mongodb/mongo
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

SERVER-83435 Extend sharding concurrency suites with unsharded collections state (#53299)

Browse Source 
GitOrigin-RevId: ac895885b45801434dfa5fbd6b792aa372484468
This commit is contained in:
Enrico Golfieri 2026-05-26 13:06:08 +02:00 committed by MongoDB Bot
parent 26d3dfa093
commit 10787747e6
25 changed files with 125 additions and 18 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

1
buildscripts/resmokeconfig/matrix_suites/generated_suites/concurrency_sharded_replication_execution_control_with_prioritization.yml
View File

@ -21,6 +21,7 @@ executor:
TestData:
implicitlyShardOnCreateCollectionOnly: true
runningWithBalancer: false
shardCollectionProbability: 0.5
fixture:
class: ShardedClusterFixture
enable_balancer: false

1
buildscripts/resmokeconfig/matrix_suites/generated_suites/concurrency_sharded_replication_last_continuous_new_old_old_new.yml
View File

@ -22,6 +22,7 @@ executor:
enableOTELTracing: false
implicitlyShardOnCreateCollectionOnly: true
runningWithBalancer: false
shardCollectionProbability: 0.5
fixture:
class: ShardedClusterFixture
enable_balancer: false

1
buildscripts/resmokeconfig/matrix_suites/generated_suites/concurrency_sharded_replication_last_lts_new_old_old_new.yml
View File

@ -22,6 +22,7 @@ executor:
enableOTELTracing: false
implicitlyShardOnCreateCollectionOnly: true
runningWithBalancer: false
shardCollectionProbability: 0.5
fixture:
class: ShardedClusterFixture
enable_balancer: false

1
buildscripts/resmokeconfig/matrix_suites/generated_suites/concurrency_sharded_replication_primary_driven_index_builds.yml
View File

@ -21,6 +21,7 @@ executor:
TestData:
implicitlyShardOnCreateCollectionOnly: true
runningWithBalancer: false
shardCollectionProbability: 0.5
fixture:
class: ShardedClusterFixture
enable_balancer: false

1
buildscripts/resmokeconfig/suites/concurrency_sharded_causal_consistency.yml
View File

@ -54,6 +54,7 @@ executor:
runningWithCausalConsistency: true
runningWithBalancer: false
implicitlyShardOnCreateCollectionOnly: true
shardCollectionProbability: 0.5
hooks:
# TODO SERVER-120162: Re-enable once all the false positive have been addressed and the test is stable.
# - class: CheckIdleCursors

1
buildscripts/resmokeconfig/suites/concurrency_sharded_causal_consistency_and_balancer.yml
View File

@ -67,6 +67,7 @@ executor:
runningWithCausalConsistency: true
runningWithBalancer: true
implicitlyShardOnCreateCollectionOnly: true
shardCollectionProbability: 0.5
hooks:
# TODO SERVER-120162: Re-enable once all the false positive have been addressed and the test is stable.

1
buildscripts/resmokeconfig/suites/concurrency_sharded_initial_sync.yml
View File

@ -126,6 +126,7 @@ executor:
killShards: true
implicitlyShardOnCreateCollectionOnly: true
useActionPermittedFile: [ContinuousInitialSync]
shardCollectionProbability: 0.5
hooks:
- class: ContinuousInitialSync

1
buildscripts/resmokeconfig/suites/concurrency_sharded_local_read_write_multi_stmt_txn.yml
View File

@ -104,6 +104,7 @@ executor:
runningWithSessions: true
traceExceptions: false
implicitlyShardOnCreateCollectionOnly: true
shardCollectionProbability: 0.5
hooks:
# TODO SERVER-120162: Re-enable once all the false positive have been addressed and the test is stable.

1
buildscripts/resmokeconfig/suites/concurrency_sharded_local_read_write_multi_stmt_txn_with_balancer.yml
View File

@ -119,6 +119,7 @@ executor:
traceExceptions: false
runningWithBalancer: true
implicitlyShardOnCreateCollectionOnly: true
shardCollectionProbability: 0.5
hooks:
# TODO SERVER-120162: Re-enable once all the false positive have been addressed and the test is stable.

