Refactor service resource limits (#4363)

porter-dev · Mar 4, 2024 · 5f0f86a · 5f0f86a
1 parent d8ed126
commit 5f0f86a
Show file tree

Hide file tree

Showing 25 changed files with 808 additions and 701 deletions.
diff --git a/dashboard/src/lib/clusters/constants.ts b/dashboard/src/lib/clusters/constants.ts
diff --git a/dashboard/src/lib/clusters/types.ts b/dashboard/src/lib/clusters/types.ts
@@ -237,6 +237,8 @@ export type ClientMachineType = {
   displayName: string;
   supportedRegions: Array<AWSRegion | GCPRegion | AzureRegion>;
   isGPU: boolean;
+  cpuCores: number;
+  ramMegabytes: number;
 };
 type PreflightCheckResolutionStep = {
   text: string;
@@ -261,7 +263,7 @@ export const nodeValidator = z.object({
 });
 export type ClientNode = {
   nodeGroupType: NodeGroupType;
-  instanceType: string;
+  instanceType: ClientMachineType;
 };
 
 // Cluster
@@ -272,6 +274,7 @@ export const clusterValidator = z.object({
   cloud_provider: cloudProviderValidator,
   cloud_provider_credential_identifier: z.string(),
   status: z.string(),
+  ingress_ip: z.string().optional().default(""),
 });
 export type SerializedCluster = z.infer<typeof clusterValidator>;
 export type ClientCluster = Omit<SerializedCluster, "cloud_provider"> & {

diff --git a/dashboard/src/lib/hooks/useCluster.ts b/dashboard/src/lib/hooks/useCluster.ts
@@ -497,7 +497,7 @@ export const useClusterNodeList = ({
       );
 
       const parsed = await z.array(nodeValidator).parseAsync(res.data);
-      return parsed
+      const nodes = parsed
         .map((n) => {
           const nodeGroupType = match(n.labels["porter.run/workload-kind"])
             .with("application", () => "APPLICATION" as const)
@@ -508,7 +508,21 @@ export const useClusterNodeList = ({
           if (nodeGroupType === "UNKNOWN") {
             return undefined;
           }
-          const instanceType = n.labels["node.kubernetes.io/instance-type"];
+          const instanceTypeName = n.labels["node.kubernetes.io/instance-type"];
+          if (!instanceTypeName) {
+            return undefined;
+          }
+          // TODO: use more node information to narrow down which cloud provider instance type list to check against
+          const instanceType =
+            CloudProviderAWS.machineTypes.find(
+              (i) => i.name === instanceTypeName
+            ) ??
+            CloudProviderAzure.machineTypes.find(
+              (i) => i.name === instanceTypeName
+            ) ??
+            CloudProviderGCP.machineTypes.find(
+              (i) => i.name === instanceTypeName
+            );
           if (!instanceType) {
             return undefined;
           }
@@ -518,6 +532,7 @@ export const useClusterNodeList = ({
           };
         })
         .filter(valueExists);
+      return nodes;
     },
     {
       refetchInterval,

diff --git a/dashboard/src/lib/hooks/usePorterYaml.ts b/dashboard/src/lib/hooks/usePorterYaml.ts
@@ -1,13 +1,16 @@
-import { useCallback, useContext, useEffect, useState } from "react";
+import { useCallback, useContext, useEffect, useMemo, useState } from "react";
 import { PorterApp } from "@porter-dev/api-contracts";
 import { useQuery } from "@tanstack/react-query";
 import { z } from "zod";
 
+import { useClusterContext } from "main/home/infrastructure-dashboard/ClusterContextProvider";
 import { serviceOverrides, type SourceOptions } from "lib/porter-apps";
-import { type DetectedServices } from "lib/porter-apps/services";
+import {
+  getServiceResourceAllowances,
+  type DetectedServices,
+} from "lib/porter-apps/services";
 
 import api from "shared/api";
-import { useClusterResources } from "shared/ClusterResourcesContext";
 import { Context } from "shared/Context";
 
