFS= read -r pid; do
+		[ -n "$pid" ] || continue
+		identity="$(process_identity "$pid")" || return 2
+		CAPTURED_SERVICE_MEMBER_IDENTITIES="${CAPTURED_SERVICE_MEMBER_IDENTITIES}${identity}\n"
+	done <<EOF
+$members
+EOF
+
+	printf '%s\n' "$members" | grep -qx "$main_pid" || return 1
+	CAPTURED_SERVICE_CGROUP_PROCS="$cgroup_procs"
+}
+
+service_quiesced() {
+	local active_state main_pid control_pid cgroup_contents current_members pid captured_identity current_identity identity_status
+
+	[ -n "$CAPTURED_SERVICE_CGROUP_PROCS" ] || return 2
+	active_state="$(service_property ActiveState 2>/dev/null)" || return 2
+	main_pid="$(service_property MainPID 2>/dev/null)" || return 2
+	control_pid="$(service_property ControlPID 2>/dev/null)" || return 2
+
+	case "$active_state" in
+		inactive|failed) ;;
+		*) return 1 ;;
+	esac
+
+	[ "$main_pid" = "0" ] || return 1
+	[ "$control_pid" = "0" ] || return 1
+
+	# Check every member captured before stop independently. This catches KillMode=process
+	# survivors even when systemd has already removed the original cgroup directory.
+	while IFS= read -r captured_identity; do
+		[ -n "$captured_identity" ] || continue
+		pid="${captured_identity%%|*}"
+		current_identity=""
+		if current_identity="$(process_identity "$pid")"; then
+			identity_status=0
+		else
+			identity_status=$?
+		fi
+		case "$identity_status" in
+			0) [ "$current_identity" != "$captured_identity" ] || return 1 ;;
+			1) ;;
+			*) return 2 ;;
+		esac
+	done <<EOF
+$(printf '%b' "$CAPTURED_SERVICE_MEMBER_IDENTITIES")
+EOF
+
+	# The original cgroup may be empty or removed after the service reaches inactive/failed with
+	# zero MainPID/ControlPID. If it still exists, it must be readable and contain no members.
+	if [ -e "$CAPTURED_SERVICE_CGROUP_PROCS" ]; then
+		[ -r "$CAPTURED_SERVICE_CGROUP_PROCS" ] || return 2
+		if ! cgroup_contents="$(cat "$CAPTURED_SERVICE_CGROUP_PROCS" 2>/dev/null)"; then
+			[ ! -e "$CAPTURED_SERVICE_CGROUP_PROCS" ] || return 2
+			cgroup_contents=""
+		fi
+		current_members="$(printf '%s\n' "$cgroup_contents" | grep -E '^[0-9]+$')" || true
+		[ -z "$current_members" ] || return 1
+	fi
+
+	return 0
+}
+
+wait_for_quiescence() {
+	local deadline=$((SECONDS + QUIESCENCE_TIMEOUT_SECONDS))
+	local result
+
+	while [ "$SECONDS" -le "$deadline" ]; do
+		if service_quiesced; then
+			return 0
+		else
+			result=$?
+		fi
+		if [ "$result" -eq 2 ]; then
+			return 2
+		fi
+
+		sleep 1
+	done
+
+	return 1
+}
+
+wait_for_started_service() {
+	local deadline=$((SECONDS + START_TIMEOUT_SECONDS))
+	local state main_pid
+
+	while [ "$SECONDS" -le "$deadline" ]; do
+		state="$(service_property ActiveState 2>/dev/null || true)"
+		main_pid="$(service_property MainPID 2>/dev/null || true)"
+
+		if [ "$state" = "active" ] && [ "$main_pid" != "0" ] && [ -n "$main_pid" ]; then
+			if [ -z "$HEALTH_URL" ] || curl -fsS --max-time 5 "$HEALTH_URL" >/dev/null 2>&1; then
+				return 0
+			fi
+		fi
+
+		sleep 1
+	done
+
+	return 1
+}
+
 # Capture once — used for all status writes so timeout detection works correctly
 STARTED_AT="$(date -Iseconds)"
 
@@ -45,7 +367,13 @@ update_status() {
 cleanup() {
 	local exit_code=$?
 
