Merge branch '20540-crunch-run-retry'

[arvados.git] / lib / crunchrun / crunchrun.go

diff --git a/lib/crunchrun/crunchrun.go b/lib/crunchrun/crunchrun.go
index 4a514f3d8966ecbdbc3e43f1a526f9427b99b7b6..7d69fb0f977d408e890cd9a21d44f91be17dd132 100644 (file)
--- a/lib/crunchrun/crunchrun.go
+++ b/lib/crunchrun/crunchrun.go
@@ -638,7 +638,7 @@ func (runner *ContainerRunner) SetupMounts() (map[string]bindmount, error) {
                        if err != nil {
                                return nil, fmt.Errorf("creating temp dir: %v", err)
                        }
-                       err = gitMount(mnt).extractTree(runner.ContainerArvClient, tmpdir, token)
+                       err = gitMount(mnt).extractTree(runner.containerClient, tmpdir, token)
                        if err != nil {
                                return nil, err
                        }
@@ -1345,7 +1345,6 @@ func (runner *ContainerRunner) CaptureOutput(bindmounts map[string]bindmount) er
 
        txt, err := (&copier{
                client:        runner.containerClient,
-               arvClient:     runner.ContainerArvClient,
                keepClient:    runner.ContainerKeepClient,
                hostOutputDir: runner.HostOutputDir,
                ctrOutputDir:  runner.Container.OutputPath,
@@ -1995,7 +1994,9 @@ func (command) RunCommand(prog string, args []string, stdin io.Reader, stdout, s
                log.Printf("%s: %v", containerUUID, err)
                return 1
        }
-       api.Retries = 8
+       // arvadosclient now interprets Retries=10 to mean
+       // Timeout=10m, retrying with exponential backoff + jitter.
+       api.Retries = 10
 
        kc, err := keepclient.MakeKeepClient(api)
        if err != nil {
@@ -2172,7 +2173,9 @@ func hpcConfData(uuid string, configFile string, stderr io.Writer) ConfigData {
                fmt.Fprintf(stderr, "error setting up arvadosclient: %s\n", err)
                return conf
        }
-       arv.Retries = 8
+       // arvadosclient now interprets Retries=10 to mean
+       // Timeout=10m, retrying with exponential backoff + jitter.
+       arv.Retries = 10
        var ctr arvados.Container
        err = arv.Call("GET", "containers", uuid, "", arvadosclient.Dict{"select": []string{"runtime_constraints"}}, &ctr)
        if err != nil {