@platformatic/runtime 2.44.4 → 2.45.0

package/config.d.ts

@@ -5,7 +5,7 @@
  * and run json-schema-to-typescript to regenerate this file.
  */
 
-export type HttpsSchemasPlatformaticDevPlatformaticRuntime2444Json = {
+export type HttpsSchemasPlatformaticDevPlatformaticRuntime2450Json = {
   [k: string]: unknown;
 } & {
   $schema?: string;
@@ -163,6 +163,7 @@ export type HttpsSchemasPlatformaticDevPlatformaticRuntime2444Json = {
     | boolean
     | {
         port?: number | string;
+        enabled?: boolean | string;
         hostname?: string;
         endpoint?: string;
         auth?: {

package/lib/config.js

@@ -70,7 +70,7 @@ async function _transformConfig (configManager, args) {
         config = join(entryPath, configFilename)
       }
 
-      const service = { id, config, path: entryPath, useHttp: !!mapping.useHttp }
+      const service = { id, config, path: entryPath, useHttp: !!mapping.useHttp, health: mapping.health }
       const existingServiceId = services.findIndex(service => service.id === id)
 
       if (existingServiceId !== -1) {

package/lib/management-api.js

@@ -36,7 +36,7 @@ async function managementApiPlugin (app, opts) {
 
   app.post('/stop', async () => {
     app.log.debug('stop services')
-    await runtime.close(true)
+    await runtime.close()
   })
 
   app.post('/restart', async () => {

package/lib/prom-server.js

@@ -3,6 +3,9 @@
 const fastify = require('fastify')
 
 async function startPrometheusServer (runtime, opts) {
+  if (opts.enabled === false) {
+    return
+  }
   const host = opts.hostname ?? '0.0.0.0'
   const port = opts.port ?? 9090
   const metricsEndpoint = opts.endpoint ?? '/metrics'

package/lib/runtime.js

@@ -26,6 +26,7 @@ const { sendViaITC, waitEventFromITC } = require('./worker/itc')
 const { RoundRobinMap } = require('./worker/round-robin-map.js')
 const {
   kId,
+  kFullId,
   kServiceId,
   kWorkerId,
   kITC,
@@ -273,7 +274,7 @@ class Runtime extends EventEmitter {
     if (this.#inspectorServer) {
       await this.#inspectorServer.close()
     }
-
+    await this.#meshInterceptor.close()
     for (const service of this.#servicesIds) {
       await this.stopService(service, silent)
     }
@@ -285,6 +286,7 @@ class Runtime extends EventEmitter {
     this.emit('restarting')
 
     await this.stop()
+    this.#meshInterceptor.restart()
     await this.start()
 
     this.emit('restarted')
@@ -296,7 +298,7 @@ class Runtime extends EventEmitter {
     return this.#status
   }
 
-  async close (fromManagementApi = false, silent = false) {
+  async close (silent = false) {
     this.#updateStatus('closing')
 
     clearInterval(this.#metricsTimeout)
@@ -304,14 +306,10 @@ class Runtime extends EventEmitter {
     await this.stop(silent)
 
     if (this.#managementApi) {
-      if (fromManagementApi) {
-        // This allow a close request coming from the management API to correctly be handled
-        setImmediate(() => {
-          this.#managementApi.close()
-        })
-      } else {
-        await this.#managementApi.close()
-      }
+      // This allow a close request coming from the management API to correctly be handled
+      setImmediate(() => {
+        this.#managementApi.close()
+      })
     }
 
     if (this.#prometheusServer) {
@@ -1006,7 +1004,7 @@ class Runtime extends EventEmitter {
     this.emit('service:init', id)
   }
 
-  async #setupWorker (config, serviceConfig, workersCount, serviceId, index) {
+  async #setupWorker (config, serviceConfig, workersCount, serviceId, index, enabled = true) {
     const { restartOnError } = config
     const workerId = `${serviceId}:${index}`
 
@@ -1096,7 +1094,7 @@ class Runtime extends EventEmitter {
       worker[kWorkerStatus] = 'exited'
       this.emit('service:worker:exited', eventPayload)
 
-      this.#cleanupWorker(workerId, worker)
+      this.#cleanupWorker(worker)
 
       if (this.#status === 'stopping') {
         return
@@ -1130,8 +1128,10 @@ class Runtime extends EventEmitter {
     })
 
     worker[kId] = workersCount > 1 ? workerId : serviceId
+    worker[kFullId] = workerId
     worker[kServiceId] = serviceId
     worker[kWorkerId] = workersCount > 1 ? index : undefined
+    worker[kWorkerStatus] = 'init'
     worker[kForwardEvents] = false
 
     if (inspectorOptions) {
@@ -1190,19 +1190,21 @@ class Runtime extends EventEmitter {
       })
     }
 
