use pooled labels array

app/vmagent/remotewrite/obfuscation.go

@@ -0,0 +1,84 @@
+package remotewrite
+
+import (
+	"crypto/sha256"
+	"encoding/hex"
+	"strings"
+	"sync"
+
+	"github.com/VictoriaMetrics/VictoriaMetrics/lib/prompb"
+	"github.com/VictoriaMetrics/VictoriaMetrics/lib/promrelabel"
+)
+
+type obfuscationCtx struct {
+	labels []prompb.Label
+}
+
+func (ctx *obfuscationCtx) Reset() {
+	promrelabel.CleanLabels(ctx.labels)
+	ctx.labels = ctx.labels[:0]
+}
+
+var obfuscationCtxPool = &sync.Pool{
+	New: func() any {
+		return &obfuscationCtx{}
+	},
+}
+
+func (rwctx *remoteWriteCtx) initObfuscationConfig() {
+	if len(*obfuscationLabels) == 0 {
+		return
+	}
+	idx := rwctx.idx
+	rwctx.obfuscationLabels = make(map[string]struct{})
+	rwObfuscationLabels := obfuscationLabels.GetOptionalArg(idx)
+	rwObfuscationLabelsList := strings.Split(rwObfuscationLabels, "^^")
+
+	for _, label := range rwObfuscationLabelsList {
+		rwctx.obfuscationLabels[label] = struct{}{}
+	}
+}
+
+func (rwctx *remoteWriteCtx) applyObfuscation(tss []prompb.TimeSeries, ctx *obfuscationCtx) []prompb.TimeSeries {
+	if len(rwctx.obfuscationLabels) == 0 || len(tss) == 0 {
+		return tss
+	}
+	cacheObfuscatedResult := make(map[string]string)
+	poolLabels := ctx.labels[:0]
+	for i := range tss {
+		ts := &tss[i]
+		labels := ts.Labels
+		j := 0
+		needToObfuscate := false
+		for ; j < len(labels); j++ {
+			label := &labels[j]
+			if _, ok := rwctx.obfuscationLabels[label.Name]; !ok {
+				continue
+			}
+			needToObfuscate = true
+			break
+		}
+		if !needToObfuscate {
+			continue
+		}
+		// Copy the label array to apply obfuscation
+		poolLabelsLen := len(poolLabels)
+		labels = append(poolLabels, labels...)
+		ts.Labels = labels[poolLabelsLen:]
+		for ; j < len(labels); j++ {
+			label := &labels[j]
+			if _, ok := rwctx.obfuscationLabels[label.Name]; !ok {
+				continue
+			}
+			if obfuscatedValue, ok := cacheObfuscatedResult[label.Value]; ok {
+				// fast path: the obfuscated result was calculated before
+				label.Value = obfuscatedValue
+			} else {
+				obfuscatedResult := sha256.Sum256([]byte(label.Value))
+				cacheObfuscatedResult[label.Value] = hex.EncodeToString(obfuscatedResult[:])
+				label.Value = cacheObfuscatedResult[label.Value]
+			}
+		}
+	}
+	return tss
+}

app/vmagent/remotewrite/remotewrite.go

@@ -102,6 +102,9 @@ var (
 		"cannot be pushed into the configured -remoteWrite.url systems in a timely manner. See https://docs.victoriametrics.com/victoriametrics/vmagent/#disabling-on-disk-persistence")
 	disableMetadataPerURL = flagutil.NewArrayBool("remoteWrite.disableMetadata", "Whether to disable sending metadata to the corresponding -remoteWrite.url. "+
 		"By default, metadata sending is controlled by the global -enableMetadata flag")
+
+	obfuscationLabels = flagutil.NewArrayString("remoteWrite.obfuscationLabels", "List of label names whose values must be obfuscated before sending to the corresponding -remoteWrite.url."+
+		"Multiple label names should be separated by `^^`, e.g. \"job^^instance,ip\". By default, label obfuscation is disabled")
 )
 
 var (
@@ -833,6 +836,8 @@ type remoteWriteCtx struct {
 	pss        []*pendingSeries
 	pssNextIdx atomic.Uint64
 
