Revert "Temporarily turn on device sync and multiple client instances"

This reverts commit 76fcd9b.

Author: neekolas

chaos/db.ts

@@ -2,9 +2,9 @@ import type { ChaosProvider } from "@chaos/provider";
 import type { WorkerManager } from "@workers/manager";
 
 export type DbChaosConfig = {
-  minLockTime: number;
-  maxLockTime: number;
-  lockInterval: number;
+  minLockTime: number; // Minimum duration in milliseconds to lock the database
+  maxLockTime: number; // Maximum duration in milliseconds to lock the database
+  lockInterval: number; // Interval in milliseconds between lock attempts
   impactedWorkerPercentage: number; // number between 0 and 100 for what % of workers to lock on each run
 };
 
@@ -14,6 +14,7 @@ export class DbChaos implements ChaosProvider {
   interval?: NodeJS.Timeout;
 
   constructor(config: DbChaosConfig) {
+    validateConfig(config);
     this.config = config;
   }
 
@@ -70,3 +71,26 @@ export class DbChaos implements ChaosProvider {
     await Promise.allSettled(Array.from(this.activeLocks.values()));
   }
 }
+
+function validateConfig(config: DbChaosConfig): void {
+  if (config.minLockTime > config.maxLockTime) {
+    throw new Error(
+      "Minimum lock time cannot be greater than maximum lock time",
+    );
+  }
+
+  if (
+    config.impactedWorkerPercentage < 0 ||
+    config.impactedWorkerPercentage > 100
+  ) {
+    throw new Error("Impacted worker percentage must be between 0 and 100");
+  }
+
+  if (!config.lockInterval) {
+    throw new Error("Lock interval must be defined");
+  }
+
+  if (config.impactedWorkerPercentage === undefined) {
+    throw new Error("Impacted worker percentage must be defined");
+  }
+}

chaos/network.ts

@@ -11,8 +11,6 @@ export type NetworkChaosConfig = {
   interval: number; // How often to apply chaos in ms
 };
 
-// import type { Worker } from "@workers/manager";
-
 export class NetworkChaos implements ChaosProvider {
   config: NetworkChaosConfig;
   interval?: NodeJS.Timeout;

chaos/provider.ts

@@ -1,5 +1,9 @@
 import { type WorkerManager } from "@workers/manager";
 
+// Generic interface for the Chaos Provider.
+// A Chaos Provider is started after the test has been setup, but before we start performing actions
+// It is stopped after the core of the test has been completed, and should remove all chaos so that final
+// validations can be performed cleanly.
 export interface ChaosProvider {
   start(workers: WorkerManager): Promise<void>;
   stop(): Promise<void>;

chaos/streams.ts

@@ -3,7 +3,7 @@ import { typeofStream, type WorkerClient } from "@workers/main";
 import type { WorkerManager } from "@workers/manager";
 
 export type StreamsConfig = {
-  cloned: boolean;
+  cloned: boolean; // Should the stream be run against the workers used in the tests, or a cloned client instance?
 };
 
 export class StreamsChaos implements ChaosProvider {

forks/cli.ts

@@ -14,6 +14,7 @@ import {
   type RuntimeConfig,
 } from "./config";
 
+// CLI options for fork testing
 interface ForkOptions {
   count: number; // Number of times to run the process (default: 100)
   cleanAll: boolean; // Clean all raw logs before starting
@@ -223,7 +224,7 @@ async function main() {
       type: "string",
       choices: ["local", "dev", "production"] as const,
       default:
-        (process.env.XMTP_ENV as "local" | "dev" | "production") || "dev",
+        (process.env.XMTP_ENV as "local" | "dev" | "production") || "local",
       describe: "XMTP environment",
     })
     .option("network-chaos-level", {

forks/config.ts

@@ -12,11 +12,6 @@ export const workerNames = [
   "random3",
   "random4",
   "random5",
-  "random1",
-  "random2",
-  "random3",
-  "random4",
-  "random5",
 ] as string[];
 