-	if [ $exit_code -ne 0 ]; then
+	if [ "$exit_code" -ne 0 ] && [ "$FINALIZED" != "true" ]; then
+		if [ -n "$ATTEMPT_ID" ] && { [ "$MIGRATION_ENTERED" = "true" ] || phase_requires_hold; }; then
+				hold_attempt "interrupted" "Update worker exited with code $exit_code during a state-changing phase"
+		elif [ -n "$ATTEMPT_ID" ]; then
+			finish_before_migration "failed" "Update process exited with code $exit_code"
+		fi
+
 		# Only write a generic failure if a specific error wasn't already recorded
 		if [ -f "$STATUS_FILE" ] && grep -q '"status": "failed"' "$STATUS_FILE" 2>/dev/null; then
 			return
@@ -56,6 +384,27 @@ cleanup() {
 
 trap cleanup EXIT
 
+handle_signal() {
+	local signal_name="$1"
+	local signal_code="$2"
+
+	if [ -n "$ATTEMPT_ID" ]; then
+		if [ "$MIGRATION_ENTERED" = "true" ] || phase_requires_hold; then
+			hold_attempt "interrupted" "Update worker received SIG${signal_name} after migration entry"
+		else
+			finish_before_migration "interrupted" "Update worker received SIG${signal_name} before migration entry"
+		fi
+	fi
+
+	update_status "failed" "failed" "Update worker received SIG${signal_name}"
+	FINALIZED="true"
+	exit "$signal_code"
+}
+
+trap 'handle_signal TERM 143' TERM
+trap 'handle_signal INT 130' INT
+trap 'handle_signal HUP 129' HUP
+
 # ──────────────────────────────────────────────────────────────
 # Image-based update (Raspbian image installs)
 # ──────────────────────────────────────────────────────────────
@@ -65,6 +414,12 @@ if [ "$INSTALL_TYPE" = "image" ]; then
 	NEW_VERSION_DIR="${IMAGE_BASE_DIR}/v${CLEAN_VERSION}"
 	CURRENT_LINK="${IMAGE_BASE_DIR}/current"
 
+	if ! acquire_attempt; then
+		update_status "failed" "failed" "An unresolved update attempt already owns the updater"
+		FINALIZED="true"
+		exit 1
+	fi
+
 	# Verify passwordless sudo is available for all required commands.
 	# Without -n, sudo may hang waiting for a password on a detached process.
 	SUDO_CMDS=(
@@ -77,8 +432,7 @@ if [ "$INSTALL_TYPE" = "image" ]; then
 
 	for cmd in "${SUDO_CMDS[@]}"; do
 		if ! sudo -n -l $cmd >/dev/null 2>&1; then
-			update_status "failed" "failed" "Passwordless sudo not available for: $cmd"
-			exit 1
+			fail_before_migration "preflight" "Passwordless sudo not available for: $cmd"
 		fi
 	done
 	PREVIOUS_TARGET=""
@@ -86,6 +440,10 @@ if [ "$INSTALL_TYPE" = "image" ]; then
 	# Save the current version for rollback
 	if [ -L "$CURRENT_LINK" ]; then
 		PREVIOUS_TARGET=$(readlink "$CURRENT_LINK")
+		case "$PREVIOUS_TARGET" in
+			/*) ;;
+			*) PREVIOUS_TARGET="${IMAGE_BASE_DIR}/${PREVIOUS_TARGET}" ;;
+		esac
 	fi
 
