flake: TestEditFiles/RelativePath

[blink-so]

Appears to affect all MCP file tools that have a relative path test. TestReadFile/RelativePath and TestEditFiles/RelativePath seen so far. Also seems to affect other subtests (TestEditFiles/NoPermissions, TestEditFiles/NoEdits), so it seems like some generic issue with making requests to the agent?

CI Run

GitHub Actions Log

Error Analysis

=== FAIL: agent TestEditFiles/RelativePath (25.00s)
    coderdtest.go:1592:
        	Error Trace:	C:/actions-runner/coder/coder/coderd/coderdtest/coderdtest.go:1592
        						C:/actions-runner/coder/coder/agent/files_test.go:707
        	Error:      	Should be true
        	Test:       	TestEditFiles/RelativePath
        	Messages:   	should be SDK error, got %!w(*xerrors.wrapError=&{do request 0xc0073c9140 {[140702362032838 140702377576688 140702395248547]}})
--- FAIL: TestEditFiles/RelativePath (25.00s)
FAIL	github.com/coder/coder/v2/agent	27.833s

Root Cause Assessment

• Test Function: TestEditFiles in agent/files_test.go:390-722
• Failing subtest: RelativePath (line 461)
• Error location: Line 707 calling coderdtest.SDKError(t, err)
• Platform: Windows (test-go-pg windows-2022)
• Pattern: Same coderdtest.go:1592 error as seen in issue flake: TestReadFile/RelativePath #982 (TestReadFile/RelativePath)
• No data races or panics detected

Assignment Analysis

Git blame results: The entire TestEditFiles function was authored by Asher (code-asher) on 2025-09-12 in commit 30330abaea.

Commit Info

• SHA: 0601cc8fa623f0dd4e44274f08271cc748ff26d9
• Author: Asher (code-asher)
• Related Issues: Similar to flake: TestReadFile/RelativePath #982 with same error pattern, suggesting systematic issue with SDK error handling in Windows CI

Reproduction Steps

This appears to be a Windows-specific timing or SDK error handling issue. The test expects an SDK error but receives a different error wrapper type.

[ethanndickson]

https://github.com/coder/coder/actions/runs/17901722952/job/50895903109?pr=19885 Same issue in diff subtest

[johnstcn]

https://github.com/coder/coder/actions/runs/18018792590/job/51270573918?pr=19969 same issue in different subtest

[johnstcn]

#1019 also looks like same root cause

[code-asher]

oof why is this error so bad

[ethanndickson]

Seen again: https://github.com/coder/coder/actions/runs/18773049043/job/53561391107?pr=20384

[flake-investigator]

New occurrence: TestReadFile/RelativePath failed with the same SDK error mismatch pattern.

CI run: https://github.com/coder/coder/actions/runs/19284307744 (job: test-go-pg-17)
Commit: e49c917bb0a885b1dc3b3f441573615b10b9f7e8 by Ethan (39577870+ethanndickson@users.noreply.github.com)
Timing: Job completed at 2025-11-12T02:35:57Z, Slack alert at ~02:39Z (same run).

Failure evidence:

=== FAIL: agent TestReadFile/RelativePath (25.00s)
    coderdtest.go:1607:
        	Error Trace:	/home/runner/work/coder/coder/coderd/coderdtest/coderdtest.go:1607
        	            				/home/runner/work/coder/coder/agent/files_test.go:266
        	Error:      	Should be true
        	Test:       	TestReadFile/RelativePath
        	Messages:   	should be SDK error, got do request: Get "http://[fd7a:115c:a1e0:4dc0:a115:ccfa:4689:c266]:4/api/v0/read-file?path=also-relative&offset=0&limit=0": context deadline exceeded

Classification: Flaky test/request routing to agent (same class as this issue). No data race or panic signatures found; other matrix jobs succeeded.

Ownership note: TestReadFile in agent/files_test.go was introduced/last substantively modified in 4bf63b4068a157cb (2025-09-09) by Asher (code-asher). Current assignment aligns with that.

Linking for visibility; no duplicate created.

[code-asher]

Added agent API request debug logs and recovery middleware in case that is causing the timeout. Closing until it comes up again.