feat: GPU HNSW by premal · Pull Request #51786 · milvus-io/milvus

premal · 2026-07-23T23:58:55Z

Summary

Adds the GPU_HNSW / GPU_HNSW_SQ index types to Milvus, backed by the faiss-native faiss::gpu::GpuIndexHNSW (through knowhere). A persisted HNSW segment is uploaded to the GPU at load time and searched on device — no GPU build path, the CPU-built graph is reused as-is. Supports fp32 / fp16 / bf16 / int8 storage across L2 / IP / COSINE.

Pairs with knowhere (GPU HNSW + accurate int8 cosine) and faiss (GpuIndexHNSW, facebookresearch/faiss#5459). Motivation: CPU HNSW plateaus at ~1,600 vec/s per node (DRAM-bandwidth bound) on our INT8_VECTOR dim-384 COSINE workload at ~500M+ rows; this gives the existing HNSW graph a GPU search path with recall parity to CPU.

What's in this PR

Index registration & param validation

internal/util/indexparamcheck/gpu_hnsw_checker.go (+ conf_adapter_mgr.go, index_type.go): GPU_HNSW / GPU_HNSW_SQ index-param validation.
client/index/common.go, pkg/util/indexparams/index_params.go: index-type plumbing.

Load-time GPU override

internal/querynodev2/segments/segment.go, services.go, internal/querycoordv2/task/executor.go: persisted type stays HNSW; a per-segment override_index_type swaps to the GPU index at load. The collection-level IndexInfoList is unchanged by design (a cluster can serve the same collection on CPU or GPU nodes).

GPU VRAM admission (the key resource change)

GPU segments must be admitted against device memory, not host RAM. This threads a GPU-memory estimate from knowhere through the C ABI into the Go loader:

internal/core/src/common/resource_c.h: add uint64_t gpu_memory_cost;
internal/core/src/index/IndexFactory.cpp: request.gpu_memory_cost = res.gpuMemoryCost;
internal/querynodev2/segments/segment_loader.go: GPU admission uses the GPU estimate with a host-RAM fallback, and reserves host memory separately:

gpuMemoryCost := estimateResult.GpuMemoryCost
if gpuMemoryCost == 0 {
    gpuMemoryCost = estimateResult.MaxMemoryCost   // fallback: conservative
}
fieldGpuMemorySize = append(fieldGpuMemorySize, gpuMemoryCost)

The reservation is conservative (≈2× file size) so admission never under-reserves; measured int8-cosine device usage is ≈1.7× file, so a node stops/rebalances before it can OOM.

knowhere pin

internal/core/thirdparty/knowhere/CMakeLists.txt: pins knowhere to the GPU-HNSW ref (vanilla faiss GpuIndexHNSW + accurate int8 cosine + faiss review fixes). Pinned to an immutable SHA rather than a branch. For an upstream merge this would be repointed to a tagged knowhere release once the knowhere PR lands (faiss → knowhere → milvus order).

Tests

internal/util/indexparamcheck/gpu_hnsw_checker_test.go, pkg/util/indexparams/index_params_test.go: param validation.
internal/querynodev2/segments/segment_loader_test.go: GPU admission sizing.
tests/python_client/testcases/indexes/{idx_gpu_hnsw,idx_gpu_hnsw_sq,test_gpu_hnsw,test_gpu_hnsw_sq}.py + test_gpu_hnsw_e2e.py: E2E build/load/search across dtypes and metrics.

Design notes / docs

Design doc: docs/design-docs/design_docs/20260619-gpu-hnsw.md (single consolidated MEP — architecture, native fp32/fp16/bf16/int8 storage, GPU VRAM admission, CPU-parity filtered search).
Load-time override keeps persisted segments portable between CPU and GPU query nodes.
GPU resource accounting is generic; only GPU_HNSW populates gpuMemoryCost under the current knowhere pin.

Hardware / build

NVIDIA GPU compute capability 7.0+ (Volta+), CUDA 12.x+; GPU Docker image (build/docker/milvus/gpu/ubuntu22.04/Dockerfile).

Related Issue

[Feature]: Support HNSW Search using GPUs #51733

Wire up GPU_HNSW / GPU_HNSW_SQ index types backed by the faiss-native faiss::gpu::GpuIndexHNSW (via knowhere), with load-time override of persisted HNSW segments to the GPU index, VRAM admission via Resource.maxMemoryCost, and a gpu_hnsw index-param checker. Pins knowhere to gpu-hnsw@97f933f3 (vanilla faiss GpuIndexHNSW + accurate int8 cosine + faiss review fixes). - IndexFactory: maxMemoryCost>0 drives GPU segment VRAM admission (only GPU_HNSW populates it under the current pin). - segment_loader: GPU segment admission sizing. - indexparamcheck: gpu_hnsw_checker (GPU_HNSW / GPU_HNSW_SQ). - Design docs: gpu-hnsw, ocq-removal (FP16/BF16 GPU-validated), filtered-search (refreshed source anchors). Python E2E + index test cases. Clean single-commit re-creation off current master (replaces the stacked gpu-hnsw-faiss-native branch). Signed-off-by: premal <premal@6sense.com> Co-Authored-By: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com>

Repin knowhere gpu-hnsw 97f933f3 -> 9c0af31f, which re-vendors the faiss #11 review fix: the native int8 DP4A layer-0 path is now selected on a per-search staged flag rather than a leftover pooled d_queries_i8 buffer, so a fp32-query search reusing a scratch slot can't score against stale int8 query data. Signed-off-by: premal <premal@6sense.com> Co-Authored-By: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com>

