Merge #150398

150398: cli(debug.zip): improve tsdump upload speed r=aa-joshi a=aa-joshi

Previously, tsdump upload to Datadog was taking more time compare to stage tsdump in roachprod. This was inadequate because it would increase the  MTTD (Mean Time To Detect) an issue. This change introduces `upload-workers` as flag to set the number of Datadog upload workers. The default value is 50. This change includes the changes around the retry configuration to further improve performance.

Epic: None
Part of: CRDB-52094
Release note: None

-----
tsdump size: 11.19GB

roachprod upload time:
<img width="980" height="77" alt="Screenshot 2025-07-17 at 12 04 01 PM" src="https://github.com/user-attachments/assets/718d5f2b-142e-438c-9008-6739295e6930" />

Tsdump upload time (before changes):
<img width="904" height="17" alt="Screenshot 2025-07-17 at 11 52 37 AM" src="https://github.com/user-attachments/assets/9b844a58-7b5b-4e67-bad0-142f6c37673b" />

Tsdump upload time with default(after changes):
<img width="903" height="20" alt="Screenshot 2025-07-17 at 11 51 34 AM" src="https://github.com/user-attachments/assets/e6556e2a-d9bd-411b-acd1-591cc9971784" />


Co-authored-by: Akshay Joshi <[email protected]>

Author: craig[bot]

pkg/cli/debug.go

@@ -1617,6 +1617,7 @@ func init() {
 	f.StringVar(&debugTimeSeriesDumpOpts.userName, "user-name", "", "name of the user to perform datadog upload")
 	f.StringVar(&debugTimeSeriesDumpOpts.storeToNodeMapYAMLFile, "store-to-node-map-file", "", "yaml file path which contains the mapping of store ID to node ID for datadog upload.")
 	f.BoolVar(&debugTimeSeriesDumpOpts.dryRun, "dry-run", false, "run in dry-run mode without making any actual uploads")
+	f.IntVar(&debugTimeSeriesDumpOpts.noOfUploadWorkers, "upload-workers", 50, "number of workers to upload the time series data in parallel")
 
 	f = debugSendKVBatchCmd.Flags()
 	f.StringVar(&debugSendKVBatchContext.traceFormat, "trace", debugSendKVBatchContext.traceFormat,

pkg/cli/tsdump.go

@@ -52,6 +52,7 @@ var debugTimeSeriesDumpOpts = struct {
 	userName               string
 	storeToNodeMapYAMLFile string
 	dryRun                 bool
+	noOfUploadWorkers      int
 }{
 	format:       tsDumpText,
 	from:         timestampValue{},
@@ -121,6 +122,7 @@ will then convert it to the --format requested in the current invocation.
 				debugTimeSeriesDumpOpts.ddApiKey,
 				100,
 				hostNameOverride,
+				debugTimeSeriesDumpOpts.noOfUploadWorkers,
 			)
 			if err != nil {
 				return err

pkg/cli/tsdump_test.go

@@ -216,11 +216,12 @@ func TestTsDumpFormatsDataDriven(t *testing.T) {
 				debugTimeSeriesDumpOpts.zendeskTicket = "zd-test"
 				debugTimeSeriesDumpOpts.organizationName = "test-org"
 				debugTimeSeriesDumpOpts.userName = "test-user"
+				debugTimeSeriesDumpOpts.noOfUploadWorkers = 50
 				var series int
 				d.ScanArgs(t, "series-threshold", &series)
 				var ddwriter, err = makeDatadogWriter(
 					defaultDDSite, d.Cmd == "format-datadog-init", "api-key", series,
-					server.Listener.Addr().String(),
+					server.Listener.Addr().String(), debugTimeSeriesDumpOpts.noOfUploadWorkers,
 				)
 				require.NoError(t, err)
 

pkg/cli/tsdump_upload.go

@@ -22,8 +22,10 @@ import (
 
 	"github.com/DataDog/datadog-api-client-go/v2/api/datadog"
 	"github.com/DataDog/datadog-api-client-go/v2/api/datadogV2"
+	"github.com/cockroachdb/cockroach/pkg/base"
 	"github.com/cockroachdb/cockroach/pkg/roachpb"
 	"github.com/cockroachdb/cockroach/pkg/ts"
+	"github.com/cockroachdb/cockroach/pkg/util/retry"
 	"github.com/cockroachdb/cockroach/pkg/util/syncutil"
 	"github.com/cockroachdb/cockroach/pkg/util/timeutil"
 	"github.com/cockroachdb/errors"
@@ -89,15 +91,21 @@ type datadogWriter struct {
 	datadogContext context.Context
 	// namePrefix sets the string to prepend to all metric names. The
 	// names are kept with `.` delimiters.
-	namePrefix     string
-	threshold      int
-	uploadTime     time.Time
-	storeToNodeMap map[string]string
-	metricTypeMap  map[string]string
+	namePrefix        string
+	threshold         int
+	uploadTime        time.Time
+	storeToNodeMap    map[string]string
+	metricTypeMap     map[string]string
+	noOfUploadWorkers int
 }
 
