X-Git-Url: https://git.arvados.org/arvados.git/blobdiff_plain/dfb9598282b677ead60f91c14f5e96405735d42f..ea3f1b8246c27a6a44edfc561f13935ef377c1cb:/lib/crunchrun/crunchrun.go

diff --git a/lib/crunchrun/crunchrun.go b/lib/crunchrun/crunchrun.go
index b0a4007f74..e05b352fae 100644
--- a/lib/crunchrun/crunchrun.go
+++ b/lib/crunchrun/crunchrun.go
@@ -870,25 +870,24 @@ func (runner *ContainerRunner) LogNodeRecord() error {
 			return err
 		}
 		return w.Close()
-	} else {
-		// Dispatched via crunch-dispatch-slurm. Look up
-		// apiserver's node record corresponding to
-		// $SLURMD_NODENAME.
-		hostname := os.Getenv("SLURMD_NODENAME")
-		if hostname == "" {
-			hostname, _ = os.Hostname()
-		}
-		_, err := runner.logAPIResponse("node", "nodes", map[string]interface{}{"filters": [][]string{{"hostname", "=", hostname}}}, func(resp interface{}) {
-			// The "info" field has admin-only info when
-			// obtained with a privileged token, and
-			// should not be logged.
-			node, ok := resp.(map[string]interface{})
-			if ok {
-				delete(node, "info")
-			}
-		})
-		return err
 	}
+	// Dispatched via crunch-dispatch-slurm. Look up
+	// apiserver's node record corresponding to
+	// $SLURMD_NODENAME.
+	hostname := os.Getenv("SLURMD_NODENAME")
+	if hostname == "" {
+		hostname, _ = os.Hostname()
+	}
+	_, err := runner.logAPIResponse("node", "nodes", map[string]interface{}{"filters": [][]string{{"hostname", "=", hostname}}}, func(resp interface{}) {
+		// The "info" field has admin-only info when
+		// obtained with a privileged token, and
+		// should not be logged.
+		node, ok := resp.(map[string]interface{})
+		if ok {
+			delete(node, "info")
+		}
+	})
+	return err
 }
 
 func (runner *ContainerRunner) logAPIResponse(label, path string, params map[string]interface{}, munge func(interface{})) (logged bool, err error) {
@@ -1074,9 +1073,10 @@ func (runner *ContainerRunner) CreateContainer() error {
 	runner.ContainerConfig.Volumes = runner.Volumes
 
 	maxRAM := int64(runner.Container.RuntimeConstraints.RAM)
-	if maxRAM < 4*1024*1024 {
-		// Docker daemon won't let you set a limit less than 4 MiB
-		maxRAM = 4 * 1024 * 1024
+	minDockerRAM := int64(16)
+	if maxRAM < minDockerRAM*1024*1024 {
+		// Docker daemon won't let you set a limit less than ~10 MiB
+		maxRAM = minDockerRAM * 1024 * 1024
 	}
 	runner.HostConfig = dockercontainer.HostConfig{
 		Binds: runner.Binds,
@@ -1431,15 +1431,20 @@ func (runner *ContainerRunner) saveLogCollection(final bool) (response arvados.C
 		// Already finalized.
 		return
 	}
-	mt, err := runner.LogCollection.MarshalManifest(".")
-	if err != nil {
-		err = fmt.Errorf("error creating log manifest: %v", err)
-		return
-	}
 	updates := arvadosclient.Dict{
-		"name":          "logs for " + runner.Container.UUID,
-		"manifest_text": mt,
+		"name": "logs for " + runner.Container.UUID,
+	}
+	mt, err1 := runner.LogCollection.MarshalManifest(".")
+	if err1 == nil {
+		// Only send updated manifest text if there was no
+		// error.
+		updates["manifest_text"] = mt
 	}
+
+	// Even if flushing the manifest had an error, we still want
+	// to update the log record, if possible, to push the trash_at
+	// and delete_at times into the future.  Details on bug
+	// #17293.
 	if final {
 		updates["is_trashed"] = true
 	} else {
@@ -1448,16 +1453,20 @@ func (runner *ContainerRunner) saveLogCollection(final bool) (response arvados.C
 		updates["delete_at"] = exp
 	}
 	reqBody := arvadosclient.Dict{"collection": updates}
+	var err2 error
 	if runner.logUUID == "" {
 		reqBody["ensure_unique_name"] = true
-		err = runner.DispatcherArvClient.Create("collections", reqBody, &response)
+		err2 = runner.DispatcherArvClient.Create("collections", reqBody, &response)
 	} else {
-		err = runner.DispatcherArvClient.Update("collections", runner.logUUID, reqBody, &response)
+		err2 = runner.DispatcherArvClient.Update("collections", runner.logUUID, reqBody, &response)
 	}
-	if err != nil {
-		return
+	if err2 == nil {
+		runner.logUUID = response.UUID
+	}
+
+	if err1 != nil || err2 != nil {
+		err = fmt.Errorf("error recording logs: %q, %q", err1, err2)
 	}
-	runner.logUUID = response.UUID
 	return
 }
 
@@ -1488,7 +1497,7 @@ func (runner *ContainerRunner) ContainerToken() (string, error) {
 	return runner.token, nil
 }
 
-// UpdateContainerComplete updates the container record state on API
+// UpdateContainerFinal updates the container record state on API
 // server to "Complete" or "Cancelled"
 func (runner *ContainerRunner) UpdateContainerFinal() error {
 	update := arvadosclient.Dict{}