feat(cchain): improve lookback metrics (#1604)

Include the chain name, and set better buckets for these metrics. issue: #1526
omni-network · Jul 30, 2024 · 216cec3 · 216cec3
1 parent b5e8d10
commit 216cec3
Show file tree

Hide file tree

Showing 3 changed files with 19 additions and 15 deletions.
diff --git a/e2e/test/attestations_test.go b/e2e/test/attestations_test.go
@@ -23,7 +23,7 @@ func TestApprovedAttestations(t *testing.T) {
 		t.Helper()
 		client, err := node.Client()
 		require.NoError(t, err)
-		cprov := provider.NewABCIProvider(client, network.ID, nil)
+		cprov := provider.NewABCIProvider(client, network.ID, netconf.ChainVersionNamer(netconf.Simnet))
 
 		ctx := context.Background()
 		for _, portal := range portals {
@@ -73,7 +73,7 @@ func TestApprovedValUpdates(t *testing.T) {
 
 		client, err := node.Client()
 		require.NoError(t, err)
-		cprov := provider.NewABCIProvider(client, network.ID, nil)
+		cprov := provider.NewABCIProvider(client, network.ID, netconf.ChainVersionNamer(netconf.Simnet))
 
 		addr, err := k1util.PubKeyToAddress(node.PrivvalKey.PubKey())
 		require.NoError(t, err)

diff --git a/lib/cchain/provider/abci.go b/lib/cchain/provider/abci.go
@@ -40,7 +40,7 @@ func NewABCIProvider(abci rpcclient.Client, network netconf.ID, chainNamer func(
 	gcl := genserve.NewQueryClient(rpcAdaptor{abci: abci})
 
 	return Provider{
-		fetch:       newABCIFetchFunc(acl, abci),
+		fetch:       newABCIFetchFunc(acl, abci, chainNamer),
 		latest:      newABCILatestFunc(acl),
 		window:      newABCIWindowFunc(acl),
 		valset:      newABCIValsetFunc(vcl),
@@ -121,14 +121,16 @@ func newABCIValsetFunc(cl vtypes.QueryClient) valsetFunc {
 	}
 }
 
-func newABCIFetchFunc(cl atypes.QueryClient, client rpcclient.Client) fetchFunc {
+func newABCIFetchFunc(cl atypes.QueryClient, client rpcclient.Client, chainNamer func(xchain.ChainVersion) string) fetchFunc {
 	return func(ctx context.Context, chainVer xchain.ChainVersion, fromOffset uint64) ([]xchain.Attestation, error) {
 		const endpoint = "fetch_attestations"
 		defer latency(endpoint)()
 
 		ctx, span := tracer.Start(ctx, spanName(endpoint))
 		defer span.End()
 
+		chainName := chainNamer(chainVer)
+
 		// try fetching from latest height
 		atts, ok, err := attsFromAtHeight(ctx, cl, chainVer, fromOffset, 0)
 		if err != nil {
@@ -137,7 +139,7 @@ func newABCIFetchFunc(cl atypes.QueryClient, client rpcclient.Client) fetchFunc
 		}
 
 		if ok {
-			fetchStepsMetrics(0, 0)
+			fetchStepsMetrics(chainName, 0, 0)
 			return atts, nil
 		}
 
@@ -154,7 +156,7 @@ func newABCIFetchFunc(cl atypes.QueryClient, client rpcclient.Client) fetchFunc
 			return []xchain.Attestation{}, nil
 		}
 
-		offsetHeight, err := searchOffsetInHistory(ctx, client, cl, chainVer, fromOffset)
+		offsetHeight, err := searchOffsetInHistory(ctx, client, cl, chainVer, chainName, fromOffset)
 		if err != nil {
 			incQueryErr(endpoint)
 			return nil, errors.Wrap(err, "searching offset in history")
@@ -340,7 +342,7 @@ func spanName(endpoint string) string {
 // searchOffsetInHistory searches the consensus state history and
 // returns a historical consensus block height that contains an approved attestation
 // for the provided chain version and fromOffset.
-func searchOffsetInHistory(ctx context.Context, client rpcclient.Client, cl atypes.QueryClient, chainVer xchain.ChainVersion, fromOffset uint64) (uint64, error) {
+func searchOffsetInHistory(ctx context.Context, client rpcclient.Client, cl atypes.QueryClient, chainVer xchain.ChainVersion, chainName string, fromOffset uint64) (uint64, error) {
 	const endpoint = "search_offset"
 	defer latency(endpoint)
 
@@ -435,7 +437,7 @@ func searchOffsetInHistory(ctx context.Context, client rpcclient.Client, cl atyp
 		}
 
 		if fromOffset >= earliestAtt.AttestOffset && fromOffset <= latestAtt.AttestOffset {
-			fetchStepsMetrics(lookbackStepsCounter, binarySearchStepsCounter)
+			fetchStepsMetrics(chainName, lookbackStepsCounter, binarySearchStepsCounter)
 			return midHeightIndex, nil
 		}
 

diff --git a/lib/cchain/provider/metrics.go b/lib/cchain/provider/metrics.go
@@ -52,24 +52,26 @@ var (
 		Help:      "Total number of query errors per endpoint. Alert if growing.",
 	}, []string{"endpoint"})
 
-	fetchLookbackSteps = promauto.NewHistogram(prometheus.HistogramOpts{
+	fetchLookbackSteps = promauto.NewHistogramVec(prometheus.HistogramOpts{
 		Namespace: "lib",
 		Subsystem: "cprovider",
 		Name:      "fetch_lookback_steps",
+		Buckets:   []float64{0, 1, 2, 4, 8, 16, 32, 64, 128, 256},
 		Help:      "Number of steps in the exponential backoff process to find a start for binary search",
-	})
+	}, []string{"chain_version"})
 
-	fetchBinarySearchSteps = promauto.NewHistogram(prometheus.HistogramOpts{
+	fetchBinarySearchSteps = promauto.NewHistogramVec(prometheus.HistogramOpts{
 		Namespace: "lib",
 		Subsystem: "cprovider",
 		Name:      "fetch_binary_search_steps",
+		Buckets:   []float64{0, 1, 2, 4, 8, 16, 32, 64, 128, 256},
 		Help:      "Number of steps in the binary search process to find the right height",
-	})
+	}, []string{"chain_version"})
 )
 
-func fetchStepsMetrics(lookbackSteps, binarySearchSteps uint64) {
-	fetchLookbackSteps.Observe(float64(lookbackSteps))
-	fetchBinarySearchSteps.Observe(float64(binarySearchSteps))
+func fetchStepsMetrics(chainName string, lookbackSteps, binarySearchSteps uint64) {
+	fetchLookbackSteps.WithLabelValues(chainName).Observe(float64(lookbackSteps))
+	fetchBinarySearchSteps.WithLabelValues(chainName).Observe(float64(binarySearchSteps))
 }
 
 func latency(endpoint string) func() {