+	obfuscationLabels map[string]struct{}
+
 	rowsPushedAfterRelabel *metrics.Counter
 	rowsDroppedByRelabel   *metrics.Counter
 
@@ -937,6 +942,7 @@ func newRemoteWriteCtx(argIdx int, remoteWriteURL *url.URL, sanitizedURL string)
 		rowsDroppedOnPushFailure:     metrics.GetOrCreateCounter(fmt.Sprintf(`vmagent_remotewrite_samples_dropped_total{path=%q,url=%q}`, queuePath, sanitizedURL)),
 	}
 	rwctx.initStreamAggrConfig()
+	rwctx.initObfuscationConfig()
 
 	return rwctx
 }
@@ -1102,6 +1108,7 @@ func (rwctx *remoteWriteCtx) tryPushMetadataInternal(mms []prompb.MetricMetadata
 func (rwctx *remoteWriteCtx) tryPushTimeSeriesInternal(tss []prompb.TimeSeries) bool {
 	var rctx *relabelCtx
 	var v *[]prompb.TimeSeries
+	var octx *obfuscationCtx
 	defer func() {
 		if rctx == nil {
 			return
@@ -1120,6 +1127,24 @@ func (rwctx *remoteWriteCtx) tryPushTimeSeriesInternal(tss []prompb.TimeSeries)
 		rctx.appendExtraLabels(tss, labelsGlobal)
 	}
 
+	if len(rwctx.obfuscationLabels) != 0 {
+		if rctx == nil {
+			shadowTss := tssPool.Get().(*[]prompb.TimeSeries)
+			tss = append(*shadowTss, tss...)
+			defer func() {
+				*shadowTss = prompb.ResetTimeSeries(tss)
+				tssPool.Put(shadowTss)
+			}()
+		}
+		octx = obfuscationCtxPool.Get().(*obfuscationCtx)
+		defer func() {
+			octx.Reset()
+			obfuscationCtxPool.Put(octx)
+		}()
+
+		tss = rwctx.applyObfuscation(tss, octx)
+	}
+
 	pss := rwctx.pss
 	idx := rwctx.pssNextIdx.Add(1) % uint64(len(pss))
 