1
buildscripts/resmokeconfig/suites/concurrency_sharded_multi_stmt_txn.yml
View File

@ -95,6 +95,7 @@ executor:
runningWithSessions: true
traceExceptions: false
implicitlyShardOnCreateCollectionOnly: true
shardCollectionProbability: 0.5
hooks:
# TODO SERVER-120162: Re-enable once all the false positive have been addressed and the test is stable.
# - class: CheckIdleCursors

1
buildscripts/resmokeconfig/suites/concurrency_sharded_multi_stmt_txn_stepdown_terminate_kill_primary.yml
View File

@ -163,6 +163,7 @@ executor:
traceExceptions: false
implicitlyShardOnCreateCollectionOnly: true
useActionPermittedFile: [ContinuousStepdown]
shardCollectionProbability: 0.5
hooks:
# We use a stepdown interval of 15 seconds because we will retry all commands in a transaction
# so we need to allow time for at most 10 operations to be re-run and then re-committed. If

1
buildscripts/resmokeconfig/suites/concurrency_sharded_multi_stmt_txn_with_balancer.yml
View File

@ -99,6 +99,7 @@ executor:
traceExceptions: false
runningWithBalancer: true
implicitlyShardOnCreateCollectionOnly: true
shardCollectionProbability: 0.5
hooks:
# TODO SERVER-120162: Re-enable once all the false positive have been addressed and the test is stable.

1
buildscripts/resmokeconfig/suites/concurrency_sharded_replication.yml
View File

@ -34,6 +34,7 @@ executor:
TestData:
runningWithBalancer: false
implicitlyShardOnCreateCollectionOnly: true
shardCollectionProbability: 0.5
hooks:
# TODO SERVER-120162: Re-enable once all the false positive have been addressed and the test is stable.
# - class: CheckIdleCursors

1
buildscripts/resmokeconfig/suites/concurrency_sharded_replication_with_balancer.yml
View File

@ -70,6 +70,7 @@ executor:
TestData:
runningWithBalancer: true
implicitlyShardOnCreateCollectionOnly: true
shardCollectionProbability: 0.5
hooks:
# TODO SERVER-120162: Re-enable once all the false positive have been addressed and the test is stable.
# - class: CheckIdleCursors

1
buildscripts/resmokeconfig/suites/concurrency_sharded_stepdown_terminate_kill_primary_with_balancer.yml
View File

@ -97,6 +97,7 @@ executor:
useActionPermittedFile: [ContinuousStepdown]
runningWithBalancer: true
implicitlyShardOnCreateCollectionOnly: true
shardCollectionProbability: 0.5
hooks:
- class: ContinuousStepdown
config_stepdown: true

1
buildscripts/resmokeconfig/suites/concurrency_sharded_with_balancer_and_config_shard.yml
View File

@ -68,6 +68,7 @@ executor:
TestData:
runningWithBalancer: true
implicitlyShardOnCreateCollectionOnly: true
shardCollectionProbability: 0.5
hooks:
# TODO SERVER-120162: Re-enable once all the false positive have been addressed and the test is stable.

1
buildscripts/resmokeconfig/suites/concurrency_sharded_with_config_shard.yml
View File

@ -56,6 +56,7 @@ executor:
TestData:
runningWithBalancer: false
implicitlyShardOnCreateCollectionOnly: true
shardCollectionProbability: 0.5
hooks:
# TODO SERVER-120162: Re-enable once all the false positive have been addressed and the test is stable.
# - class: CheckIdleCursors

1
buildscripts/resmokeconfig/suites/concurrency_sharded_with_stepdowns.yml
View File

@ -89,6 +89,7 @@ executor:
runningWithBalancer: false
implicitlyShardOnCreateCollectionOnly: true
useActionPermittedFile: [ContinuousStepdown]
shardCollectionProbability: 0.5
hooks:
- class: ContinuousStepdown
config_stepdown: true

