[OGUI-1389] Let gRPC do the reconnection on its own (#2155)

* removes the creation of a new client and shutting down of the existing one in case of AliECS hanging. * Instead it leaves gRPC to attempt the reconnection. * increases the timeout to allow more time to AliECS to reply in cases in which it gets frozen
AliceO2Group · Oct 6, 2023 · 111cd8a · 111cd8a
1 parent 689e974
commit 111cd8a
Show file tree

Hide file tree

Showing 3 changed files with 16 additions and 25 deletions.
diff --git a/Control/lib/control-core/ControlService.js b/Control/lib/control-core/ControlService.js
@@ -15,7 +15,6 @@
 const assert = require('assert');
 const path = require('path');
 const {WebSocketMessage, Log} = require('@aliceo2/web-ui');
-const GrpcProxy = require('./GrpcProxy.js');
 const log = new Log(`${process.env.npm_config_log_label ?? 'cog'}/controlservice`);
 const {errorHandler, errorLogger} = require('./../utils.js');
 const CoreUtils = require('./CoreUtils.js');
@@ -55,22 +54,16 @@ class ControlService {
    * @returns {Interval}
    */
   initiateHeartBeat() {
-    let wasInError = false;
     return setInterval(async () => {
       try {
-        await this.ctrlProx['GetEnvironments']({}, {deadline: Date.now() + 3500});
-        wasInError = false;
+        await this.ctrlProx['GetEnvironments']({}, {deadline: Date.now() + 9000});
       } catch (err) {
-        if (!wasInError) {
-          log.errorMessage('Unable to reach AliECS, attempting reconnection in silence', {
-            level: 20,
-            system: 'GUI',
-            facility: 'cog/controlservice'
-          });
-        }
-        wasInError = true;
-        this.ctrlProx.client.close()
-        this.ctrlProx = new GrpcProxy(this.coreConfig, this.O2_CONTROL_PROTO_PATH, wasInError);
+        const stateCode = this.ctrlProx.client.getChannel().getConnectivityState();
+        log.errorMessage(`Unable to reach AliECS (state: ${stateCode}), attempting reconnection`, {
+          level: 20,
+          system: 'GUI',
+          facility: 'cog/controlservice'
+        });
       }
     }, 10000);
   }

diff --git a/Control/lib/control-core/EnvCache.js b/Control/lib/control-core/EnvCache.js
@@ -27,7 +27,7 @@ class EnvCache {
   constructor(ctrlService) {
     this.ctrlService = ctrlService;
     this.cache = {};
-    this.timeout = 5000;
+    this.timeout = 9000;
     this.cacheEvictionTimeout = 5 * 60 * 1000;
     this.cacheEvictionLast = new Date();
     this.refreshInterval = setInterval(() => this.refresh(), this.timeout);

diff --git a/Control/lib/control-core/GrpcProxy.js b/Control/lib/control-core/GrpcProxy.js
@@ -30,9 +30,8 @@ class GrpcProxy {
    * https://grpc.io/grpc/node/grpc.Client.html
    * @param {Object} config - Contains configuration fields for gRPC client
    * @param {string} path - path to protofile location
-   * @param {boolean} wasInError - parameter to define if the connection is done following a failed attempt, thus logs are silent
    */
-  constructor(config, path, wasInError = false) {
+  constructor(config, path) {
     if (this._isConfigurationValid(config, path)) {
       const packageDefinition = protoLoader.loadSync(path, {longs: String, keepCase: false, arrays: true});
       const octlProto = grpcLibrary.loadPackageDefinition(packageDefinition);
@@ -42,8 +41,10 @@ class GrpcProxy {
       const options = {'grpc.max_receive_message_length': 1024 * 1024 * this._maxMessageLength}; // MB
 
       this.client = new protoService(address, credentials, options);
-      this.client.waitForReady(Date.now() + this._connectionTimeout,
-        (error) => this._logConnectionResponse(error, wasInError, address));
+      this.client.waitForReady(
+        Date.now() + this._connectionTimeout,
+        (error) => this._logConnectionResponse(error, address)
+      );
 
       // set all the available gRPC methods in object and build a separate array with names only
       this.methods = Object.keys(protoService.prototype)
@@ -135,15 +136,12 @@ class GrpcProxy {
   /**
    * 
    * @param {Error} error - error following attempt to connect to gRPC server
-   * @param {boolean} wasInError - flag to know if connection is attempted again after a failure; If yes, logging is disabled
    * @param {string} address - address on which connection was attempted
    */
-  _logConnectionResponse(error, wasInError, address) {
+  _logConnectionResponse(error, address) {
     if (error) {
-      if (!wasInError) {
-        log.error(`Connection to ${this._label} server (${address}) timeout`);
-        log.error(error.message);
-      }
+      log.error(`Connection to ${this._label} server (${address}) timeout`);
+      log.error(error.message);
 
       this.connectionError = error;
       this.isConnectionReady = false;