Resolve container-cpu-cap based on cluster type and ID

4 files changed, 28 insertions, 10 deletions

diff --git a/flags/src/main/java/com/yahoo/vespa/flags/FetchVector.java b/flags/src/main/java/com/yahoo/vespa/flags/FetchVector.java
index 37849b65adf..ede7bd6a109 100644
--- a/flags/src/main/java/com/yahoo/vespa/flags/FetchVector.java
+++ b/flags/src/main/java/com/yahoo/vespa/flags/FetchVector.java
@@ -33,9 +33,12 @@ public class FetchVector {
         /** Node type from com.yahoo.config.provision.NodeType::name, e.g. tenant, host, confighost, controller, etc. */
         NODE_TYPE,
 
-        /** Cluster type from com.yahoo.config.provision.ClusterSpec.Type::name, e.g. content, container, admin */
+        /** Cluster type from com.yahoo.config.provision.ClusterSpec.Type::value, e.g. content, container, admin */
         CLUSTER_TYPE,
 
+        /** Cluster ID from com.yahoo.config.provision.ClusterSpec.Id::value, e.g. cluster-controllers, logserver. */
+        CLUSTER_ID,
+
         /**
          * Fully qualified hostname.
          *
diff --git a/flags/src/main/java/com/yahoo/vespa/flags/PermanentFlags.java b/flags/src/main/java/com/yahoo/vespa/flags/PermanentFlags.java
index 7099124024b..251d8f67126 100644
--- a/flags/src/main/java/com/yahoo/vespa/flags/PermanentFlags.java
+++ b/flags/src/main/java/com/yahoo/vespa/flags/PermanentFlags.java
@@ -11,6 +11,8 @@ import java.time.format.DateTimeFormatter;
 import java.util.List;
 
 import static com.yahoo.vespa.flags.FetchVector.Dimension.APPLICATION_ID;
+import static com.yahoo.vespa.flags.FetchVector.Dimension.CLUSTER_ID;
+import static com.yahoo.vespa.flags.FetchVector.Dimension.CLUSTER_TYPE;
 import static com.yahoo.vespa.flags.FetchVector.Dimension.CONSOLE_USER_EMAIL;
 import static com.yahoo.vespa.flags.FetchVector.Dimension.HOSTNAME;
 import static com.yahoo.vespa.flags.FetchVector.Dimension.NODE_TYPE;
@@ -102,9 +104,9 @@ public class PermanentFlags {
     public static final UnboundDoubleFlag CONTAINER_CPU_CAP = defineDoubleFlag(
             "container-cpu-cap", 0,
             "Hard limit on how many CPUs a container may use. This value is multiplied by CPU allocated to node, so " +
-                    "to cap CPU at 200%, set this to 2, etc.",
+                    "to cap CPU at 200%, set this to 2, etc. 0 disables the cap to allow unlimited CPU.",
             "Takes effect on next node agent tick. Change is orchestrated, but does NOT require container restart",
-            HOSTNAME, APPLICATION_ID);
+            HOSTNAME, APPLICATION_ID, CLUSTER_ID, CLUSTER_TYPE);
 
     public static final UnboundListFlag<String> DISABLED_HOST_ADMIN_TASKS = defineListFlag(
             "disabled-host-admin-tasks", List.of(), String.class,
diff --git a/flags/src/main/java/com/yahoo/vespa/flags/json/DimensionHelper.java b/flags/src/main/java/com/yahoo/vespa/flags/json/DimensionHelper.java
index f109d3c950b..b18c0c43bbc 100644
--- a/flags/src/main/java/com/yahoo/vespa/flags/json/DimensionHelper.java
+++ b/flags/src/main/java/com/yahoo/vespa/flags/json/DimensionHelper.java
@@ -17,6 +17,7 @@ public class DimensionHelper {
         serializedDimensions.put(FetchVector.Dimension.HOSTNAME, "hostname");
         serializedDimensions.put(FetchVector.Dimension.APPLICATION_ID, "application");
         serializedDimensions.put(FetchVector.Dimension.NODE_TYPE, "node-type");
+        serializedDimensions.put(FetchVector.Dimension.CLUSTER_ID, "cluster-id");
         serializedDimensions.put(FetchVector.Dimension.CLUSTER_TYPE, "cluster-type");
         serializedDimensions.put(FetchVector.Dimension.VESPA_VERSION, "vespa-version");
         serializedDimensions.put(FetchVector.Dimension.CONSOLE_USER_EMAIL, "console-user-email");
diff --git a/node-admin/src/main/java/com/yahoo/vespa/hosted/node/admin/nodeagent/NodeAgentImpl.java b/node-admin/src/main/java/com/yahoo/vespa/hosted/node/admin/nodeagent/NodeAgentImpl.java
index f401b614637..ae4841f4713 100644
--- a/node-admin/src/main/java/com/yahoo/vespa/hosted/node/admin/nodeagent/NodeAgentImpl.java
+++ b/node-admin/src/main/java/com/yahoo/vespa/hosted/node/admin/nodeagent/NodeAgentImpl.java
@@ -379,13 +379,25 @@ public class NodeAgentImpl implements NodeAgent {
     }
 
     private ContainerResources getContainerResources(NodeAgentContext context) {
-        double cpuCap = noCpuCap(context.zone()) ?
-                0 :
-                context.node().owner()
-                        .map(appId -> containerCpuCap.with(FetchVector.Dimension.APPLICATION_ID, appId.serializedForm()))
-                        .orElse(containerCpuCap)
-                        .with(FetchVector.Dimension.HOSTNAME, context.node().hostname())
-                        .value() * context.vcpuOnThisHost();
+        final double cpuCap;
+        if (noCpuCap(context.zone())) {
+            cpuCap = 0.0;
+        } else {
+            DoubleFlag containerCpuCapFlag = containerCpuCap.with(FetchVector.Dimension.HOSTNAME, context.node().hostname());;
+
+            if (context.node().owner().isPresent()) {
+                containerCpuCapFlag = containerCpuCapFlag.with(FetchVector.Dimension.APPLICATION_ID,
+                                                               context.node().owner().get().serializedForm());
+            }
+
+            if (context.node().membership().isPresent()) {
+                containerCpuCapFlag = containerCpuCapFlag
+                        .with(FetchVector.Dimension.CLUSTER_TYPE, context.node().membership().get().type().value())
+                        .with(FetchVector.Dimension.CLUSTER_ID, context.node().membership().get().clusterId());
+            }
+
+            cpuCap = context.vcpuOnThisHost() * containerCpuCapFlag.value();
+        }
 
         return ContainerResources.from(cpuCap, context.vcpuOnThisHost(), context.node().memoryGb());
     }