Prototype crash-resilient TRX writes by nohwnd · Pull Request #10106 · microsoft/testfx

nohwnd · 2026-07-21T07:59:57Z

Wanted to implement both options and see what actually survives interrupted writes. Neither implementation is wired into production.

The padded file cannot provide the guarantee from the issue. Across 5,656 injected operation failures and byte cuts:

4,545 were malformed
719 were parseable but inconsistent
22 were repairable
370 were truthful

The append-only journal recovered a complete prefix in all 246 interruption cases, including restart and append after a torn final record. Publishing complete streamed snapshots through sibling-file replacement kept the destination old or new and truthful in all 3,519 observations.

For 10k synthetic results, journal plus snapshots did 10,004 flushes and wrote 18.3 MB. The padded writer did 89,959 flushes and wrote 24.6 MB. Replay keeps one record, no result collection, and no XDocument.

Changes

Add an isolated padded incremental writer with fixed pads, counters, running slots, reflow, and compaction.
Add an append-only journal with torn-tail recovery and streamed snapshot publication.
Add deterministic operation and partial-write injection, semantic TRX classification, concurrent-reader barriers, and real-file-system tests.
Cover CRLF preservation, UTF-8-safe truncation, long metadata, large output, summary data, startup overwrite, replacement sharing modes, and 10k-result diagnostics.

Verification

148 focused tests passed across net462, net472, net8.0, and net9.0.
3,432 extension unit tests passed, with 20 existing skips.
20 Release diagnostics passed.
TRX acceptance passed: 40 MTP, 6 MSTest TRX, and 6 MSTest output tests.
Debug and Release builds, full repository build, and package production passed.

This does not claim XSD, Visual Studio, Azure DevOps, third-party reader, network filesystem, antivirus, physical power-loss, or actual attachment-copying compatibility. Atomic overwrite intentionally fails closed on .NET Framework where the same replacement guarantee is not available.

LMK if this should stay as evidence only, or if I should follow with the production journal and snapshot path.

🤖

Add isolated padded in-place and journal plus snapshot writers without wiring the production lifecycle. Run deterministic operation and byte-cut interruption matrices, concurrent readers, compatibility probes, and 10k diagnostics. The evidence disproves arbitrary-termination validity for padded mutation and supports append-only journal recovery plus atomic sibling snapshots. Fix torn journal tails before resumed appends and keep prototype summary children in schema order. Co-authored-by: Copilot App <223556219+Copilot@users.noreply.github.com> 🤖

…otype

Co-authored-by: Copilot App <223556219+Copilot@users.noreply.github.com> 🤖

Copilot

Pull request overview

Prototypes and evaluates crash-resilient TRX persistence without integrating either approach into production.

Changes:

Adds padded incremental and journal/snapshot TRX prototypes.
Adds atomic file publication and fault-injection infrastructure.
Adds semantic, interruption, compatibility, and performance evidence tests.

Reviewed changes

Copilot reviewed 24 out of 24 changed files in this pull request and generated 6 comments.