 	# Guard: refuse to overwrite the currently running version
@@ -94,23 +452,20 @@ if [ "$INSTALL_TYPE" = "image" ]; then
 	RESOLVED_NEW=$(cd "$IMAGE_BASE_DIR" && realpath "$NEW_VERSION_DIR" 2>/dev/null || true)
 
 	if [ -n "$PREVIOUS_TARGET" ] && [ -n "$RESOLVED_PREV" ] && [ -n "$RESOLVED_NEW" ] && [ "$RESOLVED_PREV" = "$RESOLVED_NEW" ]; then
-		update_status "failed" "failed" "Target version v${CLEAN_VERSION} is already the active version"
-		exit 1
+		fail_before_migration "preflight" "Target version v${CLEAN_VERSION} is already the active version"
 	fi
 
 	# ── Download ──
 	update_status "downloading" "downloading"
 
 	if [ -z "$DOWNLOAD_URL" ]; then
-		update_status "failed" "failed" "No download URL provided for image update"
-		exit 1
+		fail_before_migration "downloading" "No download URL provided for image update"
 	fi
 
 	TMP_TARBALL="/tmp/smart-panel-backend-v${CLEAN_VERSION}.tar.gz"
 
 	curl -fSL -o "$TMP_TARBALL" "$DOWNLOAD_URL" 2>&1 || {
-		update_status "failed" "failed" "Download failed from ${DOWNLOAD_URL}"
-		exit 1
+		fail_before_migration "downloading" "Download failed from ${DOWNLOAD_URL}"
 	}
 
 	# ── Extract ──
@@ -121,8 +476,7 @@ if [ "$INSTALL_TYPE" = "image" ]; then
 	tar -xzf "$TMP_TARBALL" -C "$NEW_VERSION_DIR" 2>&1 || {
 		rm -rf "$NEW_VERSION_DIR"
 		rm -f "$TMP_TARBALL"
-		update_status "failed" "failed" "Failed to extract update archive"
-		exit 1
+		fail_before_migration "installing" "Failed to extract update archive"
 	}
 
 	rm -f "$TMP_TARBALL"
@@ -130,8 +484,7 @@ if [ "$INSTALL_TYPE" = "image" ]; then
 	# Create the image-install marker in the new version
 	touch "${NEW_VERSION_DIR}/.image-install" || {
 		rm -rf "$NEW_VERSION_DIR"
-		update_status "failed" "failed" "Failed to create image-install marker"
-		exit 1
+		fail_before_migration "installing" "Failed to create image-install marker"
 	}
 
 	# ── Verify dependencies ──
@@ -143,8 +496,7 @@ if [ "$INSTALL_TYPE" = "image" ]; then
 	if [ ! -d "node_modules" ]; then
 		npm install --omit=dev 2>&1 || {
 			rm -rf "$NEW_VERSION_DIR"
-			update_status "failed" "failed" "npm install failed"
-			exit 1
+			fail_before_migration "installing" "npm install failed"
 		}
 	fi
 
@@ -153,72 +505,122 @@ if [ "$INSTALL_TYPE" = "image" ]; then
 	# hanging on password prompt when running as a detached process.
 	sudo -n chown -R smart-panel:smart-panel "$NEW_VERSION_DIR" || {
 		rm -rf "$NEW_VERSION_DIR"
-		update_status "failed" "failed" "Failed to set ownership on ${NEW_VERSION_DIR}"
-		exit 1
+		fail_before_migration "installing" "Failed to set ownership on ${NEW_VERSION_DIR}"
 	}
 