 type PorterYamlStatus =
@@ -41,11 +44,18 @@ export const usePorterYaml = ({
   useDefaults?: boolean;
 }): PorterYamlStatus => {
   const { currentProject, currentCluster } = useContext(Context);
-  const { currentClusterResources } = useClusterResources();
   const [detectedServices, setDetectedServices] =
     useState<DetectedServices | null>(null);
   const [detectedName, setDetectedName] = useState<string | null>(null);
   const [porterYamlFound, setPorterYamlFound] = useState(false);
+  const { nodes } = useClusterContext();
+  const { newServiceDefaultCpuCores, newServiceDefaultRamMegabytes } =
+    useMemo(() => {
+      return getServiceResourceAllowances(
+        nodes,
+        currentProject?.sandbox_enabled
+      );
+    }, [nodes]);
 
   const { data, status } = useQuery(
     [
@@ -135,8 +145,8 @@ export const usePorterYaml = ({
         const { services, predeploy, build } = serviceOverrides({
           overrides: proto,
           useDefaults,
-          defaultCPU: currentClusterResources.defaultCPU,
-          defaultRAM: currentClusterResources.defaultRAM,
+          defaultCPU: newServiceDefaultCpuCores,
+          defaultRAM: newServiceDefaultRamMegabytes,
         });
 
         if (services.length || predeploy || build) {

diff --git a/dashboard/src/lib/porter-apps/services.ts b/dashboard/src/lib/porter-apps/services.ts
@@ -7,6 +7,8 @@ import _ from "lodash";
 import { match } from "ts-pattern";
 import { z } from "zod";
 
+import { type ClientNode } from "lib/clusters/types";
+
 import { type BuildOptions } from "./build";
 import {
   autoscalingValidator,
@@ -719,3 +721,70 @@ export function serializedServiceFromProto({
     )
     .exhaustive();
 }
+
+const SMALL_INSTANCE_UPPER_BOUND = 0.75;
+const LARGE_INSTANCE_UPPER_BOUND = 0.9;
+const NEW_SERVICE_RESOURCE_DEFAULT_MULTIPLIER = 0.125;
+
+const DEFAULT_RESOURCE_ALLOWANCES = {
+  maxCpuCores: 1.5,
+  newServiceDefaultCpuCores: 0.19,
+  maxRamMegabytes: 3100,
+  newServiceDefaultRamMegabytes: 400,
+};
+
+const DEFAULT_SANDBOX_RESOURCE_ALLOWANCES = {
+  maxCpuCores: 0.2,
+  newServiceDefaultCpuCores: 0.1,
+  maxRamMegabytes: 250,
+  newServiceDefaultRamMegabytes: 120,
+};
+
+export function getServiceResourceAllowances(
+  nodes: ClientNode[],
+  isSandboxEnabled?: boolean
+): {
+  maxCpuCores: number;
+  maxRamMegabytes: number;
+  newServiceDefaultCpuCores: number;
+  newServiceDefaultRamMegabytes: number;
+} {
+  if (isSandboxEnabled) {
+    return DEFAULT_SANDBOX_RESOURCE_ALLOWANCES;
+  }
+
+  if (nodes.length === 0) {
+    return DEFAULT_RESOURCE_ALLOWANCES;
+  }
+  const maxRamApplicationInstance = nodes
+    .filter((n) => n.nodeGroupType === "APPLICATION")
+    .reduce((max, node) =>
+      node.instanceType.ramMegabytes > max.instanceType.ramMegabytes
+        ? node
+        : max
+    );
+  const multiplier =
+    maxRamApplicationInstance.instanceType.ramMegabytes > 16000
+      ? LARGE_INSTANCE_UPPER_BOUND
+      : SMALL_INSTANCE_UPPER_BOUND;
+
+  const maxCpuCores =
+    Math.floor(
+      maxRamApplicationInstance.instanceType.cpuCores * multiplier * 2
+    ) / 2; // round to nearest half
+  const maxRamMegabytes =
+    Math.round(
+      (maxRamApplicationInstance.instanceType.ramMegabytes * multiplier) / 100
+    ) * 100; // round to nearest 100 MB
+  return {
+    maxCpuCores,
+    newServiceDefaultCpuCores: Number(
+      (maxCpuCores * NEW_SERVICE_RESOURCE_DEFAULT_MULTIPLIER).toFixed(2)
+    ), // round to hundredths place
+    maxRamMegabytes,
+    newServiceDefaultRamMegabytes:
+      Math.round(
+        (maxRamMegabytes * NEW_SERVICE_RESOURCE_DEFAULT_MULTIPLIER) / 100
+      ) * 100, // round to nearest 100 MB
+  };
+}