mongo/jstests/concurrency/fsm_libs/worker_thread.js

311 lines
15 KiB
JavaScript

import {Cluster} from "jstests/concurrency/fsm_libs/cluster.js";
import {parseConfig} from "jstests/concurrency/fsm_libs/parse_config.js";
import {SpecificSecondaryReaderMongo} from "jstests/libs/specific_secondary_reader_mongo.js";
const AsyncFunction = Object.getPrototypeOf(async function () {}).constructor;
export const workerThread = (function () {
// workloads = list of workload filenames
// args.tid = the thread identifier
// args.tenantId = the tenant id
// args.data = map of workload -> 'this' parameter passed to the FSM state functions
// args.host = the address to make a new connection to
// args.latch = CountDownLatch instance for starting all threads
// args.dbName = the database name
// args.collName = the collection name
// args.cluster = connection strings for all cluster nodes (see cluster.js for format)
// args.clusterOptions = the configuration of the cluster
// args.seed = seed for the random number generator
// args.errorLatch = CountDownLatch instance that threads count down when they error
// args.sessionOptions = the options to start a session with
// run = callback that takes a map of workloads to their associated $config
async function main(workloads, args, run) {
let myDB;
let configs = {};
let connectionString = "mongodb://" + args.host + "/?appName=tid:" + args.tid;
if (typeof args.replSetName !== "undefined") {
connectionString += "&replicaSet=" + args.replSetName;
}
// The global 'TestData' object may still be undefined if the concurrency suite isn't being
// run by resmoke.py (e.g. if it is being run via a parallel shell in the backup/restore
// tests).
TestData = TestData !== undefined ? TestData : {};
try {
// Running a callback passed through testData before running fsm worker threads.
// Can be added to yml files as the following example:
// fsmPreOverridesLoadedCallback: '
// testingReplication = true;
// await import("jstests/libs/override_methods/network_error_and_txn_override.js");
// ...
// '
if (TestData.fsmPreOverridesLoadedCallback) {
if (typeof TestData.fsmPreOverridesLoadedCallback !== "string") {
throw new Error(
`TestData.fsmPreOverridesLoadedCallback must be a string, not ${typeof TestData.fsmPreOverridesLoadedCallback}`,
);
}
const fn = new AsyncFunction(TestData.fsmPreOverridesLoadedCallback);
await fn();
}
if (typeof db !== "undefined") {
// The implicit database connection created within the thread's scope
// is unneeded, so forcibly clean it up.
globalThis.db = undefined;
gc();
}
let mongo;
if (TestData.pinningSecondary) {
mongo = new SpecificSecondaryReaderMongo(connectionString, args.secondaryHost);
} else {
mongo = new Mongo(connectionString);
}
if (typeof args.tenantId !== "undefined") {
TestData.tenantId = args.tenantId;
await import("jstests/libs/override_methods/simulate_atlas_proxy.js");
}
// Retry operations that fail due to in-progress background operations. Load this early
// so that later overrides can be retried.
await import("jstests/libs/override_methods/index_builds/implicitly_retry_on_background_op_in_progress.js");
if (typeof args.sessionOptions !== "undefined") {
let initialClusterTime;
let initialOperationTime;
// JavaScript objects backed by C++ objects (e.g. BSON values from a command
// response) do not serialize correctly when passed through the Thread
// constructor. To work around this behavior, we instead pass a stringified form
// of the JavaScript object through the Thread constructor and use eval()
// to rehydrate it.
if (typeof args.sessionOptions.initialClusterTime === "string") {
initialClusterTime = eval("(" + args.sessionOptions.initialClusterTime + ")");
// The initialClusterTime property was removed from SessionOptions in a
// later revision of the Driver's specification, so we remove the property
// and call advanceClusterTime() ourselves.
delete args.sessionOptions.initialClusterTime;
}
if (typeof args.sessionOptions.initialOperationTime === "string") {
initialOperationTime = eval("(" + args.sessionOptions.initialOperationTime + ")");
// The initialOperationTime property was removed from SessionOptions in a
// later revision of the Driver's specification, so we remove the property
// and call advanceOperationTime() ourselves.
delete args.sessionOptions.initialOperationTime;
}
const session = mongo.startSession(args.sessionOptions);
const readPreference = session.getOptions().getReadPreference();
if (readPreference && readPreference.mode === "secondary") {
// Unset the explicit read preference so set_read_preference_secondary.js can do
// the right thing based on the DB.
session.getOptions().setReadPreference(undefined);
// We import set_read_preference_secondary.js in order to avoid running
// commands against the "admin" and "config" databases via mongos with
// readPreference={mode: "secondary"} when there's only a single node in
// the CSRS.
await import("jstests/libs/override_methods/set_read_preference_secondary.js");
}
if (typeof initialClusterTime !== "undefined") {
session.advanceClusterTime(initialClusterTime);
}
if (typeof initialOperationTime !== "undefined") {
session.advanceOperationTime(initialOperationTime);
}
myDB = session.getDatabase(args.dbName);
} else {
myDB = mongo.getDB(args.dbName);
}
{
let connectionDesc = "";
// In sharded environments, mongos is acting as a proxy for the mongo shell and
// therefore has a different outbound port than the 'whatsmyuri' command returns.