Show a summary per file

File	Description
`src/Platform/Microsoft.Testing.Extensions.TrxReport/InternalAPI/InternalAPI.Unshipped.txt`	Tracks new internal APIs.
`src/Platform/Microsoft.Testing.Extensions.TrxReport/Streaming/Prototype/ITrxPrototypeFileOperations.cs`	Defines prototype file abstractions.
`src/Platform/Microsoft.Testing.Extensions.TrxReport/Streaming/Prototype/TrxIncrementalWriterPrototype.cs`	Implements padded incremental writes.
`src/Platform/Microsoft.Testing.Extensions.TrxReport/Streaming/Prototype/TrxJournalSnapshotPrototype.cs`	Implements journal replay and snapshots.
`src/Platform/Microsoft.Testing.Extensions.TrxReport/Streaming/Prototype/TrxPrototypeFileOperations.cs`	Provides real filesystem operations.
`src/Platform/Microsoft.Testing.Extensions.TrxReport/Streaming/Prototype/TrxPrototypeXmlRenderer.cs`	Renders prototype TRX XML.
`src/Platform/Microsoft.Testing.Extensions.TrxReport/Streaming/Prototype/TrxSnapshotPublisherPrototype.cs`	Publishes snapshots by sibling replacement.
`test/UnitTests/Microsoft.Testing.Extensions.UnitTests/Helpers/TrxDocumentClassifier.cs`	Classifies TRX consistency.
`test/UnitTests/Microsoft.Testing.Extensions.UnitTests/Helpers/TrxFaultInjectingFileOperations.cs`	Simulates filesystem faults.
`test/UnitTests/Microsoft.Testing.Extensions.UnitTests/Helpers/TrxPrototypeDiagnostics.cs`	Collects prototype metrics.
`test/UnitTests/Microsoft.Testing.Extensions.UnitTests/Helpers/TrxPrototypeScenarioFactory.cs`	Builds deterministic scenarios.
`test/UnitTests/Microsoft.Testing.Extensions.UnitTests/TrxCompatibilityEvidenceTests.cs`	Records compatibility boundaries.
`test/UnitTests/Microsoft.Testing.Extensions.UnitTests/TrxConcurrentReaderTests.cs`	Exercises concurrent readers.
`test/UnitTests/Microsoft.Testing.Extensions.UnitTests/TrxCrashConsistencyContractTests.cs`	Captures canonical TRX semantics.
`test/UnitTests/Microsoft.Testing.Extensions.UnitTests/TrxDocumentClassifierTests.cs`	Tests semantic classification.
`test/UnitTests/Microsoft.Testing.Extensions.UnitTests/TrxFaultInjectingFileOperationsTests.cs`	Tests fault simulation behavior.
`test/UnitTests/Microsoft.Testing.Extensions.UnitTests/TrxIncrementalWriterInterruptionTests.cs`	Sweeps padded-writer interruptions.
`test/UnitTests/Microsoft.Testing.Extensions.UnitTests/TrxIncrementalWriterPrototypeTests.cs`	Tests incremental rendering.
`test/UnitTests/Microsoft.Testing.Extensions.UnitTests/TrxJournalSnapshotPrototypeTests.cs`	Tests journal recovery and publication.
`test/UnitTests/Microsoft.Testing.Extensions.UnitTests/TrxPrototypeDiagnosticsTests.cs`	Verifies large-run diagnostics.
`test/UnitTests/Microsoft.Testing.Extensions.UnitTests/TrxPrototypeFileOperationsTests.cs`	Tests real file operations.
`test/UnitTests/Microsoft.Testing.Extensions.UnitTests/TrxPrototypeRealFileSystemTests.cs`	Exercises real replacement behavior.
`test/UnitTests/Microsoft.Testing.Extensions.UnitTests/TrxPrototypeWarningSuppressions.cs`	Suppresses prototype-test style warnings.
`test/UnitTests/Microsoft.Testing.Extensions.UnitTests/TrxSnapshotPublisherPrototypeTests.cs`	Tests reflow and atomic publication.

+        if (!_operations.Exists(_journalPath))
+        {
+            _journalTailValidated = true;
+            return;


+        Replay(
+            record =>
+            {
+                byte[] fragment = _renderer.RenderCompletedResult(record.Result, record.ExecutionId, _startTime);


+                $"Received two different test definition names ('{existingDefinitionName}' and '{definitionName}') for test id '{testId}'.");
+        }
+
+        byte[] resultBytes = _renderer.RenderCompletedResult(result, executionId, _startTime);


+        bool addDefinition = !_definitionNames.TryGetValue(testId, out string? existingDefinitionName);
+        if (!addDefinition
+            && result.TrxTestDefinitionName is not null
+            && !string.Equals(existingDefinitionName, definitionName, StringComparison.Ordinal))


+        byte[] summaryAdditions = _renderer.RenderSummaryAdditions(completion);
+        EnsureCapacity("summary", summaryAdditions.Length, _summaryBytesRemaining);


+            ValidateOutput(actual, expected, i, errors);
+            ValidateRelativeOrder(actual, ["Output", "ResultFiles"], $"Result[{i}] child order", errors);


+            Assert.Contains(
+                observation => observation.Dimension == TrxPhase3EvidenceDimension.ByteCutFault
+                        && observation.WriteRole == role
+                        && observation.Cut == observation.RequestedBytes,


+        Assert.Contains(
+            observation => observation.Scenario == "startup-absent"
+                && observation.Dimension == TrxPhase3EvidenceDimension.ByteCutFault
+                && observation.Cut == observation.RequestedBytes