-    // Store locally
-    this.#workers.set(workerId, worker)
+    if (enabled) {
+      // Store locally
+      this.#workers.set(workerId, worker)
 
-    if (serviceConfig.entrypoint) {
-      this.#entrypointId = serviceId
+      // Setup the interceptor
+      this.#meshInterceptor.route(serviceId, worker)
     }
 
-    // Setup the interceptor
-    this.#meshInterceptor.route(serviceId, worker)
-
     // Store dependencies
     const [{ dependencies }] = await waitEventFromITC(worker, 'init')
 
+    if (serviceConfig.entrypoint) {
+      this.#entrypointId = serviceId
+    }
+
     serviceConfig.dependencies = dependencies
     for (const { envVar, url } of dependencies) {
       if (envVar) {
@@ -1214,9 +1216,11 @@ class Runtime extends EventEmitter {
     worker[kConfig] = { ...serviceConfig, health }
     worker[kWorkerStatus] = 'boot'
     this.emit('service:worker:boot', eventPayload)
+
+    return worker
   }
 
-  #setupHealthCheck (worker, errorLabel) {
+  #setupHealthCheck (config, serviceConfig, workersCount, id, index, workerId, worker, errorLabel) {
     // Clear the timeout when exiting
     worker.on('exit', () => clearTimeout(worker[kHealthCheckTimer]))
 
@@ -1229,9 +1233,10 @@ class Runtime extends EventEmitter {
       const memoryUsage = heapUsed / maxHeapTotal
       const unhealthy = elu > maxELU || memoryUsage > maxHeapUsed
 
+      const serviceId = worker[kServiceId]
       this.emit('health', {
         id: worker[kId],
-        service: worker[kServiceId],
+        service: serviceId,
         worker: worker[kWorkerId],
         currentHealth: health,
         unhealthy,
@@ -1256,20 +1261,39 @@ class Runtime extends EventEmitter {
         unhealthyChecks = 0
       }
 
-      if (unhealthyChecks >= maxUnhealthyChecks) {
-        this.logger.error(
-          { elu, maxELU, memoryUsage, maxMemoryUsage: maxHeapUsed },
-          `The ${errorLabel} is unhealthy. Forcefully terminating it ...`
-        )
-        worker.terminate()
-        return
-      }
+      if (unhealthyChecks === maxUnhealthyChecks) {
+        try {
+          this.logger.error(
+            { elu, maxELU, memoryUsage, maxMemoryUsage: maxHeapUsed },
+            `The ${errorLabel} is unhealthy. Replacing it ...`
+          )
 
-      worker[kHealthCheckTimer].refresh()
+          await this.#replaceWorker(config, serviceConfig, workersCount, id, index, workerId, worker)
+        } catch (e) {
+          this.logger.error(
+            { elu, maxELU, memoryUsage, maxMemoryUsage: maxHeapUsed },
+            `Cannot replace the ${errorLabel}. Forcefully terminating it ...`
+          )
+
+          worker.terminate()
+        }
+      } else {
+        worker[kHealthCheckTimer].refresh()
+      }
     }, interval)
   }
 
-  async #startWorker (config, serviceConfig, workersCount, id, index, silent, bootstrapAttempt = 0) {
+  async #startWorker (
+    config,
+    serviceConfig,
+    workersCount,
+    id,
+    index,
+    silent,
+    bootstrapAttempt = 0,
+    worker = undefined,
+    disableRestartAttempts = false
+  ) {
     const workerId = `${id}:${index}`
     const label = this.#workerExtendedLabel(id, index, workersCount)
 
@@ -1277,7 +1301,10 @@ class Runtime extends EventEmitter {
       this.logger?.info(`Starting the ${label}...`)
     }
 
-    let worker = await this.#getWorkerById(id, index, false, false)
+    if (!worker) {
+      worker = await this.#getWorkerById(id, index, false, false)
+    }
+
     const eventPayload = { service: id, worker: index, workersCount }
 
     // The service was stopped, recreate the thread
@@ -1321,7 +1348,7 @@ class Runtime extends EventEmitter {
       if (enabled && config.restartOnError > 0) {
         worker[kHealthCheckTimer] = setTimeout(
           () => {
-            this.#setupHealthCheck(worker, label)
+            this.#setupHealthCheck(config, serviceConfig, workersCount, id, index, workerId, worker, label)
           },
           gracePeriod > 0 ? gracePeriod : 1
         )
@@ -1330,7 +1357,7 @@ class Runtime extends EventEmitter {
       // TODO: handle port allocation error here
       if (error.code === 'EADDRINUSE') throw error
 
