Expose remote cache read errors and fail on server disconnects

[seungpyoson]

Summary

• in the direct single-cache daemon path, surface non-NotFound remote read failures through the existing cache-error statistics path instead of counting them as ordinary misses
• make post-CompileStarted response I/O failures follow the documented default-fatal policy while preserving SCCACHE_IGNORE_SERVER_IO_ERROR=1

Deliberately unchanged

• startup rate-limit tolerance
• automatic read-only downgrade when a configured backend cannot write
• explicit read-only and public-cache configurations
• client-side IPC and multi-level cache fallback policies

Scope

One commit changes src/cache/cache.rs and src/commands.rs only. There are no documentation, protocol, cache-format, dependency, disk-cache, multi-level-cache, or client-side-storage changes. Existing documentation already describes the server-I/O policy.

Validation

Focused regression tests cover unexpected direct remote read errors, default-fatal post-start disconnects, and the existing opt-in local fallback. Upstream CI and maintainer review remain required before merge.

[codecov-commenter]

Codecov Report

❌ Patch coverage is 97.29730% with 1 line in your changes missing coverage. Please review.
✅ Project coverage is 73.21%. Comparing base (ffac4a5) to head (d048ff9).

Files with missing lines	Patch %	Lines
src/commands.rs	94.73%	1 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##             main    #2779      +/-   ##
==========================================
+ Coverage   70.91%   73.21%   +2.30%     
==========================================
  Files          72       72              
  Lines       37684    37695      +11     
==========================================
+ Hits        26723    27600     +877     
+ Misses      10961    10095     -866     

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[codspeed-hq]

Merging this PR will not alter performance

✅ 58 untouched benchmarks
⏩ 6 skipped benchmarks¹

---

_{Comparing seungpyoson:agent/fail-closed-cache-errors (d048ff9) with main (2a40bc3)²}

Footnotes

1. 6 benchmarks were skipped, so the baseline results were used instead. If they were deleted from the codebase, click here and archive them to remove them from the performance reports. ↩

2. No successful run was found on main (ffac4a5) during the generation of this report, so 2a40bc3 was used instead as the comparison base. There might be some changes unrelated to this pull request in this report. ↩

[seungpyoson]

Synced this PR with current main at 918b8ef. New CI, integration-test, and benchmark workflows are waiting for maintainer approval. Could a maintainer please approve the workflows and re-review the PR after they complete? The sync includes main's libc pin for the prior dist-test failure and allows CodSpeed to compare against the current runner generation.

[seungpyoson]

@sylvestre Could you take a look at sccache PR #2779 (#2779) when you have a chance? I synced it with current main today. The underlying bugs are still present, the patch merges cleanly, and the fork workflow runs are waiting for maintainer approval.

If the approach looks good, could you approve the workflows and merge it?

[seungpyoson]

Quick update: the sole non-successful check on d048ff9 was the CUDA 11.8 job, which was canceled after Ubuntu package mirror requests stalled during NVCC installation; it never reached tests. I pushed empty commit acfbdca to retrigger the workflows without source changes. The new CI, integration-test, and benchmark runs are waiting for maintainer approval.