[systemd-devel] [PATCH v3 1/4] service: add watchdog timestamp
Michael Olbrich
m.olbrich at pengutronix.de
Wed Feb 1 08:17:12 PST 2012
This patch adds WatchdogTimestamp[Monotonic] to the systemd service
D-Bus API. The timestamp is updated to the current time when the
service calls 'sd_nofity("WATCHDOG=1\n")'.
Using a timestamp instead of an 'alive' flag has two advantages:
1. No timeout is needed to define when a service is no longer alive.
This simplifies both configuration (no timeout value) and
implementation (no timeout event).
2. It is more robust. A 'dead' service might not be detected should
systemd 'forget' to reset an 'alive' flag. It is much less likely
to get a valid new timestamp if a service died.
---
changes in v2:
- adapt to changes from "d200735 dbus: more efficient implementation of properties"
- serialize timestamp (needed for correct daemon-reload/daemon-reload handling)
man/sd_notify.xml | 12 ++++++++++++
src/dbus-service.c | 5 +++++
src/service.c | 25 +++++++++++++++++++++++++
src/service.h | 2 ++
src/systemd/sd-daemon.h | 5 +++++
5 files changed, 49 insertions(+), 0 deletions(-)
diff --git a/man/sd_notify.xml b/man/sd_notify.xml
index 0209146..9797a5f 100644
--- a/man/sd_notify.xml
+++ b/man/sd_notify.xml
@@ -151,6 +151,18 @@
itself. Example:
"MAINPID=4711"</para></listitem>
</varlistentry>
+
+ <varlistentry>
+ <term>WATCHDOG=1</term>
+
+ <listitem><para>Tells systemd to
+ update the watchdog timestamp.
+ Services using this feature should do
+ this in regular intervals. A watchdog
+ framework can use the timestamps to
+ detect failed
+ services.</para></listitem>
+ </varlistentry>
</variablelist>
<para>It is recommended to prefix variable names that
diff --git a/src/dbus-service.c b/src/dbus-service.c
index e1f6370..d7529ec 100644
--- a/src/dbus-service.c
+++ b/src/dbus-service.c
@@ -43,6 +43,8 @@
" <property name=\"NotifyAccess\" type=\"s\" access=\"read\"/>\n" \
" <property name=\"RestartUSec\" type=\"t\" access=\"read\"/>\n" \
" <property name=\"TimeoutUSec\" type=\"t\" access=\"read\"/>\n" \
+ " <property name=\"WatchdogTimestamp\" type=\"t\" access=\"read\"/>\n" \
+ " <property name=\"WatchdogTimestampMonotonic\" type=\"t\" access=\"read\"/>\n" \
BUS_EXEC_COMMAND_INTERFACE("ExecStartPre") \
BUS_EXEC_COMMAND_INTERFACE("ExecStart") \
BUS_EXEC_COMMAND_INTERFACE("ExecStartPost") \
@@ -86,6 +88,7 @@ const char bus_service_invalidating_properties[] =
"ExecStop\0"
"ExecStopPost\0"
"ExecMain\0"
+ "WatchdogTimestamp\0"
"MainPID\0"
"ControlPID\0"
"StatusText\0";
@@ -112,6 +115,8 @@ static const BusProperty bus_service_properties[] = {
{ "NotifyAccess", bus_service_append_notify_access, "s", offsetof(Service, notify_access) },
{ "RestartUSec", bus_property_append_usec, "t", offsetof(Service, restart_usec) },
{ "TimeoutUSec", bus_property_append_usec, "t", offsetof(Service, timeout_usec) },
+ { "WatchdogTimestamp", bus_property_append_usec, "t", offsetof(Service, watchdog_timestamp.realtime)},
+ { "WatchdogTimestampMonotonic",bus_property_append_usec, "t", offsetof(Service, watchdog_timestamp.monotonic)},
BUS_EXEC_COMMAND_PROPERTY("ExecStartPre", offsetof(Service, exec_command[SERVICE_EXEC_START_PRE]), true ),