app/vmagent/remotewrite/remotewrite_test.go

@@ -1,6 +1,8 @@
 package remotewrite
 
 import (
+	"crypto/sha256"
+	"encoding/hex"
 	"fmt"
 	"math"
 	"reflect"
@@ -374,3 +376,124 @@ func TestCalculateHealthyRwctxIdx(t *testing.T) {
 	f(1, []int{0}, nil)
 	f(1, []int{}, []int{0})
 }
+
+func TestRemoteWriteObfuscation(t *testing.T) {
+	f := func(obfuscationLabelList string, inputTss []prompb.TimeSeries, expectedTss []prompb.TimeSeries) {
+		t.Helper()
+		rwctx := &remoteWriteCtx{
+			idx: 0,
+		}
+		defer metrics.UnregisterAllMetrics()
+		originValue := *obfuscationLabels
+		defer func() {
+			*obfuscationLabels = originValue
+		}()
+		*obfuscationLabels = []string{obfuscationLabelList}
+		rwctx.initObfuscationConfig()
+		octx := obfuscationCtx{}
+		outputTss := rwctx.applyObfuscation(inputTss, &octx)
+
+		if !reflect.DeepEqual(expectedTss, outputTss) {
+			t.Fatalf("unexpected samples;\ngot\n%v\nwant\n%v", outputTss, expectedTss)
+		}
+	}
+
+	sha256Result := func(str string) string {
+		sha256Result := sha256.Sum256([]byte(str))
+		return hex.EncodeToString(sha256Result[:])
+	}
+
+	// 1. obfuscation is not set.
+	f("",
+		[]prompb.TimeSeries{
+			{
+				Labels: []prompb.Label{
+					{Name: "ip", Value: "123"},
+					{Name: "instance", Value: "1234"},
+				},
+				Samples: []prompb.Sample{
+					{Value: 1, Timestamp: 0},
+				},
+			},
+		},
+		[]prompb.TimeSeries{
+			{
+				Labels: []prompb.Label{
+					{Name: "ip", Value: "123"},
+					{Name: "instance", Value: "1234"},
+				},
+				Samples: []prompb.Sample{
+					{Value: 1, Timestamp: 0},
+				},
+			},
+		},
+	)
+
+	// 2. obfuscate the value of "ip" label
+	f("ip",
+		[]prompb.TimeSeries{
+			{
+				Labels: []prompb.Label{
+					{Name: "ip", Value: "123"},
+					{Name: "instance", Value: "1234"},
+				},
+				Samples: []prompb.Sample{
+					{Value: 1, Timestamp: 0},
+				},
+			},
+		},
+		[]prompb.TimeSeries{
+			{
+				Labels: []prompb.Label{
+					{Name: "ip", Value: sha256Result("123")},
+					{Name: "instance", Value: "1234"},
+				},
+				Samples: []prompb.Sample{
+					{Value: 1, Timestamp: 0},
+				},
+			},
+		},
+	)
+
+	// 3. obfuscate the values of "ip" and "instance"
+	f("ip^^instance",
+		[]prompb.TimeSeries{
+			{
+				Labels: []prompb.Label{
+					{Name: "ip", Value: "123"},
+					{Name: "instance", Value: "1234"},
+				},
+				Samples: []prompb.Sample{
+					{Value: 1, Timestamp: 0},
+				},
+			},
+			{
+				Labels: []prompb.Label{
+					{Name: "job", Value: "123"},
+				},
+				Samples: []prompb.Sample{
+					{Value: 1, Timestamp: 0},
+				},
+			},
+		},
+		[]prompb.TimeSeries{
+			{
+				Labels: []prompb.Label{
+					{Name: "ip", Value: sha256Result("123")},
+					{Name: "instance", Value: sha256Result("1234")},
+				},
+				Samples: []prompb.Sample{
+					{Value: 1, Timestamp: 0},
+				},
+			},
+			{
+				Labels: []prompb.Label{
+					{Name: "job", Value: "123"},
+				},
+				Samples: []prompb.Sample{
+					{Value: 1, Timestamp: 0},
+				},
+			},
+		},
+	)
+}

docs/victoriametrics/changelog/CHANGELOG.md

@@ -25,9 +25,9 @@ The sandbox cluster installation runs under the constant load generated by
 See also [LTS releases](https://docs.victoriametrics.com/victoriametrics/lts-releases/).
 
 ## tip
+* FEATURE: [vmagent](https://docs.victoriametrics.com/victoriametrics/vmagent/): introduce obfuscation functionality for remote write. By setting `-remoteWrite.obfuscationLabels`, the values of the specific labels will be anonymized before they're sent to corresponding `-remoteWrite.url`. See [#10599](https://github.com/VictoriaMetrics/VictoriaMetrics/issues/10599).
 
 ## [v1.140.0](https://github.com/VictoriaMetrics/VictoriaMetrics/releases/tag/v1.140.0)
-
 Released at 2026-04-10
 
 **Update Note 1:** [vmsingle](https://docs.victoriametrics.com/victoriametrics/single-server-victoriametrics/) and `vmselect` in [VictoriaMetrics cluster](https://docs.victoriametrics.com/victoriametrics/cluster-victoriametrics/): [CSV export](https://docs.victoriametrics.com/victoriametrics/single-server-victoriametrics/#how-to-export-csv-data) (`/api/v1/export/csv`) now adds a header row as the first line of the response, so existing CSV-processing scripts may need to skip this header. See [#10666](https://github.com/VictoriaMetrics/VictoriaMetrics/issues/10666).