-      this.#cleanupWorker(workerId, worker)
+      this.#cleanupWorker(worker)
 
       if (worker[kWorkerStatus] !== 'exited') {
         // This prevent the exit handler to restart service
@@ -1346,7 +1373,7 @@ class Runtime extends EventEmitter {
 
       const restartOnError = config.restartOnError
 
-      if (!restartOnError) {
+      if (disableRestartAttempts || !restartOnError) {
         throw error
       }
 
@@ -1369,13 +1396,20 @@ class Runtime extends EventEmitter {
     }
   }
 
-  async #stopWorker (workersCount, id, index, silent) {
-    const worker = await this.#getWorkerById(id, index, false, false)
+  async #stopWorker (workersCount, id, index, silent, worker = undefined) {
+    if (!worker) {
+      worker = await this.#getWorkerById(id, index, false, false)
+    }
 
     if (!worker) {
       return
     }
 
+    // Starting should be aborted, discard the worker
+    if (worker[kWorkerStatus] !== 'started') {
+      return this.#discardWorker(worker)
+    }
+
     const eventPayload = { service: id, worker: index, workersCount }
 
     worker[kWorkerStatus] = 'stopping'
@@ -1419,12 +1453,23 @@ class Runtime extends EventEmitter {
     this.emit('service:worker:stopped', eventPayload)
   }
 
-  #cleanupWorker (workerId, worker) {
-    this.#workers.delete(workerId)
+  #cleanupWorker (worker) {
+    clearTimeout(worker[kHealthCheckTimer])
+
+    const currentWorker = this.#workers.get(worker[kFullId])
+
+    if (currentWorker === worker) {
+      this.#workers.delete(worker[kFullId])
+    }
 
     worker[kITC].close()
+  }
 
-    clearTimeout(worker[kHealthCheckTimer])
+  async #discardWorker (worker) {
+    worker.removeAllListeners('exit')
+    await worker.terminate()
+
+    return this.#cleanupWorker(worker)
   }
 
   #workerExtendedLabel (serviceId, workerId, workersCount) {
@@ -1476,6 +1521,39 @@ class Runtime extends EventEmitter {
     await restartPromise
   }
 
+  async #replaceWorker (config, serviceConfig, workersCount, serviceId, index, workerId, worker) {
+    let newWorker
+
+    try {
+      // Create a new worker
+      newWorker = await this.#setupWorker(config, serviceConfig, workersCount, serviceId, index, false)
+
+      // Make sure the runtime hasn't been stopped in the meanwhile
+      if (this.#status !== 'started') {
+        return this.#discardWorker(newWorker)
+      }
+
+      // Add the worker to the mesh
+      await this.#startWorker(config, serviceConfig, workersCount, serviceId, index, false, 0, newWorker, true)
+
+      // Make sure the runtime hasn't been stopped in the meanwhile
+      if (this.#status !== 'started') {
+        return this.#discardWorker(newWorker)
+      }
+
+      this.#workers.set(workerId, newWorker)
+      this.#meshInterceptor.route(serviceId, newWorker)
+
+      // Remove the old worker and then kill it
+      await sendViaITC(worker, 'removeFromMesh')
+    } catch (e) {
+      newWorker?.terminate?.()
+      throw e
+    }
+
+    await this.#stopWorker(workersCount, serviceId, index, false, worker)
+  }
+
   async #getServiceById (serviceId, ensureStarted = false, mustExist = true) {
     // If the serviceId includes the worker, properly split
     let workerId

package/lib/schema.js

@@ -3,7 +3,7 @@
 
 const telemetry = require('@platformatic/telemetry').schema
 const {
-  schemaComponents: { server, logger, health }
+  schemaComponents: { server, logger, health, healthWithoutDefaults }
 } = require('@platformatic/utils')
 
 const env = {
@@ -65,7 +65,7 @@ const services = {
         type: 'boolean'
       },
       workers,
-      health: { ...health, default: undefined },
+      health: { ...healthWithoutDefaults, default: undefined },
       arguments: {
         type: 'array',
         items: {
@@ -174,7 +174,7 @@ const platformaticRuntimeSchema = {
                 type: 'boolean'
               },
               workers,
-              health: { ...health, default: undefined },
+              health: { ...healthWithoutDefaults, default: undefined },
               preload,
               arguments: {
                 type: 'array',
@@ -359,6 +359,13 @@ const platformaticRuntimeSchema = {
             port: {
               anyOf: [{ type: 'integer' }, { type: 'string' }]
             },
+            enabled: {
+              anyOf: [{
+                type: 'boolean'
+              }, {
+                type: 'string'
+              }]
+            },
             hostname: { type: 'string' },
             endpoint: { type: 'string' },
             auth: {

package/lib/worker/itc.js

@@ -93,7 +93,7 @@ function setupITC (app, service, dispatcher) {
           await app.stop()
         }
 
