While the originally written timeout for process killing (2 seconds)
was way too short, waiting indefinitely is suboptimal as well. We're
seeing cases where the test is stuck for possibly hours in
uninterruptible sleep (IO). Wait a fairly longer selected time period
of 2 minutes, because even making progress for that long means the
machine is in bad enough state to require a good kicking and booting.
v2:
- Abort quicker if kernel is tainted (Chris)
- Correctly convert process-exists check with kill() to process-does-not-exist
Signed-off-by: Petri Latvala <petri.latvala at intel.com>
Cc: Chris Wilson <chris at chris-wilson.co.uk>
Cc: Arkadiusz Hiler <arkadiusz.hiler at intel.com>
---
runner/executor.c | 29 ++++++++++++++++++++---------
1 file changed, 20 insertions(+), 9 deletions(-)
diff --git a/runner/executor.c b/runner/executor.c
index e6086772..f36bfd3d 100644
--- a/runner/executor.c
+++ b/runner/executor.c
@@ -682,6 +682,9 @@ static int monitor_output(pid_t child,
int timeout_intervals = 1, intervals_left;
int wd_extra = 10;
int killed = 0; /* 0 if not killed, signal number otherwise */
+ int sigkill_timeout = 120;
+ int sigkill_interval = 20;
+ int sigkill_intervals_left = sigkill_timeout / sigkill_interval;
struct timespec time_beg, time_end;
unsigned long taints = 0;
bool aborting = false;
@@ -776,25 +779,33 @@ static int monitor_output(pid_t child,
if (!kill_child(killed, child))
return -1;
- intervals_left = timeout_intervals = 1;
- break;
- case SIGKILL:
/*
- * If the child still exists, and the kernel
- * hasn't oopsed, assume it is still making
- * forward progress towards exiting (i.e. still
- * freeing all of its resources).
+ * Allow the test two minutes to die
+ * on SIGKILL. If it takes more than
+ * that, we're quite likely in a
+ * scenario where we want to reboot
+ * the machine anyway.
*/
- if (kill(child, 0) == 0 && !tainted(&taints)) {
- intervals_left = 1;
+ watchdogs_set_timeout(sigkill_timeout);
+ timeout = sigkill_interval;
+ intervals_left = 1; /* Intervals handled separately for sigkill */
+ break;
+ case SIGKILL:
+ if (!tainted(&taints) && --sigkill_intervals_left) {
+ intervals_left = 1;
break;
}
/* Nothing that can be done, really. Let's tell the caller we want to abort. */
+
if (settings->log_level >= LOG_LEVEL_NORMAL) {
errf("Child refuses to die, tainted %lx. Aborting.\n",
taints);
+ if (kill(child, 0) && errno == ESRCH)
+ errf("The test process no longer exists, "
+ "but we didn't get informed of its demise...\n");
}
+
close_watchdogs(settings);
free(outbuf);
close(outfd);
--
2.19.1