feat: Ballooning monitors QEMU process (#1097) (68ff6cdd) · Commits · git-mirror / Qemu

src/balloon.py

+34 −4

Original line number	Diff line number	Diff line
		@@ -213,9 +213,24 @@ async def _get_host_qemu_guest_mem_rss(qmp: QMPClient, qemu_pid: int) -> Optiona
		# QMP helpers using qemu.qmp
		# ==========================================================

		async def qmp_wait_connected(sock_path: str, interval: int = 5) -> QMPClient:
		"""Create and connect a QMPClient, retrying until successful."""
		def _is_process_alive(pid: int) -> bool:
		"""Check if a process with the given PID is still running."""
		try:
		os.kill(pid, 0)
		return True
		except ProcessLookupError:
		return False
		except PermissionError:
		return True
		except OSError:
		return False


		async def qmp_wait_connected(sock_path: str, interval: int = 5, qemu_pid: int = -1) -> QMPClient:
		"""Create and connect a QMPClient, retrying until successful or QEMU exits."""
		while True:
		if qemu_pid > 0 and not _is_process_alive(qemu_pid):
		raise ConnectionError("QEMU process (pid %d) is no longer running" % qemu_pid)
		qmp = QMPClient("balloon-monitor")
		try:
		await qmp.connect(sock_path)
		@@ -367,7 +382,7 @@ class BalloonMonitor:
		)

		async def _qmp_connect(self) -> None:
		self.qmp = await qmp_wait_connected(self.args.qmp_sock, self.args.interval)
		self.qmp = await qmp_wait_connected(self.args.qmp_sock, self.args.interval, self.qemu_pid)
		self.event_task = asyncio.create_task(self._qmp_event_listener())
		self._record_balloon(await qmp_get_actual_balloon(self.qmp))

		@@ -389,6 +404,10 @@ class BalloonMonitor:
		log.debug("QMP event listener stopped: %s", e)

		async def _qmp_reconnect(self) -> None:
		if self.qemu_pid > 0 and not _is_process_alive(self.qemu_pid):
		log.info("QEMU process (pid %d) has exited, shutting down.", self.qemu_pid)
		self._stop.set()
		return
		if self.qmp:
		try:
		await self.qmp.disconnect()
		@@ -759,15 +778,20 @@ def main() -> None:
		datefmt="%Y-%m-%d %H:%M:%S",
		)

		qmp_logger = logging.getLogger("qemu.qmp")
		# Suppress noisy qemu.qmp internal messages (e.g. "end-of-file") unless debug is requested
		qmp_logger.setLevel(logging.WARNING)

		if args.debug is not None:
		targets = {t.strip() for t in args.debug.split(",")}
		if "all" in targets:
		logging.getLogger().setLevel(logging.DEBUG)
		qmp_logger.setLevel(logging.DEBUG)
		else:
		if "controller" in targets:
		logging.getLogger(__name__).setLevel(logging.DEBUG)
		if "qmp" in targets:
		logging.getLogger("qemu.qmp").setLevel(logging.DEBUG)
		qmp_logger.setLevel(logging.DEBUG)

		monitor = BalloonMonitor(args)
		asyncio.run(monitor.start())
		@@ -778,3 +802,9 @@ if __name__ == "__main__":
		except KeyboardInterrupt:
		log.debug("Monitor stopped by user.")
		sys.exit(0)
		except ConnectionError as e:
		log.info("Monitor exiting: %s", e)
		sys.exit(0)
		except Exception as e:
		log.error("Monitor terminated unexpectedly: %s", e, exc_info=True)
		sys.exit(1)

src/power.sh

+2 −2

Original line number	Diff line number	Diff line
		@@ -46,7 +46,7 @@ finish() {
		143 ) display="SIGTERM" ;;
		esac
		echo && error "Forcefully terminating $(app), reason: $display..."
		{ kill -9 -- "$pid" \|\| :; } 2>/dev/null
		{ disown "$pid" \|\| :; kill -9 -- "$pid" \|\| :; } 2>/dev/null
		fi
		fi
		fi
		@@ -54,7 +54,7 @@ finish() {
		mKill "${pids[@]}"
		closeNetwork

		if [ -n "$pid" ] && ! waitPid "$pid" 100; then
		if ! waitPidFile "$QEMU_PID" 10; then
		warn "Timed out while waiting for $(app) to exit!"
		fi

src/utils.sh

+39 −18

Original line number	Diff line number	Diff line
		@@ -40,23 +40,44 @@ isAlive() {
		waitPid() {
		local i=0
		local pid="$1"
		local timeout="$2"
		local timeout="${2:-10}"

		while [ -n "$pid" ] && isAlive "$pid"; do
		sleep 0.2
		i=$((i + 1))
		[ "$i" -ge "$timeout" ] && return 1
		(( i >= timeout * 5 )) && return 1
		done

		return 0
		}

		waitPidFile() {
		local i=0
		local pid=""
		local file="$1"
		local timeout="${2:-10}"

		[ ! -s "$file" ] && return 0
		! read -r pid <"$file" && return 0
		[ -z "$pid" ] && return 0

		while [ -s "$file" ] && isAlive "$pid"; do
		sleep 0.2
		i=$((i + 1))
		(( i >= timeout * 5 )) && return 1
		done

		rm -f -- "$file"
		return 0
		}

		pKill() {
		local pid="$1"
		local timeout="${2:-10}"

		{ kill -15 -- "$pid" \|\| :; } 2>/dev/null

		if ! waitPid "$pid" 50; then
		if ! waitPid "$pid" "$timeout"; then
		warn "Timed out while waiting for PID $pid"
		fi

		@@ -64,13 +85,16 @@ pKill() {
		}

		fWait() {
		local name="$1" i=0
		local i=0
		local name="$1"
		local timeout="${2:-10}"

		[ -z "$name" ] && return 0

		while pgrep -f -l "$name" >/dev/null; do
		sleep 0.2
		i=$((i + 1))
		if [ "$i" -ge 50 ]; then
		if (( i >= timeout * 5 )); then
		warn "Timed out while waiting for process: $name"
		break
		fi
		@@ -81,21 +105,25 @@ fWait() {

		fKill() {
		local name="$1"
		local timeout="${2:-10}"

		[ -z "$name" ] && return 0

		{ pkill -f "$name" \|\| :; } 2>/dev/null
		fWait "$name"
		fWait "$name" "$timeout"

		return 0
		}

		sKill() {
		local file="$1" pid=""
		local pid=""
		local file="$1"

		[ ! -s "$file" ] && return 0
		! read -r pid <"$file" && return 0
		[ -z "$pid" ] && return 0

		if [ -n "$pid" ] && isAlive "$pid"; then
		if isAlive "$pid"; then
		{ kill -15 -- "$pid" \|\| :; } 2>/dev/null
		fi

		@@ -103,24 +131,17 @@ sKill() {
		}

		mKill() {
		local pid="" files=("$@")
		local timeout=10
		local files=("$@")

		for file in "${files[@]}"; do
		sKill "$file"
		done

		for file in "${files[@]}"; do

		[ ! -s "$file" ] && continue
		! read -r pid <"$file" && continue
		[ -z "$pid" ] && continue

		if waitPid "$pid" 50; then
		rm -f -- "$file"
		else
		if ! waitPidFile "$file" "$timeout"; then
		warn "Timed out while waiting for PID file: $file"
		fi

		done

		return 0