-        dispatcher.interceptor.close()
+        await dispatcher.interceptor.close()
         itc.close()
       },
 
@@ -101,6 +101,14 @@ function setupITC (app, service, dispatcher) {
         return app.stackable.build()
       },
 
+      async removeFromMesh () {
+        return dispatcher.interceptor.close()
+      },
+
+      inject (injectParams) {
+        return app.stackable.inject(injectParams)
+      },
+
       getStatus () {
         return app.getStatus()
       },
@@ -158,10 +166,6 @@ function setupITC (app, service, dispatcher) {
         } catch (err) {
           throw new errors.FailedToRetrieveHealthError(service.id, err.message)
         }
-      },
-
-      inject (injectParams) {
-        return app.stackable.inject(injectParams)
       }
     }
   })

package/lib/worker/symbols.js

@@ -2,6 +2,7 @@
 
 const kConfig = Symbol.for('plt.runtime.config')
 const kId = Symbol.for('plt.runtime.id') // This is also used to detect if we are running in a Platformatic runtime thread
+const kFullId = Symbol.for('plt.runtime.fullId')
 const kServiceId = Symbol.for('plt.runtime.service.id')
 const kWorkerId = Symbol.for('plt.runtime.worker.id')
 const kITC = Symbol.for('plt.runtime.itc')
@@ -14,6 +15,7 @@ const kStderrMarker = '\ue002'
 module.exports = {
   kConfig,
   kId,
+  kFullId,
   kServiceId,
   kWorkerId,
   kITC,

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@platformatic/runtime",
-  "version": "2.44.4",
+  "version": "2.45.0",
   "description": "",
   "main": "index.js",
   "bin": {
@@ -20,6 +20,7 @@
     "@fastify/compress": "^8.0.0",
     "@fastify/express": "^4.0.0",
     "@fastify/formbody": "^8.0.0",
+    "autocannon": "^8.0.0",
     "borp": "^0.19.0",
     "c8": "^10.0.0",
     "eslint": "9",
@@ -36,12 +37,12 @@
     "typescript": "^5.5.4",
     "undici-oidc-interceptor": "^0.5.0",
     "why-is-node-running": "^2.2.2",
-    "@platformatic/composer": "2.44.4",
-    "@platformatic/db": "2.44.4",
-    "@platformatic/node": "2.44.4",
-    "@platformatic/service": "2.44.4",
-    "@platformatic/sql-graphql": "2.44.4",
-    "@platformatic/sql-mapper": "2.44.4"
+    "@platformatic/composer": "2.45.0",
+    "@platformatic/db": "2.45.0",
+    "@platformatic/service": "2.45.0",
+    "@platformatic/node": "2.45.0",
+    "@platformatic/sql-graphql": "2.45.0",
+    "@platformatic/sql-mapper": "2.45.0"
   },
   "dependencies": {
     "@fastify/accepts": "^5.0.0",
@@ -73,15 +74,15 @@
     "tail-file-stream": "^0.2.0",
     "thread-cpu-usage": "^0.2.0",
     "undici": "^7.0.0",
-    "undici-thread-interceptor": "^0.11.0",
+    "undici-thread-interceptor": "^0.12.2",
     "ws": "^8.16.0",
-    "@platformatic/basic": "2.44.4",
-    "@platformatic/config": "2.44.4",
-    "@platformatic/generators": "2.44.4",
-    "@platformatic/itc": "2.44.4",
-    "@platformatic/telemetry": "2.44.4",
-    "@platformatic/ts-compiler": "2.44.4",
-    "@platformatic/utils": "2.44.4"
+    "@platformatic/basic": "2.45.0",
+    "@platformatic/generators": "2.45.0",
+    "@platformatic/config": "2.45.0",
+    "@platformatic/itc": "2.45.0",
+    "@platformatic/ts-compiler": "2.45.0",
+    "@platformatic/telemetry": "2.45.0",
+    "@platformatic/utils": "2.45.0"
   },
   "scripts": {
     "test": "npm run lint && borp --concurrency=1 --timeout=300000 && tsd",

