Merge pull request vespa-engine#32314 from vespa-engine/arnej/add-log…

…server-memory-feature-flag add feature flag for tuning memory on logserver nodes
baldersheim · Sep 3, 2024 · 6c04a09 · 6c04a09
2 parents 2764ffd + 045e9c0
commit 6c04a09
Show file tree

Hide file tree

Showing 7 changed files with 41 additions and 8 deletions.
diff --git a/config-model-api/abi-spec.json b/config-model-api/abi-spec.json
@@ -1346,6 +1346,7 @@
       "public boolean logserverOtelCol()",
       "public com.yahoo.config.provision.SharedHosts sharedHosts()",
       "public com.yahoo.config.provision.NodeResources$Architecture adminClusterArchitecture()",
+      "public double logserverNodeMemory()",
       "public boolean symmetricPutAndActivateReplicaSelection()",
       "public boolean enforceStrictlyIncreasingClusterStateVersions()",
       "public boolean distributionConfigFromClusterController()",

diff --git a/config-model-api/src/main/java/com/yahoo/config/model/api/ModelContext.java b/config-model-api/src/main/java/com/yahoo/config/model/api/ModelContext.java
@@ -119,6 +119,7 @@ interface FeatureFlags {
         @ModelFeatureFlag(owners = {"olaa"}) default boolean logserverOtelCol() { return false; }
         @ModelFeatureFlag(owners = {"bratseth"}) default SharedHosts sharedHosts() { return SharedHosts.empty(); }
         @ModelFeatureFlag(owners = {"bratseth"}) default Architecture adminClusterArchitecture() { return Architecture.x86_64; }
+        @ModelFeatureFlag(owners = {"arnej"}) default double logserverNodeMemory() { return 0.0; }
         @ModelFeatureFlag(owners = {"vekterli"}) default boolean symmetricPutAndActivateReplicaSelection() { return false; }
         @ModelFeatureFlag(owners = {"vekterli"}) default boolean enforceStrictlyIncreasingClusterStateVersions() { return false; }
         @ModelFeatureFlag(owners = {"vekterli"}) default boolean distributionConfigFromClusterController() { return false; }

diff --git a/config-model/src/main/java/com/yahoo/vespa/model/application/validation/QuotaValidator.java b/config-model/src/main/java/com/yahoo/vespa/model/application/validation/QuotaValidator.java
@@ -34,8 +34,10 @@ public class QuotaValidator implements Validator {
     public void validate(Context context) {
         var zone = context.deployState().zone();
         var exclusivity = new Exclusivity(zone, context.deployState().featureFlags().sharedHosts());
+        var tuning = new CapacityPolicies.Tuning(context.deployState().featureFlags().adminClusterArchitecture(),
+                                                 context.deployState().featureFlags().logserverNodeMemory());
         var capacityPolicies = new CapacityPolicies(zone, exclusivity, context.model().applicationPackage().getApplicationId(),
-                                                    context.deployState().featureFlags().adminClusterArchitecture());
+                                                    tuning);
         var quota = context.deployState().getProperties().quota();
         quota.maxClusterSize().ifPresent(maxClusterSize -> validateMaxClusterSize(maxClusterSize, context.model()));
         quota.budgetAsDecimal().ifPresent(budget -> validateBudget(budget, context, capacityPolicies));

diff --git a/config-provisioning/src/main/java/com/yahoo/config/provision/CapacityPolicies.java b/config-provisioning/src/main/java/com/yahoo/config/provision/CapacityPolicies.java
@@ -17,16 +17,29 @@
  */
 public class CapacityPolicies {
 
+    public record Tuning(Architecture adminClusterArchitecture,
+                         double logserverMemoryGiB)
+    {
+        double logserverMem(double v) {
+            double override = logserverMemoryGiB();
+            return (override > 0) ? override : v;
+        }
+    }
+
     private final Zone zone;
     private final Exclusivity exclusivity;
     private final ApplicationId applicationId;
-    private final Architecture adminClusterArchitecture;
+    private final Tuning tuning;
 
     public CapacityPolicies(Zone zone, Exclusivity exclusivity, ApplicationId applicationId, Architecture adminClusterArchitecture) {
+        this(zone, exclusivity, applicationId, new Tuning(adminClusterArchitecture, 0.0));
+    }
+
+    public CapacityPolicies(Zone zone, Exclusivity exclusivity, ApplicationId applicationId, Tuning tuning) {
         this.zone = zone;
         this.exclusivity = exclusivity;
         this.applicationId = applicationId;
-        this.adminClusterArchitecture = adminClusterArchitecture;
+        this.tuning = tuning;
     }
 
     public Capacity applyOn(Capacity capacity, boolean exclusive) {
@@ -92,6 +105,7 @@ public NodeResources specifyFully(NodeResources resources, ClusterSpec clusterSp
     }
 
     private NodeResources defaultResources(ClusterSpec clusterSpec) {
+        var adminClusterArchitecture = tuning.adminClusterArchitecture();
         if (clusterSpec.type() == ClusterSpec.Type.admin) {
             if (exclusivity.allocation(clusterSpec)) {
                 return smallestExclusiveResources().with(adminClusterArchitecture);
@@ -134,14 +148,14 @@ private NodeResources clusterControllerResources(ClusterSpec clusterSpec, Archit
 
     private NodeResources logserverResources(Architecture architecture) {
         if (zone.cloud().name() == CloudName.AZURE)
-            return new NodeResources(2, 4, 50, 0.3);
+            return new NodeResources(2, tuning.logserverMem(4.0), 50, 0.3);
 
         if (zone.cloud().name() == CloudName.GCP)
-            return new NodeResources(1, 4, 50, 0.3);
+            return new NodeResources(1, tuning.logserverMem(4.0), 50, 0.3);
 
         return architecture == Architecture.arm64
-                ? new NodeResources(0.5, 2.5, 50, 0.3)
-                : new NodeResources(0.5, 2, 50, 0.3);
+                ? new NodeResources(0.5, tuning.logserverMem(2.5), 50, 0.3)
+                : new NodeResources(0.5, tuning.logserverMem(2.0), 50, 0.3);
     }
 