docs/victoriametrics/vmagent.md

@@ -289,14 +289,16 @@ flowchart TB
       H1 --> H2[per-url <a href="https://docs.victoriametrics.com/victoriametrics/stream-aggregation">aggregation</a><br><b>-remoteWrite.streamAggr.config</b><br><b>-remoteWrite.streamAggr.dedupInterval</b>]
       H2 --> H3["per-url <a href="https://docs.victoriametrics.com/victoriametrics/vmagent/#calculating-disk-space-for-persistence-queue">queue</a> (default: enabled)<br><b>-remoteWrite.disableOnDiskQueue</b>"]
       H3 --> H4[<a href="https://docs.victoriametrics.com/victoriametrics/vmagent/#adding-labels-to-metrics">add extra labels</a><br><b>-remoteWrite.label</b>]
-      H4 --> H5[[push to <b>-remoteWrite.url</b>]]
+      H4 --> H5[<a href="https://docs.victoriametrics.com/victoriametrics/vmagent/#obfuscating-label-values">obfuscate labels</a><br><b>-remoteWrite.obfuscationLabels</b>]
+      H5 --> H6[[push to <b>-remoteWrite.url</b>]]
 
       %% Right branch
       G --> R1[per-url <a href="https://docs.victoriametrics.com/victoriametrics/relabeling/">relabeling</a><br><b>-remoteWrite.urlRelabelConfig</b>]
       R1 --> R2[per-url <a href="https://docs.victoriametrics.com/victoriametrics/stream-aggregation">aggregation</a><br><b>-remoteWrite.streamAggr.config</b><br><b>-remoteWrite.streamAggr.dedupInterval</b>]
       R2 --> R3["per-url <a href="https://docs.victoriametrics.com/victoriametrics/vmagent/#calculating-disk-space-for-persistence-queue">queue</a> (default: enabled)<br><b>-remoteWrite.disableOnDiskQueue</b>"]
       R3 --> R4[<a href="https://docs.victoriametrics.com/victoriametrics/vmagent/#adding-labels-to-metrics">add extra labels</a><br><b>-remoteWrite.label</b>]
-      R4 --> R5[[push to <b>-remoteWrite.url</b>]]
+      R4 --> R5[<a href="https://docs.victoriametrics.com/victoriametrics/vmagent/#obfuscating-label-values">obfuscate labels</a><br><b>-remoteWrite.obfuscationLabels</b>]
+      R5 --> R6[[push to <b>-remoteWrite.url</b>]]
 ```
 
 Scraping has additional settings that can be applied before samples are pushed to the processing pipeline above:
@@ -529,6 +531,24 @@ Extra labels can be added to metrics collected by `vmagent` via the following me
   /path/to/vmagent -remoteWrite.url=http://127.0.0.1:8428/api/v1/write?extra_label="env=prod"
   ```
 
+## Obfuscating label values
+
+Before sending metrics to `-remoteWrite.url`, `vmagent` can anonymize the values of specific labels in the metrics via setting `-remoteWrite.obfuscationLabels`.
+Sometimes, some of the `-remoteWrite.url` may point to external services, such as monitoring service vendor outside the department or company. For security and compliance requirements, 
+obfuscating the specific label values (e.g. ip, host, datacenter, etc.) before sending them to these external services will be useful.
+
+Use `-remoteWrite.obfuscationLabels` to specify the labels that need to be obfuscated before sending to `-remoteWrite.url`. Multiple labels should be separated by `^^`:
+
+```sh
+./vmagent \
+  -remoteWrite.url=http://<external-service1> \
+  -remoteWrite.obfuscationLabels='instance^^datacenter' \
+  -remoteWrite.url=http://<external-service2> \
+  -remoteWrite.obfuscationLabels='instance' \
+  -remoteWrite.url=http://<internal-service> \
+  -remoteWrite.obfuscationLabels='' 
+```
+
 ## Automatically generated metrics
 
 `vmagent` automatically generates the following metrics per each scrape of every [Prometheus-compatible target](#how-to-collect-metrics-in-prometheus-format)