You can not select more than 25 topics
Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.
340 lines
13 KiB
340 lines
13 KiB
8 months ago
|
From 5458256264c97eee521caf07c705f549a0f0bd55 Mon Sep 17 00:00:00 2001
|
||
|
From: Lennart Poettering <lennart@poettering.net>
|
||
|
Date: Thu, 9 Aug 2018 16:26:27 +0200
|
||
|
Subject: [PATCH] core: rework StopWhenUnneeded= logic
|
||
|
|
||
|
Previously, we'd act immediately on StopWhenUnneeded= when a unit state
|
||
|
changes. With this rework we'll maintain a queue instead: whenever
|
||
|
there's the chance that StopWhenUneeded= might have an effect we enqueue
|
||
|
the unit, and process it later when we have nothing better to do.
|
||
|
|
||
|
This should make the implementation a bit more reliable, as the unit notify event
|
||
|
cannot immediately enqueue tons of side-effect jobs that might
|
||
|
contradict each other, but we do so only in a strictly ordered fashion,
|
||
|
from the main event loop.
|
||
|
|
||
|
This slightly changes the check when to consider a unit "unneeded".
|
||
|
Previously, we'd assume that a unit in "deactivating" state could also
|
||
|
be cleaned up. With this new logic we'll only consider units unneeded
|
||
|
that are fully up and have no job queued. This means that whenever
|
||
|
there's something pending for a unit we won't clean it up.
|
||
|
|
||
|
(cherry picked from commit a3c1168ac293f16d9343d248795bb4c246aaff4a)
|
||
|
|
||
|
Resolves: #1798046
|
||
|
---
|
||
|
src/core/manager.c | 43 ++++++++++++++++
|
||
|
src/core/manager.h | 3 ++
|
||
|
src/core/unit.c | 122 +++++++++++++++++++++++++--------------------
|
||
|
src/core/unit.h | 7 +++
|
||
|
4 files changed, 120 insertions(+), 55 deletions(-)
|
||
|
|
||
|
diff --git a/src/core/manager.c b/src/core/manager.c
|
||
|
index 0eae7d46fb..4c04896aaa 100644
|
||
|
--- a/src/core/manager.c
|
||
|
+++ b/src/core/manager.c
|
||
|
@@ -1211,6 +1211,45 @@ static unsigned manager_dispatch_gc_job_queue(Manager *m) {
|
||
|
return n;
|
||
|
}
|
||
|
|
||
|
+static unsigned manager_dispatch_stop_when_unneeded_queue(Manager *m) {
|
||
|
+ unsigned n = 0;
|
||
|
+ Unit *u;
|
||
|
+ int r;
|
||
|
+
|
||
|
+ assert(m);
|
||
|
+
|
||
|
+ while ((u = m->stop_when_unneeded_queue)) {
|
||
|
+ _cleanup_(sd_bus_error_free) sd_bus_error error = SD_BUS_ERROR_NULL;
|
||
|
+ assert(m->stop_when_unneeded_queue);
|
||
|
+
|
||
|
+ assert(u->in_stop_when_unneeded_queue);
|
||
|
+ LIST_REMOVE(stop_when_unneeded_queue, m->stop_when_unneeded_queue, u);
|
||
|
+ u->in_stop_when_unneeded_queue = false;
|
||
|
+
|
||
|
+ n++;
|
||
|
+
|
||
|
+ if (!unit_is_unneeded(u))
|
||
|
+ continue;
|
||
|
+
|
||
|
+ log_unit_debug(u, "Unit is not needed anymore.");
|
||
|
+
|
||
|
+ /* If stopping a unit fails continuously we might enter a stop loop here, hence stop acting on the
|
||
|
+ * service being unnecessary after a while. */
|
||
|
+
|
||
|
+ if (!ratelimit_below(&u->auto_stop_ratelimit)) {
|
||
|
+ log_unit_warning(u, "Unit not needed anymore, but not stopping since we tried this too often recently.");
|
||
|
+ continue;
|
||
|
+ }
|
||
|
+
|
||
|
+ /* Ok, nobody needs us anymore. Sniff. Then let's commit suicide */
|
||
|
+ r = manager_add_job(u->manager, JOB_STOP, u, JOB_FAIL, &error, NULL);
|
||
|
+ if (r < 0)
|
||
|