28
jstests/concurrency/fsm_workloads/crud/move_collection_with_crud.js
View File

@ -14,6 +14,7 @@
*/
import {ShardingTopologyHelpers} from "jstests/concurrency/fsm_workload_helpers/catalog_and_routing/sharding_topology_helpers.js";
import {FixtureHelpers} from "jstests/libs/fixture_helpers.js";
export const $config = (function () {
const kTotalWorkingDocuments = 500;
@ -55,6 +56,8 @@ export const $config = (function () {
};
let currentShard = currentShardFn();
let shards = ShardingTopologyHelpers.getShardNames(mongosConn);
// If the collection was never tracked, currentShard is undefined and any shard is a valid
// moveCollection destination.
let destinationShards = shards.filter(function (shard) {
if (shard !== currentShard) {
return shard;
@ -136,13 +139,24 @@ export const $config = (function () {
function setup(db, collName, _cluster) {
const ns = db + "." + collName;
print(`Started unshardCollection on ${ns}`);
assert.commandWorked(db.adminCommand({unshardCollection: ns}));
print(`Finished unshardCollection on ${ns}`);
// Calculate the primary shard
let unshardedColl = db.getSiblingDB("config").collections.findOne({_id: ns});
let chunk = db.getSiblingDB("config").chunks.findOne({uuid: unshardedColl.uuid});
this.primaryShard = chunk.shard;
const result = db.adminCommand({unshardCollection: ns});
if (result.ok) {
print(`Finished unshardCollection on ${ns}`);
let unshardedColl = db.getSiblingDB("config").collections.findOne({_id: ns});
let chunk = db.getSiblingDB("config").chunks.findOne({uuid: unshardedColl.uuid});
this.primaryShard = chunk.shard;
} else if (
result.code === ErrorCodes.NamespaceNotFound &&
FixtureHelpers.maySkipImplicitSharding() &&
FixtureHelpers.isUntracked(db.getCollection(collName))
) {
// When implicit sharding may be skipped, the collection may not have been sharded, so
// the database may not be registered in the cluster catalog.
this.primaryShard = undefined;
print(`unshardCollection skipped for ${ns}: collection is not tracked in cluster catalog`);
} else {
assert.commandWorked(result);
}
const coll = db.getCollection(collName);
assert.commandWorked(coll.insert(createDocuments(kTotalWorkingDocuments)));

8
jstests/concurrency/fsm_workloads/ddl/rename_collection/collection_uuid.js
View File

@ -49,6 +49,9 @@ const runCommandInLoop = function (db, namespace, cmdName, cmdObj, data, expecte
ErrorCodes.BackgroundOperationInProgressForNamespace,
ErrorCodes.ReshardCollectionInProgress,
ErrorCodes.QueryPlanKilled,
ErrorCodes.MaxTimeMSExpired,
// A rename can starve behind another concurrent DDL operation holding the namespace lock.
ErrorCodes.LockBusy,
// StaleConfig is usually retried by the mongos, but in situations where multiple errors
// have ocurred on the same batch and MultipleErrorsOcurred is returned, one of the errors
// could be StaleConfig and the other could be one that mongos does not retry the batch on.
@ -273,9 +276,12 @@ export const $config = (function () {
let otherDbShard;
if (numShards > 1) {
// currDb may be null if implicit sharding is skipped, leaving the database
// unregistered in config.databases (no sharded collection created yet).
const currDb = db.getSiblingDB("config")["databases"].findOne({_id: db.getName()});
const primaryShard = currDb ? currDb.primary : null;
shardNames.some((shard) => {
if (shard != currDb.primary) {
if (shard != primaryShard) {
otherDbShard = shard;
return false;
}

7
jstests/concurrency/fsm_workloads/ddl/rename_collection/collection_uuid_sharded.js
View File

@ -16,6 +16,7 @@ import {
$config as $baseConfig,
testCommand,
} from "jstests/concurrency/fsm_workloads/ddl/rename_collection/collection_uuid.js";
import {FixtureHelpers} from "jstests/libs/fixture_helpers.js";
export const $config = extendWorkload($baseConfig, function ($config, $super) {
const origStates = Object.keys($config.states);
@ -24,6 +25,12 @@ export const $config = extendWorkload($baseConfig, function ($config, $super) {
shardingCommands: function shardingCommands(db, collName) {
const namespace = db.getName() + "." + collName;
// When implicit sharding may be skipped, the collection may be unsharded and absent
// from the cluster catalog. All commands below require it to be tracked.
if (FixtureHelpers.maySkipImplicitSharding() && FixtureHelpers.isUntracked(db, this.collUUID)) {
return;
}
// ShardCollection should fail as the collection is already sharded.
let shardCollectionCmd = {
shardCollection: namespace,