 // Operations configuration - enable/disable specific operations
@@ -28,10 +23,11 @@ export const epochRotationOperations = {
 export const otherOperations = {
   createInstallation: true, // creates a new installation for a random worker
   sendMessage: true, // sends a message to the group
-  sync: false, // syncs the group
+  sync: true, // syncs the group
 };
 export const randomInboxIdsCount = 50; // How many inboxIds to use randomly in the add/remove operations
 export const installationCount = 2; // How many installations to use randomly in the createInstallation operations
+export const maxConsecutiveFailures = 5; // Maximum number of times we throw an error verifying the results of a batch of operations
 export const testName = "forks";
 
 // Network chaos configuration

forks/forks.test.ts

@@ -13,6 +13,7 @@ import {
   epochRotationOperations,
   getConfigFromEnv,
   installationCount,
+  maxConsecutiveFailures,
   multinodeContainers,
   NODE_VERSION,
   otherOperations,
@@ -149,7 +150,8 @@ describe(testName, () => {
           let currentEpoch = 0n;
           let numConsecutiveFailures = 0;
 
-          while (currentEpoch < targetEpoch && numConsecutiveFailures < 5) {
+          // Run until we reach the target epoch or we hit 5 consecutive failures.
+          while (currentEpoch < targetEpoch) {
             const parallelOperationsArray = Array.from(
               { length: parallelOperations },
               () =>
@@ -198,6 +200,8 @@ describe(testName, () => {
                   }
                 })(),
             );
+
+            // We want to wait for all operations to complete, but ignore any errors which may be caused by the chaos
             const results = await Promise.allSettled(parallelOperationsArray);
             for (const result of results) {
               if (result.status === "rejected") {
@@ -221,6 +225,9 @@ describe(testName, () => {
             } catch (e) {
               console.error(`Group ${groupIndex + 1} operation failed:`, e);
               numConsecutiveFailures++;
+              if (numConsecutiveFailures >= maxConsecutiveFailures) {
+                throw e;
+              }
             }
           }
 
@@ -232,6 +239,7 @@ describe(testName, () => {
     } catch (e: any) {
       const msg = `Error during fork testing: ${e}`;
       console.error(msg);
+      // This will fail the test if there were too many failures
       expect.fail(msg);
     } finally {
       clearInterval(verifyInterval);

helpers/versions.ts

@@ -105,15 +105,6 @@ export const AgentVersionList = [
 
 // Node SDK version list
 export const VersionList = [
-  {
-    Client: Client440,
-    Conversation: Conversation440,
-    Dm: Dm440,
-    Group: Group440,
-    nodeSDK: "4.4.0",
-    nodeBindings: "1.6.1",
-    auto: true,
-  },
   {
     Client: Client430Dev,
     Conversation: Conversation430Dev,
@@ -123,6 +114,15 @@ export const VersionList = [
     nodeBindings: "1.7.0",
     auto: true,
   },
+  {
+    Client: Client440,
+    Conversation: Conversation440,
+    Dm: Dm440,
+    Group: Group440,
+    nodeSDK: "4.4.0",
+    nodeBindings: "1.6.1",
+    auto: true,
+  },
   {
     Client: Client430,
     Conversation: Conversation430,
@@ -275,8 +275,6 @@ export const regressionClient = async (
       loggingLevel,
       apiUrl,
       appVersion: APP_VERSION,
-      historySyncUrl: "http://localhost:5558",
-      disableDeviceSync: false,
       codecs: [new ReactionCodec(), new ReplyCodec()],
     });
   } catch (error) {

multinode/docker-compose.yml

@@ -113,16 +113,16 @@ services:
       - mlsdb
       - validation
 
-  validation:
-    image: ghcr.io/xmtp/mls-validation-service:main
-    platform: linux/amd64
-
   history-server:
     image: ghcr.io/xmtp/message-history-server:main
     platform: linux/amd64
     ports:
       - 5558:5558
 
+  validation:
+    image: ghcr.io/xmtp/mls-validation-service:main
+    platform: linux/amd64
+
   db:
     image: postgres:13
     environment:

scripts/revokeInstallations.ts

@@ -2,8 +2,7 @@ import { existsSync } from "node:fs";
 import { readFile } from "node:fs/promises";
 import { join } from "node:path";
 import { createSigner } from "@helpers/client";
-import { type XmtpEnv } from "@helpers/versions";
-import { Client } from "@xmtp/node-sdk";
+import { Client, type XmtpEnv } from "@helpers/versions";
 
 // Check Node.js version
 const nodeVersion = process.versions.node;