feat(cli): servers and workers in cluster mode (#13601)

* feat(cli): servers and workers in cluster mode

* allow percentage

.changeset/gorgeous-islands-double.md

@@ -0,0 +1,6 @@
+---
+"@medusajs/cli": patch
+"@medusajs/medusa": patch
+---
+
+feat(cli): servers and workers in cluster mode

packages/cli/medusa-cli/src/create-cli.ts

@@ -416,9 +416,19 @@ function buildLocalCommands(cli, isLocalProject) {
               : `Set port. Defaults to ${defaultPort}`,
           })
           .option(`cluster`, {
-            type: `number`,
+            type: `string`,
             describe:
-              "Start the Node.js server in cluster mode. You can specify the number of cpus to use, which defaults to (env.CPUS)",
+              "Start the Node.js server in cluster mode. Specify the number of CPUs to use or a percentage (e.g., 50%). Defaults to the number of available CPUs.",
+          })
+          .option("workers", {
+            type: "string",
+            default: "0",
+            describe: "Number of worker processes in cluster mode or a percentage of cluster size (e.g., 25%).",
+          })
+          .option("servers", {
+            type: "string",
+            default: "0",
+            describe: "Number of server processes in cluster mode or a percentage of cluster size (e.g., 25%).",
           }),
       handler: handlerP(
         getCommandHandler(`start`, (args, cmd) => {

packages/medusa/src/commands/start.ts

@@ -26,6 +26,32 @@ const EVERY_SIXTH_HOUR = "0 */6 * * *"
 const CRON_SCHEDULE = EVERY_SIXTH_HOUR
 const INSTRUMENTATION_FILE = "instrumentation"
 
+function parseValueOrPercentage(value: string, base: number): number {
+  if (typeof value !== "string") {
+    throw new Error(`Invalid value: ${value}. Must be a string.`)
+  }
+
+  const trimmed = value.trim()
+  if (trimmed.endsWith("%")) {
+    const percent = parseFloat(trimmed.slice(0, -1))
+    if (isNaN(percent)) {
+      throw new Error(`Invalid percentage: ${value}`)
+    }
+    if (percent < 0 || percent > 100) {
+      throw new Error(`Percentage must be between 0 and 100: ${value}`)
+    }
+    return Math.round((percent / 100) * base)
+  } else {
+    const num = parseInt(trimmed, 10)
+    if (isNaN(num) || num < 0) {
+      throw new Error(
+        `Invalid number: ${value}. Must be a non-negative integer.`
+      )
+    }
+    return num
+  }
+}
+
 /**
  * Imports the "instrumentation.js" file from the root of the
  * directory and invokes the register function. The existence
@@ -142,9 +168,30 @@ async function start(args: {
   host?: string
   port?: number
   types?: boolean
-  cluster?: number
+  cluster?: string
+  workers?: string
+  servers?: string
 }) {
-  const { port = 9000, host, directory, types } = args
+  const {
+    port = 9000,
+    host,
+    directory,
+    types,
+    cluster: clusterSize,
+    workers,
+    servers,
+  } = args
+
+  const maxCpus = os.cpus().length
+  const clusterSizeNum = clusterSize
+    ? parseValueOrPercentage(clusterSize, maxCpus)
+    : maxCpus
+  const serversCount = servers
+    ? parseValueOrPercentage(servers, clusterSizeNum)
+    : 0
+  const workersCount = workers
+    ? parseValueOrPercentage(workers, clusterSizeNum)
+    : 0
 
   async function internalStart(generateTypes: boolean) {
     track("CLI_START")
@@ -261,12 +308,17 @@ async function start(args: {
    * cluster mode
    */
   if ("cluster" in args) {
-    const maxCpus = os.cpus().length
-    const cpus = args.cluster ?? maxCpus
+    const cpus = clusterSizeNum
+    const numCPUs = Math.min(maxCpus, cpus)
+
+    if (serversCount + workersCount > numCPUs) {
+      throw new Error(
+        `Sum of servers (${serversCount}) and workers (${workersCount}) cannot exceed cluster size (${numCPUs})`
+      )
+    }
 
     if (cluster.isPrimary) {
       let isShuttingDown = false
-      const numCPUs = Math.min(maxCpus, cpus)
       const killMainProccess = () => process.exit(0)
       const gracefulShutDown = () => {
         isShuttingDown = true
@@ -274,8 +326,14 @@ async function start(args: {
 
       for (let index = 0; index < numCPUs; index++) {
         const worker = cluster.fork()
+        let workerMode: "server" | "worker" | "shared" = "shared"
+        if (index < serversCount) {
+          workerMode = "server"
+        } else if (index < serversCount + workersCount) {
+          workerMode = "worker"
+        }
         worker.on("online", () => {
-          worker.send({ index })
+          worker.send({ index, workerMode })
         })
       }
 
@@ -291,6 +349,10 @@ async function start(args: {
       process.on("SIGINT", gracefulShutDown)
     } else {
       process.on("message", async (msg: any) => {
+        if (msg.workerMode) {
+          process.env.MEDUSA_WORKER_MODE = msg.workerMode
+        }
+
         if (msg.index > 0) {
           process.env.PLUGIN_ADMIN_UI_SKIP_CACHE = "true"
         }