+ log_unit_warning_errno(u, r, "Failed to enqueue stop job, ignoring: %s", bus_error_message(&error, r));
|
||
|
+ }
|
||
|
+
|
||
|
+ return n;
|
||
|
+}
|
||
|
+
|
||
|
static void manager_clear_jobs_and_units(Manager *m) {
|
||
|
Unit *u;
|
||
|
|
||
|
@@ -1228,6 +1267,7 @@ static void manager_clear_jobs_and_units(Manager *m) {
|
||
|
assert(!m->cleanup_queue);
|
||
|
assert(!m->gc_unit_queue);
|
||
|
assert(!m->gc_job_queue);
|
||
|
+ assert(!m->stop_when_unneeded_queue);
|
||
|
|
||
|
assert(hashmap_isempty(m->jobs));
|
||
|
assert(hashmap_isempty(m->units));
|
||
|
@@ -2824,6 +2864,9 @@ int manager_loop(Manager *m) {
|
||
|
if (manager_dispatch_cgroup_realize_queue(m) > 0)
|
||
|
continue;
|
||
|
|
||
|
+ if (manager_dispatch_stop_when_unneeded_queue(m) > 0)
|
||
|
+ continue;
|
||
|
+
|
||
|
if (manager_dispatch_dbus_queue(m) > 0)
|
||
|
continue;
|
||
|
|
||
|
diff --git a/src/core/manager.h b/src/core/manager.h
|
||
|
index fa47952d24..40568d3c8b 100644
|
||
|
--- a/src/core/manager.h
|
||
|
+++ b/src/core/manager.h
|
||
|
@@ -130,6 +130,9 @@ struct Manager {
|
||
|
/* Target units whose default target dependencies haven't been set yet */
|
||
|
LIST_HEAD(Unit, target_deps_queue);
|
||
|
|
||
|
+ /* Units that might be subject to StopWhenUnneeded= clean-up */
|
||
|
+ LIST_HEAD(Unit, stop_when_unneeded_queue);
|
||
|
+
|
||
|
sd_event *event;
|
||
|
|
||
|
/* This maps PIDs we care about to units that are interested in. We allow multiple units to he interested in
|
||
|
diff --git a/src/core/unit.c b/src/core/unit.c
|
||
|
index e1f5e6f7bd..40f138d25c 100644
|
||
|
--- a/src/core/unit.c
|
||
|
+++ b/src/core/unit.c
|
||
|
@@ -438,6 +438,22 @@ void unit_add_to_dbus_queue(Unit *u) {
|
||
|
u->in_dbus_queue = true;
|
||
|
}
|
||
|
|
||
|
+void unit_add_to_stop_when_unneeded_queue(Unit *u) {
|
||
|
+ assert(u);
|
||
|
+
|
||
|
+ if (u->in_stop_when_unneeded_queue)
|
||
|
+ return;
|
||
|
+
|
||
|
+ if (!u->stop_when_unneeded)
|
||
|
+ return;
|
||
|
+
|
||
|
+ if (!UNIT_IS_ACTIVE_OR_RELOADING(unit_active_state(u)))
|
||
|
+ return;
|
||
|
+
|
||
|
+ LIST_PREPEND(stop_when_unneeded_queue, u->manager->stop_when_unneeded_queue, u);
|
||
|
+ u->in_stop_when_unneeded_queue = true;
|
||
|
+}
|
||
|
+
|
||
|
static void bidi_set_free(Unit *u, Hashmap *h) {
|
||
|
Unit *other;
|
||
|
Iterator i;
|
||
|
@@ -634,6 +650,9 @@ void unit_free(Unit *u) {
|
||
|
if (u->in_target_deps_queue)
|
||
|
LIST_REMOVE(target_deps_queue, u->manager->target_deps_queue, u);
|
||
|
|
||
|
+ if (u->in_stop_when_unneeded_queue)
|
||
|
+ LIST_REMOVE(stop_when_unneeded_queue, u->manager->stop_when_unneeded_queue, u);
|
||
|
+
|
||
|
safe_close(u->ip_accounting_ingress_map_fd);
|
||
|
safe_close(u->ip_accounting_egress_map_fd);
|
||
|
|
||
|
@@ -1950,55 +1969,71 @@ bool unit_can_reload(Unit *u) {
|
||
|
return UNIT_VTABLE(u)->reload;
|
||
|
}
|
||
|
|
||
|
-static void unit_check_unneeded(Unit *u) {
|
||
|
-
|
||
|
- _cleanup_(sd_bus_error_free) sd_bus_error error = SD_BUS_ERROR_NULL;
|
||
|
-
|
||
|
- static const UnitDependency needed_dependencies[] = {
|
||
|
+bool unit_is_unneeded(Unit *u) {
|
||
|
+ static const UnitDependency deps[] = {
|
||
|
UNIT_REQUIRED_BY,
|
||
|
UNIT_REQUISITE_OF,
|
||
|
UNIT_WANTED_BY,
|
||
|
UNIT_BOUND_BY,
|
||
|
};
|
||
|
-
|
||
|
- unsigned j;
|
||
|
- int r;
|
||
|