if (!Cluster.isSharded(args.clusterOptions) && !TestData.testingReplicaSetEndpoint) {
let res = assert.commandWorked(myDB.runCommand({whatsmyuri: 1}));
const myUri = res.you;
res = assert.commandWorked(myDB.adminCommand({currentOp: 1, client: myUri}));
connectionDesc = ", conn:" + res.inprog[0].desc;
}
const printOriginal = print;
print = function () {
const printArgs = Array.from(arguments);
const prefix = "[tid:" + args.tid + connectionDesc + "]";
printArgs.unshift(prefix);
return printOriginal(printArgs.join(" "));
};
jsTestLog = function (msg) {
if (typeof msg === "object") {
msg = tojson(msg);
}
assert.eq(typeof msg, "string", "Received: " + msg);
let msgs = msg.split("\n");
msgs = msgs.map((msg) => "[tid:" + args.tid + "] " + msg);
const printMsgs = ["----", ...msgs, "----"].map((s) => `[jsTest] ${s}`);
printOriginal(`\n\n${printMsgs.join("\n")}\n\n`);
};
}
if (TestData.shardsAddedRemoved) {
await import("jstests/libs/override_methods/implicitly_retry_on_shard_transition_errors.js");
}
if (TestData.runningWithConfigStepdowns || TestData.runningWithShardStepdowns || TestData.killShards) {
await import("jstests/libs/override_methods/implicitly_retry_crud_on_no_progress_made.js");
}
if (TestData.runningWithBalancer && !TestData.shardsAddedRemoved) {
// Skipping the import on shard transitions because it's already imported under
// implicitly_retry_on_shard_transition_errors.js
await import("jstests/libs/override_methods/implicitly_retry_on_migration_in_progress.js");
}
if (TestData.fuzzRuntimeParams) {
await import(
"jstests/libs/override_methods/implicitly_retry_on_conflicting_operation_during_fuzztest.js"
);
}
if (Cluster.isReplication(args.clusterOptions)) {
if (
args.clusterOptions.hasOwnProperty("sharded") &&
args.clusterOptions.sharded.hasOwnProperty("stepdownOptions") &&
args.clusterOptions.sharded.stepdownOptions.shardStepdown
) {
const newOptions = {
alwaysInjectTransactionNumber: true,
defaultReadConcernLevel: "majority",
logRetryAttempts: true,
overrideRetryAttempts: 3,
};
Object.assign(TestData, newOptions);
assert(!TestData.hasOwnProperty("networkErrorAndTxnOverrideConfig"), TestData);
TestData.networkErrorAndTxnOverrideConfig = {retryOnNetworkErrors: true};
await import("jstests/libs/override_methods/network_error_and_txn_override.js");
if (TestData.killShards) {
// TODO (SERVER-107404): Remove the override below.
await import("jstests/libs/override_methods/implicitly_retry_resharding.js");
}
}
// Operations that run after a "dropDatabase" command has been issued may fail with
// a "DatabaseDropPending" error response if they would create a new collection on
// that database while we're waiting for a majority of nodes in the replica set to
// confirm it has been dropped. We load the
// implicitly_retry_on_database_drop_pending.js file to make it so that the clients
// started by the concurrency framework automatically retry their operation in the
// face of this particular error response.
await import("jstests/libs/override_methods/implicitly_retry_on_database_drop_pending.js");
}
if (TestData.defaultReadConcernLevel || TestData.defaultWriteConcern) {
await import("jstests/libs/override_methods/set_read_and_write_concerns.js");
}
for (const workload of workloads) {
const {$config} = await import(workload);
let config = parseConfig($config); // to normalize
// Copy any modifications that were made to $config.data
// during the setup function of the workload (see caveat
// below).
// XXX: Changing the order of extend calls causes problems
// for workloads that reference $super.
// Suppose you have workloads A and B, where workload B extends
// workload A. The $config.data of workload B can define a
// function that closes over the $config object of workload A
// (known as $super to workload B). This reference is lost when
// the config object is serialized to BSON, which results in
// undefined variables in the derived workload.
let data = Object.extend({}, args.data[workload], true);
data = Object.extend(data, config.data, true);
// Object.extend() defines all properties added to the destination object as
// configurable, enumerable, and writable. To prevent workloads from changing
// the iterations and threadCount properties in their state functions, we
// redefine them here as non-configurable and non-writable.
Object.defineProperties(data, {
"iterations": {configurable: false, writable: false, value: data.iterations},
"threadCount": {configurable: false, writable: false, value: data.threadCount},
});
data.tid = args.tid;
configs[workload] = {
data: data,
db: myDB,
collName: args.collName,
cluster: args.cluster,
iterations: data.iterations,
passConnectionCache: config.passConnectionCache,
startState: config.startState,
states: config.states,
tid: args.tid,
transitions: config.transitions,
errorLatch: args.errorLatch,
numThreads: args.numThreads,
};
}
args.latch.countDown();
// Converts any exceptions to a return status. In order for the
// parent thread to call countDown() on our behalf, we must throw
// an exception. Nothing prior to (and including) args.latch.countDown()
// should be wrapped in a try/catch statement.
try {
args.latch.await(); // wait for all threads to start
Random.setRandomSeed(args.seed);
await run(configs);
return {ok: 1};
} catch (e) {
jsTest.log("Thread failed. Other threads will stop execution on next iteration.");
args.errorLatch.countDown();
return {
ok: 0,
err: e.toString(),
stack: e.stack,
tid: args.tid,
workloads: workloads,
};
}
} finally {
// Kill this worker thread's session to ensure any possible idle cursors left open by
// the workload are closed.
// TODO SERVER-74993: Remove this.
try {
let session = myDB.getSession();
if (session) {
myDB.runCommand({killSessions: [session.getSessionId()]});
}
} catch (e) {
// Ignore errors from killSessions.
jsTest.log("Error running killSessions: " + e);
}
// Avoid retention of connection object
configs = null;
myDB = null;
gc();
}
}
return {main: main};
})();