create a CDS cluster per model (#5916)

SeldonIO · Sep 18, 2024 · 36f6189 · 36f6189
1 parent d28c58c
commit 36f6189
Show file tree

Hide file tree

Showing 3 changed files with 5 additions and 38 deletions.
diff --git a/scheduler/pkg/envoy/processor/hash.go b/scheduler/pkg/envoy/processor/hash.go
diff --git a/scheduler/pkg/envoy/processor/incremental.go b/scheduler/pkg/envoy/processor/incremental.go
@@ -196,7 +196,6 @@ func (p *IncrementalProcessor) setListeners() error {
 // newSnapshotVersion increments the current snapshotVersion
 // and returns as a string.
 func (p *IncrementalProcessor) newSnapshotVersion() string {
-
 	// Reset the snapshotVersion if it ever hits max size.
 	if p.snapshotVersion == math.MaxInt64 {
 		p.snapshotVersion = 0
@@ -256,7 +255,7 @@ func (p *IncrementalProcessor) updateEnvoyForModelVersion(modelRouteName string,
 		return
 	}
 
-	clusterNameBase := server.Name + "_" + computeHashKeyForList(assignment)
+	clusterNameBase := modelVersion.GetMeta().GetName() + "_" + strconv.FormatInt(int64(modelVersion.GetVersion()), 10)
 	httpClusterName := clusterNameBase + "_http"
 	grpcClusterName := clusterNameBase + "_grpc"
 	p.xdsCache.AddCluster(httpClusterName, modelRouteName, modelVersion.GetModel().GetMeta().GetName(), modelVersion.GetVersion(), false)

diff --git a/scheduler/pkg/envoy/processor/incremental_test.go b/scheduler/pkg/envoy/processor/incremental_test.go
@@ -229,7 +229,7 @@ func TestUpdateEnvoyForModelVersion(t *testing.T) {
 			},
 			traffic:          100,
 			expectedRoutes:   2,
-			expectedClusters: 2,
+			expectedClusters: 4,
 		},
 		{
 			name: "TwoRoutesDifferentClusters",
@@ -593,7 +593,7 @@ func TestEnvoySettings(t *testing.T) {
 				createTestModel("model3", "server", 1, []int{1}, 1, []store.ModelReplicaState{store.Available}),
 				createTestExperiment("exp", []string{"model1", "model2"}, getStrPtr("model1"), getStrPtr("model3")),
 			},
-			numExpectedClusters: 4,
+			numExpectedClusters: 6,
 			numExpectedRoutes:   4,
 			experimentActive:    true,
 			experimentExists:    true,
@@ -607,7 +607,7 @@ func TestEnvoySettings(t *testing.T) {
 				createTestModel("model3", "server", 1, []int{1}, 1, []store.ModelReplicaState{store.Available}),
 				createTestPipeline("pipe", []string{"model1", "model2", "model3"}, 1),
 			},
-			numExpectedClusters:  4,
+			numExpectedClusters:  6,
 			numExpectedRoutes:    3,
 			numExpectedPipelines: 1,
 		},
@@ -710,7 +710,7 @@ func TestRollingUpdate(t *testing.T) {
 				createTestModel("model", "server", 1, []int{0}, 1, []store.ModelReplicaState{store.Available}),
 				createTestModel("model", "server", 2, []int{0, 1}, 2, []store.ModelReplicaState{store.Available, store.Loading}),
 			},
-			numExpectedClusters: 2,
+			numExpectedClusters: 4,
 			numExpectedRoutes:   1,
 			numTrafficSplits:    map[string]int{"model": 2},
 		},