20
jstests/concurrency/fsm_workloads/query/timeseries/timeseries_agg_out.js
View File

@ -25,7 +25,9 @@ import {$config as $baseConfig} from "jstests/concurrency/fsm_workloads/query/ag
import {
areViewlessTimeseriesEnabled,
getTimeseriesBucketsColl,
isTrackedTimeseries,
} from "jstests/core/timeseries/libs/viewless_timeseries_util.js";
import {FixtureHelpers} from "jstests/libs/fixture_helpers.js";
export const $config = extendWorkload($baseConfig, function ($config, $super) {
const timeFieldName = "time";
@ -39,6 +41,8 @@ export const $config = extendWorkload($baseConfig, function ($config, $super) {
*/
$config.states.query = function query(db, collName) {
jsTestLog(`Running query: coll=${collName} out=${this.outputCollName}`);
const sourceCollectionWasUntrackedBeforeQuery =
FixtureHelpers.maySkipImplicitSharding() && !isTrackedTimeseries(db.getCollection(collName));
const res = db[collName].runCommand({
aggregate: collName,
pipeline: [
@ -79,7 +83,7 @@ export const $config = extendWorkload($baseConfig, function ($config, $super) {
// TODO (SERVER-88275) a moveCollection can cause the original collection to be dropped and
// re-created with a different uuid, causing the aggregation to fail with QueryPlannedKilled
// when the mongos is fetching data from the shard using getMore(). Remove
// the interruptedQueryErrors from allowedErrorCodes once this bug is being addressed
// the interruptedQueryErrors from allowedErrorCodes once this bug is being addressed.
if (TestData.runningWithBalancer) {
allowedErrorCodes = allowedErrorCodes.concat(interruptedQueryErrors);
// On slow builds with the balancer enabled, it is possible for the router to exhaust
@ -87,6 +91,13 @@ export const $config = extendWorkload($baseConfig, function ($config, $super) {
// to the client.
allowedErrorCodes.push(ErrorCodes.StaleConfig);
}
// When the source collection is untracked, we might encounter a concurrent movePrimary
// which drops and recreates the collection on the donor shard, triggering the same class of
// interruption (QueryPlanKilled / NamespaceNotFound).
if (sourceCollectionWasUntrackedBeforeQuery) {
allowedErrorCodes.push(ErrorCodes.NamespaceNotFound);
allowedErrorCodes.push(ErrorCodes.QueryPlanKilled);
}
assert.commandWorkedOrFailedWithCode(res, allowedErrorCodes);
if (res.ok) {
@ -165,6 +176,13 @@ export const $config = extendWorkload($baseConfig, function ($config, $super) {
if (TestData.runningWithBalancer) {
allowedErrorCodes.push(ErrorCodes.StaleConfig);
}
if (FixtureHelpers.maySkipImplicitSharding()) {
// In suites that only sometimes implicitly track the timeseries collection, the router
// could exhaust the retry attempts on StaleConfig because we repeatedly drop and
// re-create the collection, causing its shard version to change continuously.
allowedErrorCodes.push(ErrorCodes.StaleConfig);
}
assert.commandWorkedOrFailedWithCode(
db.createCollection(this.outputCollName, {
timeseries: {timeField: timeFieldName, metaField: metaFieldName},

17
jstests/concurrency/fsm_workloads/random_update_delete/random_update_delete_with_move_collection.js
View File

@ -20,6 +20,7 @@
import {extendWorkload} from "jstests/concurrency/fsm_libs/extend_workload.js";
import {ShardingTopologyHelpers} from "jstests/concurrency/fsm_workload_helpers/catalog_and_routing/sharding_topology_helpers.js";
import {randomUpdateDelete} from "jstests/concurrency/fsm_workload_modifiers/random_update_delete.js";
import {FixtureHelpers} from "jstests/libs/fixture_helpers.js";
const $baseConfig = {
threadCount: 5,
@ -43,7 +44,21 @@ const $baseConfig = {
// The runner will implicitly shard the collection if we are in a sharded cluster, so
// unshard it.
assert.commandWorked(db.adminCommand({unshardCollection: `${db}.${collName}`}));
const ns = `${db}.${collName}`;
const result = db.adminCommand({unshardCollection: ns});
if (!result.ok) {
// When implicit sharding is skipped, the collection may not have been sharded,
// so the database may not be registered in the cluster catalog. In that case the
// collection is already effectively unsharded and no action is needed.
const collectionWasAlreadyUntracked =
result.code === ErrorCodes.NamespaceNotFound &&
FixtureHelpers.maySkipImplicitSharding() &&
FixtureHelpers.isUntracked(db.getCollection(collName));
// Ignore the error if the collection was already untracked. Otherwise, force the test to fail.
if (!collectionWasAlreadyUntracked) {
assert.commandWorked(result);
}
}
const bulk = db[collName].initializeUnorderedBulkOp();
for (let i = 0; i < this.threadCount * 200; ++i) {
bulk.insert({_id: i});

26
jstests/concurrency/fsm_workloads/random_update_delete/random_update_delete_with_resharding.js
View File

@ -19,6 +19,7 @@
*/
import {extendWorkload} from "jstests/concurrency/fsm_libs/extend_workload.js";
import {randomUpdateDelete} from "jstests/concurrency/fsm_workload_modifiers/random_update_delete.js";
import {FixtureHelpers} from "jstests/libs/fixture_helpers.js";
const $baseConfig = {
threadCount: 5,
@ -79,14 +80,23 @@ export const $config = extendWorkload($partialConfig, function ($config, $super)
$config.states.reshardCollection = function reshardCollection(db, collName, connCache) {
const namespace = `${db}.${collName}`;
jsTestLog(`Attempting to reshard collection ${namespace}`);
const result = assert.commandWorked(
db.adminCommand({
reshardCollection: namespace,
key: this.getShardKey(collName),
numInitialChunks: 1,
forceRedistribution: true,
}),
);
const result = db.adminCommand({
reshardCollection: namespace,
key: this.getShardKey(collName),
numInitialChunks: 1,
forceRedistribution: true,
});
if (
result.code === ErrorCodes.NamespaceNotFound &&
FixtureHelpers.maySkipImplicitSharding() &&
FixtureHelpers.isUntracked(db.getCollection(collName))
) {
// When implicit sharding is skipped, the collection may not have been sharded at
// setup time, making reshardCollection legitimately fail with NamespaceNotFound.
jsTestLog(`reshardCollection skipped for ${namespace}: collection is not sharded`);
return;
}
assert.commandWorked(result);
jsTestLog(`Reshard collection result for ${namespace}: ${tojson(result)}`);
};

19
jstests/libs/fixture_helpers.js
View File

@ -107,6 +107,23 @@ export var FixtureHelpers = (function () {
return isSharded(coll) || isUnsplittable(coll);
}
/**
* Looks for an entry in the sharding catalog for the given collection to check whether it is
* absent.
*
* TODO (SERVER-86443): remove this utility once all collections are tracked.
*/
function isUntracked(collOrDb, collUUID = undefined) {
if (collUUID !== undefined) {
return collOrDb.getSiblingDB("config").collections.findOne({uuid: collUUID}) === null;
}
return !isTracked(collOrDb);
}
function maySkipImplicitSharding() {
return typeof TestData.shardCollectionProbability !== "undefined" && TestData.shardCollectionProbability < 1;
}
/**
* Returns an array with the shardIds that own data for the given collection.
*/
@ -290,6 +307,8 @@ export var FixtureHelpers = (function () {
isSharded: isSharded,
isUnsplittable: isUnsplittable,
isTracked: isTracked,
isUntracked: isUntracked,
maySkipImplicitSharding: maySkipImplicitSharding,
areCollectionsColocated: areCollectionsColocated,
getShardsOwningDataForCollection: getShardsOwningDataForCollection,
getTopologyTime: getTopologyTime,