X-Git-Url: https://git.arvados.org/arvados.git/blobdiff_plain/14d8cfa18f28586ad296e3e598d0a2a536be0613..6b4d02513c17b3de39578f7b3658ea178206fdd4:/lib/crunchrun/docker.go

diff --git a/lib/crunchrun/docker.go b/lib/crunchrun/docker.go
index eee8f1d76a..8d8cdfc8ba 100644
--- a/lib/crunchrun/docker.go
+++ b/lib/crunchrun/docker.go
@@ -4,6 +4,7 @@
 package crunchrun
 
 import (
+	"context"
 	"fmt"
 	"io"
 	"io/ioutil"
@@ -17,12 +18,24 @@ import (
 	dockertypes "github.com/docker/docker/api/types"
 	dockercontainer "github.com/docker/docker/api/types/container"
 	dockerclient "github.com/docker/docker/client"
-	"golang.org/x/net/context"
 )
 
 // Docker daemon won't let you set a limit less than ~10 MiB
 const minDockerRAM = int64(16 * 1024 * 1024)
 
+// DockerAPIVersion is the API version we use to communicate with the
+// docker service.  The oldest OS we support is Ubuntu 18.04 (bionic)
+// which originally shipped docker 1.17.12 / API 1.35 so there is no
+// reason to use an older API version.  See
+// https://dev.arvados.org/issues/15370#note-38 and
+// https://docs.docker.com/engine/api/.
+const DockerAPIVersion = "1.35"
+
+// Number of consecutive "inspect container" failures before
+// concluding Docker is unresponsive, giving up, and cancelling the
+// container.
+const dockerWatchdogThreshold = 3
+
 type dockerExecutor struct {
 	containerUUID    string
 	logf             func(string, ...interface{})
@@ -37,7 +50,7 @@ type dockerExecutor struct {
 func newDockerExecutor(containerUUID string, logf func(string, ...interface{}), watchdogInterval time.Duration) (*dockerExecutor, error) {
 	// API version 1.21 corresponds to Docker 1.9, which is
 	// currently the minimum version we want to support.
-	client, err := dockerclient.NewClient(dockerclient.DefaultDockerHost, "1.21", nil, nil)
+	client, err := dockerclient.NewClient(dockerclient.DefaultDockerHost, DockerAPIVersion, nil, nil)
 	if watchdogInterval < 1 {
 		watchdogInterval = time.Minute
 	}
@@ -217,17 +230,17 @@ func (e *dockerExecutor) Wait(ctx context.Context) (int, error) {
 				// kill it.
 				return
 			} else if err != nil {
-				e.logf("Error inspecting container: %s", err)
-				watchdogErr <- err
-				return
+				watchdogErr <- fmt.Errorf("error inspecting container: %s", err)
 			} else if ctr.State == nil || !(ctr.State.Running || ctr.State.Status == "created") {
-				watchdogErr <- fmt.Errorf("Container is not running: State=%v", ctr.State)
-				return
+				watchdogErr <- fmt.Errorf("container is not running: State=%v", ctr.State)
+			} else {
+				watchdogErr <- nil
 			}
 		}
 	}()
 
 	waitOk, waitErr := e.dockerclient.ContainerWait(ctx, e.containerID, dockercontainer.WaitConditionNotRunning)
+	errors := 0
 	for {
 		select {
 		case waitBody := <-waitOk:
@@ -242,7 +255,16 @@ func (e *dockerExecutor) Wait(ctx context.Context) (int, error) {
 			return -1, ctx.Err()
 
 		case err := <-watchdogErr:
-			return -1, err
+			if err == nil {
+				errors = 0
+			} else {
+				e.logf("docker watchdog: %s", err)
+				errors++
+				if errors >= dockerWatchdogThreshold {
+					e.logf("docker watchdog: giving up")
+					return -1, err
+				}
+			}
 		}
 	}
 }