Increase WatchdogInterval and WatchdogThreshold
authorPeter Amstutz <peter.amstutz@curii.com>
Fri, 26 May 2023 19:26:12 +0000 (15:26 -0400)
committerPeter Amstutz <peter.amstutz@curii.com>
Thu, 1 Jun 2023 19:49:31 +0000 (15:49 -0400)
Quick workaround for customer, we should add these to the config file.

refs #20595

Arvados-DCO-1.1-Signed-off-by: Peter Amstutz <peter.amstutz@curii.com>

lib/crunchrun/docker.go

index 8d8cdfc8ba620a4317d4f48bf0f654c04cf58bef..b402637e91cfe6f4726ad0a8dd4e52e604db6ae7 100644 (file)
@@ -34,7 +34,7 @@ const DockerAPIVersion = "1.35"
 // Number of consecutive "inspect container" failures before
 // concluding Docker is unresponsive, giving up, and cancelling the
 // container.
-const dockerWatchdogThreshold = 3
+const dockerWatchdogThreshold = 5
 
 type dockerExecutor struct {
        containerUUID    string
@@ -52,7 +52,7 @@ func newDockerExecutor(containerUUID string, logf func(string, ...interface{}),
        // currently the minimum version we want to support.
        client, err := dockerclient.NewClient(dockerclient.DefaultDockerHost, DockerAPIVersion, nil, nil)
        if watchdogInterval < 1 {
-               watchdogInterval = time.Minute
+               watchdogInterval = time.Minute * 2
        }
        return &dockerExecutor{
                containerUUID:    containerUUID,