Merge cloudbots codes from TOT to all release branches
BUG=b:336556314
[cloudbots][cft] Add IsAlive check when reboot through cloudbots
BUG=b:325553505
TEST=cros_sdk cros-workon --host start cros-dut, cros_sdk sudo emerge cros-dut, bniche-test cros-tool-runner on cloudbot VM
Change-Id: Id4b738e61d104295c7bd162c28f066caa1ce557e
Reviewed-on: https://chromium-review.googlesource.com/c/chromiumos/platform/dev-util/+/5388456
Tested-by: Bryan Ni <bniche@google.com>
Reviewed-by: Azizur Rahman <azrahman@google.com>
Reviewed-by: Sergey Fetisov <sfetisov@google.com>
Reviewed-by: Derek Beckett <dbeckett@chromium.org>
Commit-Queue: Bryan Ni <bniche@google.com>
(cherry picked from commit a7464ae6717a6d4a9ad7161c5e981494656842b9)
Reviewed-on: https://chromium-review.googlesource.com/c/chromiumos/platform/dev-util/+/5513107
diff --git a/src/chromiumos/test/dut/cmd/cros-dut/dutserver.go b/src/chromiumos/test/dut/cmd/cros-dut/dutserver.go
index 1fe01af..c26f6d7 100644
--- a/src/chromiumos/test/dut/cmd/cros-dut/dutserver.go
+++ b/src/chromiumos/test/dut/cmd/cros-dut/dutserver.go
@@ -279,6 +279,33 @@
// by waiting for the client connection to shutdown or a timeout.
s.logger.Printf("Waiting for reboot to complete.")
+ // On cloudbots, the Wait() never returned even after the dut has
+ // already been rebooted. This caused timeout and test failures.
+ // By checking connection IsAlive() periodically, the Wait() does return and
+ // test can continue as expected.
+ isCloudbot := false
+ if id, found := os.LookupEnv("SWARMING_BOT_ID"); found && strings.HasPrefix(id, "cloudbots-") {
+ isCloudbot = true
+ }
+ if isCloudbot {
+ go func() {
+ t := time.NewTimer(15 * time.Second)
+ for {
+ select {
+ case <-t.C:
+ s.logger.Printf("IsConnectionAlive=%v", s.connection.IsAlive())
+ t.Reset(10 * time.Second)
+ return
+ case <-ctx.Done():
+ if !t.Stop() {
+ <-t.C
+ }
+ return
+ }
+ }
+ }()
+ }
+
wait := make(chan interface{})
go func() {
s.logger.Printf("Waiting for reboot: Connection wait.")