 func makeDatadogWriter(
-	ddSite string, init bool, apiKey string, threshold int, hostNameOverride string,
+	ddSite string,
+	init bool,
+	apiKey string,
+	threshold int,
+	hostNameOverride string,
+	noOfUploadWorkers int,
 ) (*datadogWriter, error) {
 	currentTime := getCurrentTime()
 
@@ -128,6 +136,9 @@ func makeDatadogWriter(
 	ctx = context.WithValue(ctx, datadog.ContextServerVariables, map[string]string{
 		"site": host,
 	})
+
+	// The Datadog retry configuration is used when we receive error codes
+	// 429 and >= 500 from the Datadog.
 	configuration := datadog.NewConfiguration()
 	configuration.RetryConfiguration.EnableRetry = true
 	configuration.RetryConfiguration.BackOffMultiplier = 1
@@ -140,16 +151,17 @@ func makeDatadogWriter(
 	}
 
 	return &datadogWriter{
-		datadogContext: ctx,
-		apiClient:      apiClient,
-		apiKey:         apiKey,
-		uploadID:       newTsdumpUploadID(currentTime),
-		init:           init,
-		namePrefix:     "crdb.tsdump.", // Default pre-set prefix to distinguish these uploads.
-		threshold:      threshold,
-		uploadTime:     currentTime,
-		storeToNodeMap: make(map[string]string),
-		metricTypeMap:  metricTypeMap,
+		datadogContext:    ctx,
+		apiClient:         apiClient,
+		apiKey:            apiKey,
+		uploadID:          newTsdumpUploadID(currentTime),
+		init:              init,
+		namePrefix:        "crdb.tsdump.", // Default pre-set prefix to distinguish these uploads.
+		threshold:         threshold,
+		uploadTime:        currentTime,
+		storeToNodeMap:    make(map[string]string),
+		metricTypeMap:     metricTypeMap,
+		noOfUploadWorkers: noOfUploadWorkers,
 	}, nil
 }
 
@@ -346,11 +358,25 @@ func (d *datadogWriter) flush(data []datadogV2.MetricSeries) error {
 	}
 
 	api := datadogV2.NewMetricsApi(d.apiClient)
-	_, _, err := api.SubmitMetrics(d.datadogContext, datadogV2.MetricPayload{
-		Series: data,
-	}, datadogV2.SubmitMetricsOptionalParameters{
-		ContentEncoding: datadogV2.METRICCONTENTENCODING_GZIP.Ptr(),
-	})
+	// The retry configuration is used when we receive any error code from upload.
+	// We have seen 408 error codes from Datadog when the upload is too large which
+	// is not handled by the default retry configuration that Datadog API client provides.
+	retryOpts := base.DefaultRetryOptions()
+	retryOpts.MaxBackoff = 20 * time.Millisecond
+	retryOpts.MaxRetries = 100
+	err := error(nil)
+
+	for retryAttempts := retry.Start(retryOpts); retryAttempts.Next(); {
+		_, _, err = api.SubmitMetrics(d.datadogContext, datadogV2.MetricPayload{
+			Series: data,
+		}, datadogV2.SubmitMetricsOptionalParameters{
+			ContentEncoding: datadogV2.METRICCONTENTENCODING_GZIP.Ptr(),
+		})
+
+		if err == nil {
+			return nil
+		}
+	}
 	if err != nil {
 		fmt.Printf("error submitting metrics to datadog: %v\n", err)
 	}
@@ -417,11 +443,7 @@ func (d *datadogWriter) upload(fileName string) error {
 		metricsUploadState.isSingleUploadSucceeded = true
 	})
 
-	// Note(davidh): This was previously set at 1000 and we'd get regular
-	// 400s from Datadog with the cryptic `Unable to decompress payload`
-	// error. We reduced this to 20 and was able to upload a 3.2GB tsdump
-	// in 6m20s without any errors.
-	for i := 0; i < 20; i++ {
+	for i := 0; i < d.noOfUploadWorkers; i++ {
 		go func() {
 			for data := range ch {
 				emittedMetrics, err := d.emitDataDogMetrics(data)