Add validator validating heap size calculation

2 files changed, 52 insertions, 0 deletions

diff --git a/config-model/src/main/java/com/yahoo/vespa/model/application/validation/JvmHeapSizeValidator.java b/config-model/src/main/java/com/yahoo/vespa/model/application/validation/JvmHeapSizeValidator.java
new file mode 100644
index 00000000000..2c5e0db14b9
--- /dev/null
+++ b/config-model/src/main/java/com/yahoo/vespa/model/application/validation/JvmHeapSizeValidator.java
@@ -0,0 +1,51 @@
+// Copyright Yahoo. Licensed under the terms of the Apache 2.0 license. See LICENSE in the project root.
+
+package com.yahoo.vespa.model.application.validation;
+
+import com.yahoo.config.model.deploy.DeployState;
+import com.yahoo.vespa.model.VespaModel;
+
+import java.util.logging.Level;
+
+/**
+ * Validates that the container node flavour has enough resources to run configured ONNX models.
+ *
+ * @author bjorncs
+ */
+public class JvmHeapSizeValidator extends Validator {
+
+    @Override
+    public void validate(VespaModel model, DeployState ds) {
+        if (!ds.featureFlags().dynamicHeapSize()) return;
+        if (!ds.isHostedTenantApplication(model.getAdmin().getApplicationType())) return;
+
+        model.getContainerClusters().forEach((clusterId, appCluster) -> {
+            var mp = appCluster.getMemoryPercentage().orElse(null);
+            if (mp == null) return;
+            if (mp.availableMemoryGb().isEmpty()) {
+                ds.getDeployLogger().log(Level.FINE, "Host resources unknown or percentage overridden with 'allocated-memory'");
+                return;
+            }
+            long jvmModelCost = appCluster.onnxModelCost().aggregatedModelCostInBytes();
+            if (jvmModelCost > 0) {
+                int percentLimit = 10;
+                if (mp.percentage() < percentLimit) {
+                    throw new IllegalArgumentException(
+                            ("Allocated percentage of memory of JVM in cluster '%s' is too low (%d%% < %d%%). " +
+                                    "Estimated cost of ONNX models is %.2fGB. Either use a node flavor with more memory or use less expensive models. " +
+                                    "You may override this validation by specifying 'allocated-memory' (https://docs.vespa.ai/en/performance/container-tuning.html#jvm-heap-size).")
+                                            .formatted(clusterId, mp.percentage(), percentLimit, jvmModelCost / (1024D * 1024 * 1024)));
+                }
+                double gbLimit = 0.4;
+                double availableMemoryGb = mp.availableMemoryGb().getAsDouble();
+                if (availableMemoryGb < gbLimit) {
+                    throw new IllegalArgumentException(
+                            ("Allocated memory to JVM in cluster '%s' is too low (%.2fGB < %.2fGB). " +
+                                    "Estimated cost of ONNX models is %.2fGB. Either use a node flavor with more memory or use less expensive models. " +
+                                    "You may override this validation by specifying 'allocated-memory' (https://docs.vespa.ai/en/performance/container-tuning.html#jvm-heap-size).")
+                                            .formatted(clusterId, availableMemoryGb, gbLimit, jvmModelCost / (1024D * 1024 * 1024)));
+                }
+            }
+        });
+    }
+}
diff --git a/config-model/src/main/java/com/yahoo/vespa/model/application/validation/Validation.java b/config-model/src/main/java/com/yahoo/vespa/model/application/validation/Validation.java
index 53a553ee624..b9ecf7c2d22 100644
--- a/config-model/src/main/java/com/yahoo/vespa/model/application/validation/Validation.java
+++ b/config-model/src/main/java/com/yahoo/vespa/model/application/validation/Validation.java
@@ -87,6 +87,7 @@ public class Validation {
         new AccessControlFilterExcludeValidator().validate(model, deployState);
         new CloudUserFilterValidator().validate(model, deployState);
         new CloudHttpConnectorValidator().validate(model, deployState);
+        new JvmHeapSizeValidator().validate(model, deployState);
 
         additionalValidators.forEach(v -> v.validate(model, deployState));