status_updates.json

{
  "entries": [
    {
      "date": "2026-04-01",
      "severity": "info",
      "title": "Windows debug build fixed (proxy DLL link race)",
      "body": "PR #10684 (remove slang-binaries) exposed a latent link race between slang-proxy (slang.dll) and slang-dispatcher (slang.exe) — both have OUTPUT_NAME slang, causing .ilk and .pdb collisions when Ninja parallelizes the link steps. Fixed by PR #10698 (/INCREMENTAL:NO) and PR #10702 (PDB_NAME slang-proxy). Sccache populate workflow should also recover.",
      "author": "jkiviluoto-nv",
      "visible": false
    },
    {
      "date": "2026-03-25",
      "severity": "info",
      "title": "Linux GPU persistence mode enabled",
      "body": "Linux CI runner VMs now enable GPU persistence mode (nvidia-smi -pm 1) at boot. This prevents NVML state corruption that caused intermittent 'Failed to initialize NVML: Unknown Error' during multi-server GPU test runs in Docker containers. Three other CI crash sources were also fixed: Mesa/LLVM removal from container (#10618, #10646), DeviceCache cleanup (#10595), and CountOfIntVal null guard (#10535). See #10596 for full investigation.",
      "author": "jkiviluoto-nv",
      "visible": false
    },
    {
      "date": "2026-03-19",
      "severity": "warning",
      "title": "Intermittent Linux GPU test failures",
      "body": "Linux GPU CI tests (debug and release) intermittently fail with test-server segfaults when running with parallel test-servers. Affects ~50% of merge queue runs since PR #9925 merged. The auto-retry mechanism handles most failures. Root cause is a race condition in parallel shader compilation — being investigated. See #10561 for details.",
      "author": "jkiviluoto-nv",
      "visible": false
    },
    {
      "date": "2026-03-18",
      "severity": "warning",
      "title": "SlangPy Windows CI: 5090 runners removed",
      "body": "RTX 5090 runners removed from SlangPy nvrgfx group due to a driver regression causing D3D12 raytrace and Vulkan shader cache test failures. Only affects recently updated 5090 drivers. Windows CI capacity is reduced until the driver issue is resolved. Being debugged.",
      "author": "jkiviluoto-nv",
      "visible": false
    },
    {
      "date": "2026-03-18",
      "severity": "warning",
      "title": "SlangPy cross-repo CI flaky on Windows",
      "body": "SlangPy ci-latest-slang workflow failing intermittently on Windows due to a driver issue on runner 2u1g-b650-0826. Affects D3D12 raytrace tests and Vulkan shader cache test. Runner has been removed from the nvrgfx group. Driver issue is being debugged.",
      "author": "jkiviluoto-nv",
      "visible": false
    },
    {
      "date": "2026-03-16",
      "severity": "info",
      "title": "SlangPy merge queue fix deployed",
      "body": "SlangPy Tests now run correctly in the merge queue. The trigger workflow was missing a merge_group event handler, causing PRs to be ejected after a 2-hour timeout. Fixed in slang#10534 and slangpy#875.",
      "author": "jkiviluoto-nv",
      "visible": false
    },
    {
      "date": "2026-03-13",
      "severity": "warning",
      "title": "Merge queue was blocked for ~16 hours due to SlangPy Tests rollout",
      "body": "Between Mar 12 15:13 UTC and Mar 13 ~10:30 UTC, no PRs could enter the merge queue. The 'SlangPy Tests' required status check was added with source set to 'GitHub Actions', but the status is posted via the commit status API (not a check run), so GitHub didn't match them. Resolved by changing branch protection to accept 'any source'. Additionally, PRs opened before the change had no status and were manually triggered.",
      "author": "jkiviluoto-nv",
      "visible": false
    },
    {
      "date": "2026-03-12",
      "severity": "info",
      "title": "SlangPy tests now run via cross-repo dispatch",
      "body": "SlangPy integration tests are now triggered automatically on every Slang PR via cross-repository dispatch to the SlangPy repository. Tests run against SlangPy tip-of-tree (Linux + Windows Release). The 'SlangPy Tests' commit status is a required check for merging. Debug and macOS coverage continues via nightly runs. Note: PRs that were open before this change need a new push or rebase to trigger the check. Maintainers can also trigger it manually via Actions → Trigger SlangPy CI.",
      "author": "jkiviluoto-nv",
      "visible": false
    },
    {
      "date": "2026-03-12",
      "severity": "info",
      "title": "Status page is now live",
      "body": "This is a test entry to verify the new CI status page.",
      "author": "jkiviluoto-nv",
      "visible": false
    }
  ]
}