+ size_t j;
|
||
|
|
||
|
assert(u);
|
||
|
|
||
|
- /* If this service shall be shut down when unneeded then do
|
||
|
- * so. */
|
||
|
-
|
||
|
if (!u->stop_when_unneeded)
|
||
|
- return;
|
||
|
+ return false;
|
||
|
|
||
|
- if (!UNIT_IS_ACTIVE_OR_ACTIVATING(unit_active_state(u)))
|
||
|
- return;
|
||
|
+ /* Don't clean up while the unit is transitioning or is even inactive. */
|
||
|
+ if (!UNIT_IS_ACTIVE_OR_RELOADING(unit_active_state(u)))
|
||
|
+ return false;
|
||
|
+ if (u->job)
|
||
|
+ return false;
|
||
|
|
||
|
- for (j = 0; j < ELEMENTSOF(needed_dependencies); j++) {
|
||
|
+ for (j = 0; j < ELEMENTSOF(deps); j++) {
|
||
|
Unit *other;
|
||
|
Iterator i;
|
||
|
void *v;
|
||
|
|
||
|
- HASHMAP_FOREACH_KEY(v, other, u->dependencies[needed_dependencies[j]], i)
|
||
|
- if (unit_active_or_pending(other) || unit_will_restart(other))
|
||
|
- return;
|
||
|
- }
|
||
|
+ /* If a dependending unit has a job queued, or is active (or in transitioning), or is marked for
|
||
|
+ * restart, then don't clean this one up. */
|
||
|
|
||
|
- /* If stopping a unit fails continuously we might enter a stop
|
||
|
- * loop here, hence stop acting on the service being
|
||
|
- * unnecessary after a while. */
|
||
|
- if (!ratelimit_below(&u->auto_stop_ratelimit)) {
|
||
|
- log_unit_warning(u, "Unit not needed anymore, but not stopping since we tried this too often recently.");
|
||
|
- return;
|
||
|
+ HASHMAP_FOREACH_KEY(v, other, u->dependencies[deps[j]], i) {
|
||
|
+ if (u->job)
|
||
|
+ return false;
|
||
|
+
|
||
|
+ if (!UNIT_IS_INACTIVE_OR_FAILED(unit_active_state(other)))
|
||
|
+ return false;
|
||
|
+
|
||
|
+ if (unit_will_restart(other))
|
||
|
+ return false;
|
||
|
+ }
|
||
|
}
|
||
|
|
||
|
- log_unit_info(u, "Unit not needed anymore. Stopping.");
|
||
|
+ return true;
|
||
|
+}
|
||
|
|
||
|
- /* Ok, nobody needs us anymore. Sniff. Then let's commit suicide */
|
||
|
- r = manager_add_job(u->manager, JOB_STOP, u, JOB_FAIL, &error, NULL);
|
||
|
- if (r < 0)
|
||
|
- log_unit_warning_errno(u, r, "Failed to enqueue stop job, ignoring: %s", bus_error_message(&error, r));
|
||
|
+static void check_unneeded_dependencies(Unit *u) {
|
||
|
+
|
||
|
+ static const UnitDependency deps[] = {
|
||
|
+ UNIT_REQUIRES,
|
||
|
+ UNIT_REQUISITE,
|
||
|
+ UNIT_WANTS,
|
||
|
+ UNIT_BINDS_TO,
|
||
|
+ };
|
||
|
+ size_t j;
|
||
|
+
|
||
|
+ assert(u);
|
||
|
+
|
||
|
+ /* Add all units this unit depends on to the queue that processes StopWhenUnneeded= behaviour. */
|
||
|
+
|
||
|
+ for (j = 0; j < ELEMENTSOF(deps); j++) {
|
||
|
+ Unit *other;
|
||
|
+ Iterator i;
|
||
|
+ void *v;
|
||
|
+
|
||
|
+ HASHMAP_FOREACH_KEY(v, other, u->dependencies[deps[j]], i)
|
||
|
+ unit_add_to_stop_when_unneeded_queue(other);
|
||
|
+ }
|
||
|
}
|
||
|
|
||
|
static void unit_check_binds_to(Unit *u) {
|
||
|
@@ -2098,29 +2133,6 @@ static void retroactively_stop_dependencies(Unit *u) {
|
||
|
manager_add_job(u->manager, JOB_STOP, other, JOB_REPLACE, NULL, NULL);
|
||
|
}
|
||
|
|
||
|
-static void check_unneeded_dependencies(Unit *u) {
|
||
|
- Unit *other;
|
||
|
- Iterator i;
|
||
|
- void *v;
|
||
|
-
|
||
|
- assert(u);
|
||
|
- assert(UNIT_IS_INACTIVE_OR_DEACTIVATING(unit_active_state(u)));
|
||
|
-
|
||
|
- /* Garbage collect services that might not be needed anymore, if enabled */