     // The lowest amount of resources that can be exclusive allocated (i.e. a matching host flavor for this exists)

diff --git a/configserver/src/main/java/com/yahoo/vespa/config/server/deploy/ModelContextImpl.java b/configserver/src/main/java/com/yahoo/vespa/config/server/deploy/ModelContextImpl.java
@@ -208,6 +208,7 @@ public static class FeatureFlags implements ModelContext.FeatureFlags {
         private final boolean logserverOtelCol;
         private final SharedHosts sharedHosts;
         private final Architecture adminClusterArchitecture;
+        private final double logserverNodeMemory;
         private final boolean symmetricPutAndActivateReplicaSelection;
         private final boolean enforceStrictlyIncreasingClusterStateVersions;
         private final boolean launchApplicationAthenzService;
@@ -258,6 +259,7 @@ public FeatureFlags(FlagSource source, ApplicationId appId, Version version) {
             this.logserverOtelCol = Flags.LOGSERVER_OTELCOL_AGENT.bindTo(source).with(appId).with(version).value();
             this.sharedHosts = PermanentFlags.SHARED_HOST.bindTo(source).with( appId).with(version).value();
             this.adminClusterArchitecture = Architecture.valueOf(PermanentFlags.ADMIN_CLUSTER_NODE_ARCHITECTURE.bindTo(source).with(appId).with(version).value());
+            this.logserverNodeMemory = PermanentFlags.LOGSERVER_NODE_MEMORY.bindTo(source).with(appId).with(version).value();
             this.symmetricPutAndActivateReplicaSelection = Flags.SYMMETRIC_PUT_AND_ACTIVATE_REPLICA_SELECTION.bindTo(source).with(appId).with(version).value();
             this.enforceStrictlyIncreasingClusterStateVersions = Flags.ENFORCE_STRICTLY_INCREASING_CLUSTER_STATE_VERSIONS.bindTo(source).with(appId).with(version).value();
             this.launchApplicationAthenzService = Flags.LAUNCH_APPLICATION_ATHENZ_SERVICE.bindTo(source).with(appId).with(version).value();
@@ -314,6 +316,7 @@ public FeatureFlags(FlagSource source, ApplicationId appId, Version version) {
         @Override public boolean logserverOtelCol() { return logserverOtelCol; }
         @Override public SharedHosts sharedHosts() { return sharedHosts; }
         @Override public Architecture adminClusterArchitecture() { return adminClusterArchitecture; }
+        @Override public double logserverNodeMemory() { return logserverNodeMemory; }
         @Override public boolean symmetricPutAndActivateReplicaSelection() { return symmetricPutAndActivateReplicaSelection; }
         @Override public boolean enforceStrictlyIncreasingClusterStateVersions() { return enforceStrictlyIncreasingClusterStateVersions; }
         @Override public boolean distributionConfigFromClusterController() { return distributionConfigFromClusterController; }

diff --git a/flags/src/main/java/com/yahoo/vespa/flags/PermanentFlags.java b/flags/src/main/java/com/yahoo/vespa/flags/PermanentFlags.java
@@ -329,6 +329,12 @@ public class PermanentFlags {
             value -> Set.of("any", "arm64", "x86_64").contains(value),
             INSTANCE_ID);
 
+    public static final UnboundDoubleFlag LOGSERVER_NODE_MEMORY = defineDoubleFlag(
+            "logserver-node-memory", 0.0,
+            "Amount of memory (in GiB) to allocate for logserver nodes",
+            "Takes effect on allocation from node repository",
+            INSTANCE_ID);
+
     public static final UnboundListFlag<String> CLOUD_ACCOUNTS = defineListFlag(
             "cloud-accounts", List.of(), String.class,
             "A list of 12-digit AWS account IDs that are valid for the given tenant",

diff --git a/node-repository/src/main/java/com/yahoo/vespa/hosted/provision/NodeRepository.java b/node-repository/src/main/java/com/yahoo/vespa/hosted/provision/NodeRepository.java
@@ -215,7 +215,13 @@ public CapacityPolicies capacityPoliciesFor(ApplicationId applicationId) {
                 .bindTo(flagSource)
                 .with(INSTANCE_ID, applicationId.serializedForm())
                 .value();
-        return new CapacityPolicies(zone, exclusivity(), applicationId, Architecture.valueOf(adminClusterNodeArchitecture));
+        double logserverMemory = PermanentFlags.LOGSERVER_NODE_MEMORY
+                .bindTo(flagSource)
+                .with(INSTANCE_ID, applicationId.serializedForm())
+                .value();
+        var tuning = new CapacityPolicies.Tuning(Architecture.valueOf(adminClusterNodeArchitecture),
+                                                 logserverMemory);
+        return new CapacityPolicies(zone, exclusivity(), applicationId, tuning);
     }
 
     /**