Allow download retrying if the computed and configured checksums differ.

- We have identified that downloading images under poor networking conditions
from datastores might fail silently due to the SDK libraries we use for
for Azure and AWS. This happens when we are installing new applications.
- Another observation we made is that when we manually try re-installing
the applications, they eventually succeed in the installation.
- With this patch, we automate retrying the image download a fixed number
of times (5 by default) if the computed versus the configure checksums differ.

Signed-off-by: Ioannis Sfakianakis <[email protected]>

pkg/pillar/cmd/downloader/downloader.go

@@ -398,9 +398,18 @@ func handleModify(ctx *downloaderContext, key string,
 		status.Expired, status.Name)
 
 	// If RefCount from zero to non-zero and status has error
-	// or status is not downloaded then do install
-	if config.RefCount != 0 && (status.HasError() || status.State != types.DOWNLOADED) {
+	// or status is not downloaded or retry then do install
+	if config.RefCount != 0 && (status.HasError() ||
+		status.State != types.DOWNLOADED || config.DoRetry) {
 		log.Functionf("handleModify installing %s", config.Name)
+		if config.DoRetry {
+			log.Functionf("handleModify retry download %s", config.Name)
+			status.CurrentSize = 0
+			status.Size = 0
+			status.Progress = 0
+			status.State = types.DOWNLOADING
+			publishDownloaderStatus(ctx, status)
+		}
 		handleCreate(ctx, config, status, key, receiveChan)
 	} else if status.RefCount != config.RefCount {
 		log.Functionf("handleModify RefCount change %s from %d to %d",

pkg/pillar/cmd/volumemgr/handledownloader.go

@@ -64,6 +64,7 @@ func AddOrRefcountDownloaderConfig(ctx *volumemgrContext, blob types.BlobStatus)
 		Size:            size,
 		Target:          locFilename,
 		RefCount:        refCount,
+		DoRetry:         false,
 	}
 	log.Functionf("AddOrRefcountDownloaderConfig: DownloaderConfig: %+v", n)
 	publishDownloaderConfig(ctx, &n)
@@ -105,6 +106,24 @@ func MaybeRemoveDownloaderConfig(ctx *volumemgrContext, imageSha string) {
 	log.Functionf("MaybeRemoveDownloaderConfig done for %s", imageSha)
 }
 
+func retryDownload(ctx *volumemgrContext, imageSha string) {
+	m := lookupDownloaderConfig(ctx, imageSha)
+	if m == nil {
+		log.Functionf("retryDownload: config missing for %s",
+			imageSha)
+		return
+	}
+	if m.RefCount == 0 {
+		log.Fatalf("retryDownload: Attempting to retry when "+
+			"RefCount is 0. Image Details - Name: %s, ImageSha: %s, ",
+			m.Name, m.ImageSha256)
+	}
+	m.DoRetry = true
+
+	publishDownloaderConfig(ctx, m)
+	log.Functionf("retryDownload done for %s", imageSha)
+}
+
 func publishDownloaderConfig(ctx *volumemgrContext,
 	config *types.DownloaderConfig) {
 

pkg/pillar/cmd/volumemgr/updatestatus.go

@@ -227,6 +227,22 @@ func doUpdateContentTree(ctx *volumemgrContext, status *types.ContentTreeStatus)
 				blobErrorEntities = append(blobErrorEntities, &types.ErrorEntity{EntityID: blob.Sha256, EntityType: types.ErrorEntityContentBlob})
 
 				leftToProcess = true
+				if blob.RetryCount < blobDownloadMaxRetries {
+
+					// Remove VerifyImage config and retry download
+					MaybeRemoveVerifyImageConfig(ctx, blobSha)
+					retryDownload(ctx, blobSha)
+
+					// Increment retry count
+					blob.RetryCount++
+					blob.State = types.DOWNLOADING
+					// Remove VerifyImageConfig reference
+					blob.HasVerifierRef = false
+					blob.ClearErrorWithSource()
+
+					log.Errorf("EVE failed to verify Blob(%s), retrying %d/%d ...", blobSha, blob.RetryCount, blobDownloadMaxRetries)
+					publishBlobStatus(ctx, blob)
+				}
 			}
 		}
 

pkg/pillar/cmd/volumemgr/volumemgr.go

@@ -42,7 +42,10 @@ const (
 	blankVolumeFormat = zconfig.Format_RAW // format of blank volume TODO: make configurable
 )
 
-var volumeFormat = make(map[string]zconfig.Format)
+var (
+	blobDownloadMaxRetries = 5 // Unless from GlobalConfig
+	volumeFormat           = make(map[string]zconfig.Format)
+)
 
 type volumemgrContext struct {
 	agentbase.AgentBase
@@ -746,6 +749,10 @@ func handleGlobalConfigImpl(ctxArg interface{}, key string,
 	gcp := agentlog.HandleGlobalConfig(log, ctx.subGlobalConfig, agentName,
 		ctx.CLIParams().DebugOverride, logger)
 	if gcp != nil {
+		// Set max retries for blob download from global config
+		if gcp.GlobalValueInt(types.BlobDownloadMaxRetries) != 0 {
+			blobDownloadMaxRetries = int(gcp.GlobalValueInt(types.BlobDownloadMaxRetries))
+		}
 		maybeUpdateConfigItems(ctx, gcp)
 		ctx.globalConfig = gcp
 		ctx.GCInitialized = true

pkg/pillar/types/blob.go

@@ -47,6 +47,7 @@ type BlobStatus struct {
 	Progress uint
 	// ErrorAndTimeWithSource provide common error handling capabilities
 	ErrorAndTimeWithSource
+	RetryCount int
 }
 
 const (

pkg/pillar/types/downloadertypes.go

@@ -23,6 +23,7 @@ type DownloaderConfig struct {
 	Size            uint64 // In bytes
 	FinalObjDir     string // final Object Store
 	RefCount        uint
+	DoRetry         bool
 }
 
 func (config DownloaderConfig) Key() string {

pkg/pillar/types/global.go

@@ -220,6 +220,10 @@ const (
 	// ports for image downloads.
 	DownloadMaxPortCost GlobalSettingKey = "network.download.max.cost"
 
+	// BlobDownloadMaxRetries global setting key
+	// how many times EVE will retry to download a blob if its checksum is not verified
+	BlobDownloadMaxRetries GlobalSettingKey = "blob.download.max.retries"
+
 	// Bool Items
 	// UsbAccess global setting key
 	UsbAccess GlobalSettingKey = "debug.enable.usb"
@@ -948,6 +952,7 @@ func NewConfigItemSpecMap() ConfigItemSpecMap {
 	// LogRemainToSendMBytes - Default is 2 Gbytes, minimum is 10 Mbytes
 	configItemSpecMap.AddIntItem(LogRemainToSendMBytes, 2048, 10, 0xFFFFFFFF)
 	configItemSpecMap.AddIntItem(DownloadMaxPortCost, 0, 0, 255)
+	configItemSpecMap.AddIntItem(BlobDownloadMaxRetries, 5, 1, 10)
 
 	// Goroutine Leak Detection section
 	configItemSpecMap.AddIntItem(GoroutineLeakDetectionThreshold, 5000, 1, 0xFFFFFFFF)