 	# ── Stop service ──
 	update_status "stopping" "stopping"
-	sudo -n systemctl stop smart-panel 2>/dev/null || true
+	write_attempt "active" "stopping" "" "false" "$MIGRATION_ENTERED"
+	if ! capture_service_identity; then
+		hold_attempt "stopping" "Could not capture service process/cgroup identity before stop"
+		update_status "failed" "failed" "Service process/cgroup identity could not be captured; update is held for recovery"
+		FINALIZED="true"
+		exit 1
+	fi
+
+	if ! sudo -n systemctl stop "$SERVICE_NAME" 2>/dev/null; then
+		hold_attempt "stopping" "Service stop failed; writer quiescence is unknown"
+		update_status "failed" "failed" "Could not stop ${SERVICE_NAME}; update is held for recovery"
+		FINALIZED="true"
+		exit 1
+	fi
+
+	if ! wait_for_quiescence; then
+		hold_attempt "stopping" "Service did not reach verified quiescence before migration"
+		update_status "failed" "failed" "Service quiescence could not be verified; update is held for recovery"
+		FINALIZED="true"
+		exit 1
+	fi
+
+	write_attempt "active" "quiesced" "" "false" "$MIGRATION_ENTERED"
 
 	# ── Switch symlink (atomic on same filesystem) ──
 	sudo -n ln -sfn "$NEW_VERSION_DIR" "$CURRENT_LINK" || {
-		# Revert on failure
-		if [ -n "$PREVIOUS_TARGET" ]; then
-			sudo -n ln -sfn "$PREVIOUS_TARGET" "$CURRENT_LINK"
+		# Revert only when the old link, target cleanup and old-service health are all verified.
+		if restore_before_migration "$PREVIOUS_TARGET" "$CURRENT_LINK" "$NEW_VERSION_DIR"; then
+			fail_before_migration "switching" "Failed to switch version symlink"
 		fi
-		sudo -n systemctl start smart-panel 2>/dev/null || true
-		update_status "failed" "failed" "Failed to switch version symlink"
+
+		hold_attempt "switching" "Failed to switch version symlink and recovery start was not verified"
+		update_status "failed" "failed" "Failed to switch version symlink; update is held for recovery"
+		FINALIZED="true"
 		exit 1
 	}
 
 	# ── Run database migrations ──
 	update_status "migrating" "migrating"
-
 	ENV_FILE="/etc/smart-panel/environment"
 
-	if [ -f "${NEW_VERSION_DIR}/dist/dataSource.js" ]; then
-		(
-			set -a
-			# shellcheck source=/dev/null
-			[ -f "$ENV_FILE" ] && . "$ENV_FILE"
-			set +a
-			cd "$NEW_VERSION_DIR"
-			node node_modules/typeorm/cli.js migration:run -d dist/dataSource.js
-		) 2>&1 || {
-			# Migration failed — revert symlink if possible
-			update_status "stopping" "stopping"
-			sudo -n systemctl stop smart-panel 2>/dev/null || true
-
-			if [ -n "$PREVIOUS_TARGET" ]; then
-				if sudo -n ln -sfn "$PREVIOUS_TARGET" "$CURRENT_LINK"; then
-					rm -rf "$NEW_VERSION_DIR"
-					update_status "failed" "failed" "Database migration failed — reverted to previous version"
-				else
-					# Symlink revert failed — keep new version so system stays bootable
-					update_status "failed" "failed" "Database migration failed — symlink revert also failed"
-				fi
-			else
-				# No previous version — keep current in place so system stays bootable
-				update_status "failed" "failed" "Database migration failed — no previous version to revert to"
-			fi
+	if [ ! -f "${NEW_VERSION_DIR}/dist/dataSource.js" ]; then
+		hold_attempt "migrating" "Target data source is missing; update is held for recovery"
+		update_status "failed" "failed" "Target data source is missing; update is held for recovery"
+		FINALIZED="true"
+		exit 1
+	fi
 