+            report.FaultObservations);
+        Assert.Contains(
+            observation => observation.WriteRole == TrxPhase3WriteRole.ResultTail
+                && observation.Cut == observation.RequestedBytes


+        EnsureDeleteShareIsCompatible(source, temporaryPath);
+
+        long preLength = _files.TryGetValue(destinationPath, out VirtualFileEntry? destination)
+            ? destination.Bytes.LongLength


+        foreach (TrxFileOperationKind required in new[]
+                 {
+                     TrxFileOperationKind.Seek,
+                     TrxFileOperationKind.Write,
+                     TrxFileOperationKind.Flush,
+                 })
+        {
+            if (baseline.Any(operation => operation.Kind == required))
+            {
+                Assert.AreEqual(
+                    baseline.Count(operation => operation.Kind == required),
+                    operationCases.Count(observation => observation.OperationKind == required),
+                    $"{scenario}: {required}");
+            }
+        }


+                _temporaryDirectory = System.IO.Path.Combine(
+                    System.IO.Path.GetTempPath(),
+                    $"trx-incremental-prototype-{Guid.NewGuid():N}");


+                    $"trx-incremental-prototype-{Guid.NewGuid():N}");
+                Directory.CreateDirectory(_temporaryDirectory);
+                operations = new TrxPrototypeFileOperations();
+                Path = System.IO.Path.Combine(_temporaryDirectory, "results.trx");


+        {
+            string uniqueName = Guid.NewGuid().ToString("N");
+            Path = subdirectoryName is null
+                ? System.IO.Path.Combine(System.IO.Path.GetTempPath(), uniqueName)


+            string uniqueName = Guid.NewGuid().ToString("N");
+            Path = subdirectoryName is null
+                ? System.IO.Path.Combine(System.IO.Path.GetTempPath(), uniqueName)
+                : System.IO.Path.Combine(System.IO.Path.GetTempPath(), subdirectoryName, uniqueName);


+            catch
+            {
+            }


nohwnd · 2026-07-21T10:03:28Z

Closing the investigation after recording the findings and recommendation in #10016. The prototype disproved the padded writer guarantee and supports journal plus atomic snapshots instead. This branch remains available as evidence, but the prototype is intentionally not wired into production and is not intended for merge.

🤖

nohwnd and others added 6 commits July 20, 2026 22:01

Merge remote-tracking branch 'origin/main' into nohwnd-trx-crash-prot…

865da8b

…otype

Fix TRX prototype file encoding

8088c26

Co-authored-by: Copilot App <223556219+Copilot@users.noreply.github.com> 🤖

Merge origin/main into TRX crash prototype

4108f84

Co-authored-by: Copilot App <223556219+Copilot@users.noreply.github.com> 🤖

Fix TRX prototype evidence

4877d68

Co-authored-by: Copilot App <223556219+Copilot@users.noreply.github.com> 🤖

Merge origin/main into TRX evidence fix

83b7a64

Co-authored-by: Copilot App <223556219+Copilot@users.noreply.github.com> 🤖

Copilot AI review requested due to automatic review settings July 21, 2026 07:59

Copilot started reviewing on behalf of nohwnd July 21, 2026 08:00 View session

Copilot AI reviewed Jul 21, 2026

View reviewed changes

github-code-quality Bot found potential problems Jul 21, 2026

View reviewed changes

nohwnd mentioned this pull request Jul 21, 2026

Crash-resilient TRX: an always-loadable padded incremental writer #10016

Closed

nohwnd closed this Jul 21, 2026

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Uh oh!

Prototype crash-resilient TRX writes#10106

Prototype crash-resilient TRX writes#10106
nohwnd wants to merge 6 commits into
mainfrom
nohwnd-trx-crash-prototype

nohwnd commented Jul 21, 2026

Uh oh!

Copilot AI left a comment

Uh oh!

nohwnd commented Jul 21, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

		byte[] summaryAdditions = _renderer.RenderSummaryAdditions(completion);
		EnsureCapacity("summary", summaryAdditions.Length, _summaryBytesRemaining);

		ValidateOutput(actual, expected, i, errors);
		ValidateRelativeOrder(actual, ["Output", "ResultFiles"], $"Result[{i}] child order", errors);

Uh oh!

Conversation

nohwnd commented Jul 21, 2026

Changes

Verification

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Pull request overview

Reviewed changes

Uh oh!

nohwnd commented Jul 21, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants