GODRIVER-3638 Prohibit using failpoints on sharded topologies.

[matthewdale]

GODRIVER-3638

Summary

• If SetFailPoint is called when mtest is connected to a sharded topology.
• Add an mtest option AllowFailPointsOnSharded to allow using failpoints on sharded topologies.

Background & Motivation

On sharded topologies, failpoints are applied to only a single mongoS. If the driver is connected to multiple mongoS instances, there's a possibility a different mongoS will be selected for a subsequent command. In that case, the failpoint is effectively ignored, leading to a test failure that is extremely difficult to diagnose.

Note that if we implement GODRIVER-3328, we no longer need to prohibit using failpoints on sharded topologies.

[mongodb-drivers-pr-bot]

API Change Report

No changes found!

[mongodb-drivers-pr-bot]

🧪 Performance Results

Commit SHA: de83205

The following benchmark tests for version 689ebf5509236500071bcbf9 had statistically significant changes (i.e., |z-score| > 1.96):

Benchmark	Measurement	% Change	Patch Value	Stable Region	H-Score	Z-Score
BenchmarkMultiInsertLargeDocument	total_time_seconds	17.5686	1.2747	Avg: 1.0842 Med: 1.0816 Stdev: 0.0488	0.8534	3.9068
BenchmarkBSONDeepDocumentDecoding	ops_per_second_min	-12.7165	1724.9141	Avg: 1976.2210 Med: 1984.2671 Stdev: 121.0901	0.7361	-2.0754
BenchmarkSingleFindOneByID	ns_per_op	12.3064	280211.0000	Avg: 249505.8032 Med: 248200.0000 Stdev: 7712.0254	0.8686	3.9815
BenchmarkSingleFindOneByID	ops_per_second_med	-11.4774	3644.3414	Avg: 4116.8474 Med: 4138.1822 Stdev: 116.3377	0.8713	-4.0615
BenchmarkSingleRunCommand	total_bytes_allocated	-11.2053	87749080.0000	Avg: 98822432.8182 Med: 99663608.0000 Stdev: 5104085.6817	0.7753	-2.1695
BenchmarkSingleRunCommand	total_mem_allocs	-10.8790	950527.0000	Avg: 1066557.3750 Med: 1075540.5000 Stdev: 54558.0767	0.7703	-2.1267
BenchmarkLargeDocInsertOne	ops_per_second_med	-8.7716	5208.2113	Avg: 5708.9783 Med: 5716.2701 Stdev: 153.3932	0.8383	-3.2646
BenchmarkSingleFindOneByID	ops_per_second_max	-6.3893	4330.8229	Avg: 4626.4169 Med: 4626.5025 Stdev: 106.1006	0.8106	-2.7860
BenchmarkBSONFullDocumentEncoding	total_bytes_allocated	-5.4923	229160736.0000	Avg: 242478287.0909 Med: 242695476.0000 Stdev: 3720589.4273	0.8429	-3.5794
BenchmarkBSONFullDocumentEncoding	total_mem_allocs	-5.4816	1323641.0000	Avg: 1400405.4773 Med: 1401957.0000 Stdev: 21670.3891	0.8412	-3.5424
BenchmarkBSONFullDocumentEncoding	ns_per_op	5.2213	26972.0000	Avg: 25633.5909 Med: 25543.5000 Stdev: 379.8047	0.8432	3.5239
BenchmarkBSONFullDocumentEncoding	ops_per_second_med	-3.6557	40217.1727	Avg: 41743.1955 Med: 41743.2250 Stdev: 579.9676	0.7902	-2.6312
BenchmarkBSONFullDocumentEncoding	ops_per_second_max	-3.3123	41720.5557	Avg: 43149.8004 Med: 43119.2521 Stdev: 527.6487	0.7955	-2.7087
BenchmarkSingleRunCommand	allocated_bytes_per_op	0.1269	12212.0000	Avg: 12196.5217 Med: 12196.0000 Stdev: 4.3991	0.8537	3.5185
BenchmarkBSONFlatDocumentEncoding	allocated_bytes_per_op	0.0980	6278.0000	Avg: 6271.8523 Med: 6272.0000 Stdev: 3.0078	0.7342	2.0439

For a comprehensive view of all microbenchmark results for this PR's commit, please check out the Evergreen perf task for this patch.

[Copilot]

Pull Request Overview

This PR implements a safety mechanism to prevent unreliable failpoint usage on sharded MongoDB topologies by prohibiting SetFailPoint calls unless explicitly allowed through a new option.

• Adds AllowFailPointsOnSharded() option to bypass the failpoint prohibition on sharded topologies
• Modifies existing tests to either exclude sharded topologies or explicitly allow failpoints when needed
• Implements runtime validation in SetFailPoint to prevent accidental failpoint usage on sharded clusters

Reviewed Changes

Copilot reviewed 8 out of 8 changed files in this pull request and generated no comments.

Show a summary per file

File	Description
internal/integration/mtest/options.go	Adds AllowFailPointsOnSharded() method to enable failpoint usage on sharded topologies
internal/integration/mtest/mongotest.go	Implements failpoint validation logic and stores the allowFailPointsOnSharded flag
internal/integration/server_selection_prose_test.go	Uses AllowFailPointsOnSharded() for a test that requires failpoints on sharded topology
internal/integration/retryable_writes_prose_test.go	Excludes sharded topology from one test and allows failpoints for another
internal/integration/retryable_reads_prose_test.go	Excludes sharded topology from one test and allows failpoints for another
internal/integration/sdam_prose_test.go	Excludes sharded topology from RTT monitoring test
internal/integration/csot_prose_test.go	Excludes sharded topology from multiple GridFS timeout tests
internal/integration/crud_prose_test.go	Excludes sharded topology from bulk write error handling tests

---

_{Tip: Customize your code reviews with copilot-instructions.md. Create the file or learn how to get started.
You can also share your feedback on Copilot code review for a chance to win a $100 gift card. Take the survey.}