-			sudo -n systemctl start smart-panel 2>/dev/null || true
-			exit 1
+	MIGRATION_ENTERED="true"
+	if ! write_attempt "active" "migrating" "" "false" "true"; then
+		hold_attempt "migrating" "Could not persist migration-entry checkpoint"
+		update_status "failed" "failed" "Could not persist migration-entry checkpoint; update is held for recovery"
+		FINALIZED="true"
+		exit 1
+	fi
+
+	MIGRATION_LOG="${ATTEMPT_DIR}/${ATTEMPT_ID}.migration.log"
+	: > "$MIGRATION_LOG"
+	chmod 600 "$MIGRATION_LOG"
+	write_attempt "active" "migrating" "" "false" "true"
+
+	set +e
+	(
+		set -a
+		# shellcheck source=/dev/null
+		[ -f "$ENV_FILE" ] && . "$ENV_FILE"
+		set +a
+		cd "$NEW_VERSION_DIR"
+		node node_modules/typeorm/cli.js migration:run -d dist/dataSource.js
+	) 2>&1 | awk -v output="$MIGRATION_LOG" '
+		BEGIN { limit = 65536; bytes = 0 }
+		{
+			if (bytes < limit) {
+				line = $0 ORS;
+				remaining = limit - bytes;
+				if (length(line) > remaining) line = substr(line, 1, remaining);
+				printf "%s", line >> output;
+				bytes += length(line);
+			}
 		}
+	'
+	pipeline_status=("${PIPESTATUS[@]}")
+	migration_exit_code="${pipeline_status[0]}"
+	capture_exit_code="${pipeline_status[1]}"
+	set -e
+
+	if [ "$capture_exit_code" -ne 0 ]; then
+		hold_attempt "migration_capture_failed" "Could not retain bounded migration diagnostics"
+		update_status "failed" "failed" "Migration diagnostics could not be retained; update is held for recovery"
+		FINALIZED="true"
+		exit 1
+	fi
+
+	if [ "$migration_exit_code" -eq 0 ]; then
+		write_attempt "active" "migrated" "" "false" "true"
+	else
+		hold_attempt "migration_failed" "Database migration failed with exit code ${migration_exit_code}"
+		update_status "failed" "failed" "Database migration failed; update is held for recovery"
+		FINALIZED="true"
+		exit "$migration_exit_code"
 	fi
 
 	# ── Start service ──
 	update_status "starting" "starting"
-	sudo -n systemctl start smart-panel 2>&1 || {
-		# Try to revert if start fails
-		if [ -n "$PREVIOUS_TARGET" ]; then
-			sudo -n ln -sfn "$PREVIOUS_TARGET" "$CURRENT_LINK"
-			sudo -n systemctl start smart-panel 2>/dev/null || true
-		fi
-		update_status "failed" "failed" "Failed to start service after update"
+	write_attempt "active" "starting" "" "false" "true"
+	if ! sudo -n systemctl start "$SERVICE_NAME" 2>&1 || ! wait_for_started_service; then
+		hold_attempt "start_failed" "Target service did not reach verified health after migration"
+		update_status "failed" "failed" "Target service failed to start or become healthy; update is held for recovery"
+		FINALIZED="true"
 		exit 1
-	}
+	fi
 
 	# ── Cleanup old versions (keep max 2 previous) ──
 	cd "$IMAGE_BASE_DIR"
@@ -239,7 +641,9 @@ if [ "$INSTALL_TYPE" = "image" ]; then
 	fi
 
 	# ── Mark complete ──
+	release_attempt
 	update_status "complete" "complete"
+	FINALIZED="true"
 	trap - EXIT
 	exit 0
 fi

package/dist/modules/system/services/update-executor.service.d.ts

@@ -19,7 +19,16 @@ export declare class UpdateExecutorService implements OnModuleInit {
     private progressHighWaterMark;
     constructor(updateService: UpdateService, privilegedWorker: PrivilegedWorkerService, notifications: NotificationsService);
     onModuleInit(): Promise<void>;
+    private readAttempt;
+    private isAttemptUnresolved;
+    private isAttemptOwnerAlive;
+    private readStatusFile;
+    private waitForAttemptSettlement;
+    private reconcileCompletedAttempt;
+    private reconcileLiveAttempt;
     private checkPendingUpdateStatus;
+    private processPendingStatus;
+    private cleanupPendingStatus;
     startUpdate(targetVersion: string): Promise<void>;
     private watchUpdateJob;
     private writeStatusFile;