From: Tom Clegg <tom@tomclegg.ca>
Date: Tue, 1 Sep 2020 20:22:13 +0000 (-0400)
Subject: 16795: Fix false-positive bug detection.
X-Git-Tag: 2.1.0~89^2~4
X-Git-Url: https://git.arvados.org/arvados.git/commitdiff_plain/76182f26191190c405077106becdde149c0ad7c5

16795: Fix false-positive bug detection.

The dispatcher simulation test occasionally fails with this error when
a crunch-run stub is killed, and the same container is rescheduled on
the same node before the first attempt's startup-phase sleep expires:

    bugf("[test] StubDriver bug or caller bug: pid %d exiting, running[%s]==%d", pid, uuid, svm.running[uuid])
... Error: [test] StubDriver bug or caller bug: pid 9 exiting, running[zzzzz-dz642-000000000000184]==0

Arvados-DCO-1.1-Signed-off-by: Tom Clegg <tom@tomclegg.ca>
---

diff --git a/lib/dispatchcloud/test/stub_driver.go b/lib/dispatchcloud/test/stub_driver.go
index 41eb20763c..2dcd6c1283 100644
--- a/lib/dispatchcloud/test/stub_driver.go
+++ b/lib/dispatchcloud/test/stub_driver.go
@@ -268,13 +268,13 @@ func (svm *StubVM) Exec(env map[string]string, command string, stdin io.Reader,
 		logger.Printf("[test] starting crunch-run stub")
 		go func() {
 			var ctr arvados.Container
-			var started, completed bool
+			var started, completed, killed bool
 			defer func() {
 				logger.Print("[test] exiting crunch-run stub")
 				svm.Lock()
 				defer svm.Unlock()
 				if svm.running[uuid] != pid {
-					if !completed {
+					if !completed && !killed {
 						bugf := svm.sis.driver.Bugf
 						if bugf == nil {
 							bugf = logger.Warnf
@@ -305,7 +305,7 @@ func (svm *StubVM) Exec(env map[string]string, command string, stdin io.Reader,
 			time.Sleep(time.Duration(math_rand.Float64()*20) * time.Millisecond)
 
 			svm.Lock()
-			killed := svm.running[uuid] != pid
+			killed = svm.running[uuid] != pid
 			svm.Unlock()
 			if killed || wantCrashEarly {
 				return