manager: Only invoke a single sigchld per unit within a cleanup cycle

By default, each iteration of manager_dispatch_sigchld() results in a unit level
sigchld event being invoked. For scope units, this results in a scope_sigchld_event()
which can seemingly stall for workloads that have a large number of PIDs within the
scope. The stall exhibits itself as a SIG_0 being initiated for each u->pids entry
as a result of pid_is_unwaited().

v2:
This patch resolves this condition by only paying to cost of a sigchld in the underlying
scope unit once per sigchld iteration. A new "sigchldgen" member resides within the
Unit struct. The Manager is incremented via the sd event loop, accessed via
sd_event_get_iteration, and the Unit member is set to the same value as the manager each
time that a sigchld event is invoked. If the Manager iteration value and Unit member
match, the sigchld event is not invoked for that iteration.

3 files changed, 15 insertions, 2 deletions

diff --git a/src/core/manager.c b/src/core/manager.c
index 012aa6cd53..1323df7d88 100644
--- a/src/core/manager.c
+++ b/src/core/manager.c
@@ -1716,16 +1716,25 @@ static int manager_dispatch_notify_fd(sd_event_source *source, int fd, uint32_t
 }
 
 static void invoke_sigchld_event(Manager *m, Unit *u, const siginfo_t *si) {
+        uint64_t iteration;
+
         assert(m);
         assert(u);
         assert(si);
 
+        sd_event_get_iteration(m->event, &iteration);
+
         log_unit_debug(u, "Child "PID_FMT" belongs to %s", si->si_pid, u->id);
 
         unit_unwatch_pid(u, si->si_pid);
 
-        if (UNIT_VTABLE(u)->sigchld_event)
-                UNIT_VTABLE(u)->sigchld_event(u, si->si_pid, si->si_code, si->si_status);
+        if (UNIT_VTABLE(u)->sigchld_event) {
+                if (set_size(u->pids) <= 1 || iteration != u->sigchldgen) {
+                        UNIT_VTABLE(u)->sigchld_event(u, si->si_pid, si->si_code, si->si_status);
+                        u->sigchldgen = iteration;
+                } else
+                        log_debug("%s already issued a sigchld this iteration %llu, skipping. Pids still being watched %d", u->id, iteration, set_size(u->pids));
+         }
 }
 
 static int manager_dispatch_sigchld(Manager *m) {
diff --git a/src/core/unit.c b/src/core/unit.c
index 0a1a5321df..8e5395361d 100644
--- a/src/core/unit.c
+++ b/src/core/unit.c
@@ -100,6 +100,7 @@ Unit *unit_new(Manager *m, size_t size) {
         u->on_failure_job_mode = JOB_REPLACE;
         u->cgroup_inotify_wd = -1;
         u->job_timeout = USEC_INFINITY;
+        u->sigchldgen = 0;
 
         RATELIMIT_INIT(u->start_limit, m->default_start_limit_interval, m->default_start_limit_burst);
         RATELIMIT_INIT(u->auto_stop_ratelimit, 10 * USEC_PER_SEC, 16);
diff --git a/src/core/unit.h b/src/core/unit.h
index 08a927962d..c41011ed9d 100644
--- a/src/core/unit.h
+++ b/src/core/unit.h
@@ -162,6 +162,9 @@ struct Unit {
          * process SIGCHLD for */
         Set *pids;
 
+        /* Used in sigchld event invocation to avoid repeat events being invoked */
+        uint64_t sigchldgen;
+
         /* Used during GC sweeps */
         unsigned gc_marker;