BUS_EXEC_COMMAND_PROPERTY("ExecStart", offsetof(Service, exec_command[SERVICE_EXEC_START]), true ),
BUS_EXEC_COMMAND_PROPERTY("ExecStartPost", offsetof(Service, exec_command[SERVICE_EXEC_START_POST]), true ),
diff --git a/src/service.c b/src/service.c
index 4dcd306..e107179 100644
--- a/src/service.c
+++ b/src/service.c
@@ -205,6 +205,19 @@ static void service_connection_unref(Service *s) {
unit_ref_unset(&s->accept_socket);
}
+static void service_stop_watchdog(Service *s) {
+ assert(s);
+
+ s->watchdog_timestamp.realtime = 0;
+ s->watchdog_timestamp.monotonic = 0;
+}
+
+static void service_reset_watchdog(Service *s) {
+ assert(s);
+
+ dual_timestamp_get(&s->watchdog_timestamp);
+}
+
static void service_done(Unit *u) {
Service *s = SERVICE(u);
@@ -1476,6 +1489,9 @@ static void service_set_state(Service *s, ServiceState state) {
service_connection_unref(s);
}
+ if (state == SERVICE_STOP)
+ service_stop_watchdog(s);
+
/* For the inactive states unit_notify() will trim the cgroup,
* but for exit we have to do that ourselves... */
if (state == SERVICE_EXITED && UNIT(s)->manager->n_reloading <= 0)
@@ -2411,6 +2427,9 @@ static int service_serialize(Unit *u, FILE *f, FDSet *fds) {
unit_serialize_item_format(u, f, "main-exec-status-status", "%i", s->main_exec_status.status);
}
}
+ if (dual_timestamp_is_set(&s->watchdog_timestamp)) {
+ dual_timestamp_serialize(f, "watchdog-timestamp", &s->watchdog_timestamp);
+ }
return 0;
}
@@ -2511,6 +2530,8 @@ static int service_deserialize_item(Unit *u, const char *key, const char *value,
dual_timestamp_deserialize(value, &s->main_exec_status.start_timestamp);
else if (streq(key, "main-exec-status-exit"))
dual_timestamp_deserialize(value, &s->main_exec_status.exit_timestamp);
+ else if (streq(key, "watchdog-timestamp"))
+ dual_timestamp_deserialize(value, &s->watchdog_timestamp);
else
log_debug("Unknown serialization key '%s'", key);
@@ -3069,6 +3090,10 @@ static void service_notify_message(Unit *u, pid_t pid, char **tags) {
}
}
+ if (strv_find(tags, "WATCHDOG=1")) {
+ log_debug("%s: got WATCHDOG=1", u->id);
+ service_reset_watchdog(s);
+ }
/* Notify clients about changed status or main pid */
unit_add_to_dbus_queue(u);
diff --git a/src/service.h b/src/service.h
index 0b4f8be..dbae68b 100644
--- a/src/service.h
+++ b/src/service.h
@@ -100,6 +100,8 @@ struct Service {
usec_t restart_usec;
usec_t timeout_usec;
+ dual_timestamp watchdog_timestamp;
+
ExecCommand* exec_command[_SERVICE_EXEC_COMMAND_MAX];
ExecContext exec_context;
diff --git a/src/systemd/sd-daemon.h b/src/systemd/sd-daemon.h
index eb2a606..7b664bf 100644
--- a/src/systemd/sd-daemon.h
+++ b/src/systemd/sd-daemon.h
@@ -217,6 +217,11 @@ int sd_is_mq(int fd, const char *path);
MAINPID=... The main pid of a daemon, in case systemd did not
fork off the process itself. Example: "MAINPID=4711"
+ WATCHDOG=1 Tells systemd to update the watchdog timestamp.
+ Services using this feature should do this in
+ regular intervals. A watchdog framework can use the
+ timestamps to detect failed services.
+
Daemons can choose to send additional variables. However, it is
recommended to prefix variable names not listed above with X_.
--
1.7.7.3
More information about the systemd-devel
mailing list