package/schema.json

@@ -1,5 +1,5 @@
 {
-  "$id": "https://schemas.platformatic.dev/@platformatic/runtime/2.44.4.json",
+  "$id": "https://schemas.platformatic.dev/@platformatic/runtime/2.45.0.json",
   "$schema": "http://json-schema.org/draft-07/schema#",
   "type": "object",
   "properties": {
@@ -78,7 +78,6 @@
                 "type": "object",
                 "properties": {
                   "enabled": {
-                    "default": true,
                     "anyOf": [
                       {
                         "type": "boolean"
@@ -89,7 +88,6 @@
                     ]
                   },
                   "interval": {
-                    "default": 30000,
                     "anyOf": [
                       {
                         "type": "number",
@@ -101,7 +99,6 @@
                     ]
                   },
                   "gracePeriod": {
-                    "default": 30000,
                     "anyOf": [
                       {
                         "type": "number",
@@ -113,7 +110,6 @@
                     ]
                   },
                   "maxUnhealthyChecks": {
-                    "default": 10,
                     "anyOf": [
                       {
                         "type": "number",
@@ -125,7 +121,6 @@
                     ]
                   },
                   "maxELU": {
-                    "default": 0.99,
                     "anyOf": [
                       {
                         "type": "number",
@@ -138,7 +133,6 @@
                     ]
                   },
                   "maxHeapUsed": {
-                    "default": 0.99,
                     "anyOf": [
                       {
                         "type": "number",
@@ -151,7 +145,6 @@
                     ]
                   },
                   "maxHeapTotal": {
-                    "default": 4294967296,
                     "anyOf": [
                       {
                         "type": "number",
@@ -249,7 +242,6 @@
             "type": "object",
             "properties": {
               "enabled": {
-                "default": true,
                 "anyOf": [
                   {
                     "type": "boolean"
@@ -260,7 +252,6 @@
                 ]
               },
               "interval": {
-                "default": 30000,
                 "anyOf": [
                   {
                     "type": "number",
@@ -272,7 +263,6 @@
                 ]
               },
               "gracePeriod": {
-                "default": 30000,
                 "anyOf": [
                   {
                     "type": "number",
@@ -284,7 +274,6 @@
                 ]
               },
               "maxUnhealthyChecks": {
-                "default": 10,
                 "anyOf": [
                   {
                     "type": "number",
@@ -296,7 +285,6 @@
                 ]
               },
               "maxELU": {
-                "default": 0.99,
                 "anyOf": [
                   {
                     "type": "number",
@@ -309,7 +297,6 @@
                 ]
               },
               "maxHeapUsed": {
-                "default": 0.99,
                 "anyOf": [
                   {
                     "type": "number",
@@ -322,7 +309,6 @@
                 ]
               },
               "maxHeapTotal": {
-                "default": 4294967296,
                 "anyOf": [
                   {
                     "type": "number",
@@ -485,7 +471,6 @@
             "type": "object",
             "properties": {
               "enabled": {
-                "default": true,
                 "anyOf": [
                   {
                     "type": "boolean"
@@ -496,7 +481,6 @@
                 ]
               },
               "interval": {
-                "default": 30000,
                 "anyOf": [
                   {
                     "type": "number",
@@ -508,7 +492,6 @@
                 ]
               },
               "gracePeriod": {
-                "default": 30000,
                 "anyOf": [
                   {
                     "type": "number",
@@ -520,7 +503,6 @@
                 ]
               },
               "maxUnhealthyChecks": {
-                "default": 10,
                 "anyOf": [
                   {
                     "type": "number",
@@ -532,7 +514,6 @@
                 ]
               },
               "maxELU": {
-                "default": 0.99,
                 "anyOf": [
                   {
                     "type": "number",
@@ -545,7 +526,6 @@
                 ]
               },
               "maxHeapUsed": {
-                "default": 0.99,
                 "anyOf": [
                   {
                     "type": "number",
@@ -558,7 +538,6 @@
                 ]
               },
               "maxHeapTotal": {
-                "default": 4294967296,
                 "anyOf": [
                   {
                     "type": "number",
@@ -1137,6 +1116,16 @@
                 }
               ]
             },
+            "enabled": {
+              "anyOf": [
+                {
+                  "type": "boolean"
+                },
+                {
+                  "type": "string"
+                }
+              ]
+            },
             "hostname": {
               "type": "string"
             },