|
||
|
- HASHMAP_FOREACH_KEY(v, other, u->dependencies[UNIT_REQUIRES], i)
|
||
|
- if (!UNIT_IS_INACTIVE_OR_DEACTIVATING(unit_active_state(other)))
|
||
|
- unit_check_unneeded(other);
|
||
|
- HASHMAP_FOREACH_KEY(v, other, u->dependencies[UNIT_WANTS], i)
|
||
|
- if (!UNIT_IS_INACTIVE_OR_DEACTIVATING(unit_active_state(other)))
|
||
|
- unit_check_unneeded(other);
|
||
|
- HASHMAP_FOREACH_KEY(v, other, u->dependencies[UNIT_REQUISITE], i)
|
||
|
- if (!UNIT_IS_INACTIVE_OR_DEACTIVATING(unit_active_state(other)))
|
||
|
- unit_check_unneeded(other);
|
||
|
- HASHMAP_FOREACH_KEY(v, other, u->dependencies[UNIT_BINDS_TO], i)
|
||
|
- if (!UNIT_IS_INACTIVE_OR_DEACTIVATING(unit_active_state(other)))
|
||
|
- unit_check_unneeded(other);
|
||
|
-}
|
||
|
-
|
||
|
void unit_start_on_failure(Unit *u) {
|
||
|
Unit *other;
|
||
|
Iterator i;
|
||
|
@@ -2423,7 +2435,7 @@ void unit_notify(Unit *u, UnitActiveState os, UnitActiveState ns, UnitNotifyFlag
|
||
|
}
|
||
|
|
||
|
/* stop unneeded units regardless if going down was expected or not */
|
||
|
- if (UNIT_IS_INACTIVE_OR_DEACTIVATING(ns))
|
||
|
+ if (UNIT_IS_INACTIVE_OR_FAILED(ns))
|
||
|
check_unneeded_dependencies(u);
|
||
|
|
||
|
if (ns != os && ns == UNIT_FAILED) {
|
||
|
@@ -2483,7 +2495,7 @@ void unit_notify(Unit *u, UnitActiveState os, UnitActiveState ns, UnitNotifyFlag
|
||
|
|
||
|
if (!MANAGER_IS_RELOADING(u->manager)) {
|
||
|
/* Maybe we finished startup and are now ready for being stopped because unneeded? */
|
||
|
- unit_check_unneeded(u);
|
||
|
+ unit_add_to_stop_when_unneeded_queue(u);
|
||
|
|
||
|
/* Maybe we finished startup, but something we needed has vanished? Let's die then. (This happens when
|
||
|
* something BindsTo= to a Type=oneshot unit, as these units go directly from starting to inactive,
|
||
|
diff --git a/src/core/unit.h b/src/core/unit.h
|
||
|
index 99755823eb..595ee88d43 100644
|
||
|
--- a/src/core/unit.h
|
||
|
+++ b/src/core/unit.h
|
||
|
@@ -212,6 +212,9 @@ typedef struct Unit {
|
||
|
/* Target dependencies queue */
|
||
|
LIST_FIELDS(Unit, target_deps_queue);
|
||
|
|
||
|
+ /* Queue of units with StopWhenUnneeded set that shell be checked for clean-up. */
|
||
|
+ LIST_FIELDS(Unit, stop_when_unneeded_queue);
|
||
|
+
|
||
|
/* PIDs we keep an eye on. Note that a unit might have many
|
||
|
* more, but these are the ones we care enough about to
|
||
|
* process SIGCHLD for */
|
||
|
@@ -322,6 +325,7 @@ typedef struct Unit {
|
||
|
bool in_cgroup_realize_queue:1;
|
||
|
bool in_cgroup_empty_queue:1;
|
||
|
bool in_target_deps_queue:1;
|
||
|
+ bool in_stop_when_unneeded_queue:1;
|
||
|
|
||
|
bool sent_dbus_new_signal:1;
|
||
|
|
||
|
@@ -615,6 +619,7 @@ void unit_add_to_dbus_queue(Unit *u);
|
||
|
void unit_add_to_cleanup_queue(Unit *u);
|
||
|
void unit_add_to_gc_queue(Unit *u);
|
||
|
void unit_add_to_target_deps_queue(Unit *u);
|
||
|
+void unit_add_to_stop_when_unneeded_queue(Unit *u);
|
||
|
|
||
|
int unit_merge(Unit *u, Unit *other);
|
||
|
int unit_merge_by_name(Unit *u, const char *other);
|
||
|
@@ -751,6 +756,8 @@ bool unit_type_supported(UnitType t);
|
||
|
|
||
|
bool unit_is_pristine(Unit *u);
|
||
|
|
||
|
+bool unit_is_unneeded(Unit *u);
|
||
|
+
|
||
|
pid_t unit_control_pid(Unit *u);
|
||
|
pid_t unit_main_pid(Unit *u);
|
||
|
|