…-only today Comment-only. Addresses the Devin/Magnitude review note on IndexFactory's VectorMemIndexLoadResource: document that GPU_HNSW is currently the sole index type populating res.maxMemoryCost (> 0), so the peak-based admission branch is effectively GPU_HNSW-scoped today while remaining generic for any future index that sets maxMemoryCost. No behavior change; does not affect the v4 image (gpu-hnsw-faiss-native-v4, built from b9d094e). Signed-off-by: premal <premal@6sense.com> Co-Authored-By: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com>

…9d92eb1) Repoints KNOWHERE_VERSION to 99d92eb1, which re-vendors the faiss GpuIndexHNSW on-device label-conversion kernel (search() path). No behavior change for milvus: the querynode uses the searchHost/searchHostInt8 path, which was already free of the D2H->H2D label round-trip. Keeps the pin in sync with faiss gpu-hnsw dce385d9. Signed-off-by: premal <premal@6sense.com> Co-Authored-By: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com>

Sync gpu-hnsw with upstream milvus master (78 commits). Re-pin knowhere to 6si/knowhere 2959fe3d (gpu-hnsw merged up to knowhere main), which now carries the expected<T> iterator error-code interface (milvus-io#1699) that master's segcore adaptation depends on. Resolved segment_loader.go by adopting master's prepareIndexLoadParams() helper (our side only added a doc comment; the helper preserves DiskANN/bitmap + AppendPrepareLoadParams logic). Co-Authored-By: Devin <devin@cognition.ai> Signed-off-by: premal <premal@6sense.com> Co-Authored-By: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com>

…here 0d01cac0 Knowhere now reports a device (VRAM) footprint distinct from the host transient peak. Plumb it through: add gpu_memory_cost to the C LoadResourceRequest, set it from Resource::gpuMemoryCost in VectorIndexLoadResource, expose it as ResourceEstimate.GpuMemoryCost, and use it for the querynode GPU admission reservation (fieldGpuMemorySize), falling back to MaxMemoryCost when a GPU index does not report a distinct device cost. Without this, GPU admission reused the host transient MaxMemoryCost (raised to 4x file for int8-cosine to fix the host OOM), which would over-reserve VRAM (~2.3x the real ~1.7x/file device growth) and falsely reject GPU loads. Re-pin knowhere to 0d01cac0 (chunked cosine reconstruct + metric-aware host estimate + separate gpuMemoryCost + faiss_gpu_hnsw PIC build fix). Signed-off-by: premal <premal@6sense.com> Co-Authored-By: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com>

Fold the filtered-search MEP and the OCQ-removal/native-fp16-bf16 decision record into 20260619-gpu-hnsw.md and drop the two separate docs. Update the two python test docstrings that referenced the removed filtered-search doc. Signed-off-by: premal <premal@6sense.com> Co-Authored-By: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com>

sre-ci-robot · 2026-07-23T23:59:09Z

[APPROVALNOTIFIER] This PR is NOT APPROVED

This pull-request has been approved by: premal
To complete the pull request process, please assign xiaofan-luan after the PR has been reviewed.
You can assign the PR to them by writing /assign @xiaofan-luan in a comment when ready.

The full list of commands accepted by this bot can be found here.

Details

Needs approval from an approver in each of these files:

OWNERS

Approvers can indicate their approval by writing /approve in a comment
Approvers can cancel approval by writing /approve cancel in a comment

mergify · 2026-07-23T23:59:35Z

@premal

Invalid PR Title Format Detected

Your PR submission does not adhere to our required standards. To ensure clarity and consistency, please meet the following criteria:

Title Format: The PR title must begin with one of these prefixes:

feat: for introducing a new feature.
fix: for bug fixes.
enhance: for improvements to existing functionality.
test: for add tests to existing functionality.
doc: for modifying documentation.
auto: for the pull request from bot.
build(deps): for dependency updates from Dependabot.

Description Requirement: The PR must include a non-empty description, detailing the changes and their impact.

Required Title Structure:

[Type]: [Description of the PR]

Where Type is one of feat, fix, enhance, test or doc.

Example:

enhance: improve search performance significantly

Please review and update your PR to comply with these guidelines.

sre-ci-robot · 2026-07-24T00:00:08Z

[ci-v2-notice]
Notice: New ci-v2 system is enabled for this PR.

To rerun ci-v2 checks, comment with:

/ci-rerun-code-check // for ci-v2/code-check
/ci-rerun-code-check-macos // for Code Checker MacOS (GitHub Actions)
/ci-rerun-build // for ci-v2/build
/ci-rerun-build-all // for ci-v2/build-all (multi-arch builds)
/ci-rerun-buildenv // for ci-v2/build-env (build milvus-env builder images; update .env after the new tag is ready)
/ci-rerun-ut-integration // for ci-v2/ut-integration, will rerun ci-v2/build
/ci-rerun-ut-go // for ci-v2/ut-go, will rerun ci-v2/build
/ci-rerun-ut-cpp // for ci-v2/ut-cpp
/ci-rerun-ut // for all ci-v2/ut-integration, ci-v2/ut-go, ci-v2/ut-cpp, will rerun ci-v2/build
/ci-rerun-e2e-default // for ci-v2/e2e-default
/ci-rerun-e2e-amd // for ci-v2/e2e-amd (e2e pool dispatcher)
/ci-rerun-e2e-dist-wp // for ci-v2/e2e-dist-wp (Tencent distributed woodpecker-service boundary)
/ci-rerun-build-ut-cov // for ci-v2/build-ut-cov (build + unit tests in one pipeline)
/ci-rerun-gosdk // for ci-v2/go-sdk (Go SDK E2E tests, ARM)
/ci-rerun-gosdk-std-wp // for ci-v2/go-sdk-std-wp (Go SDK E2E, standalone + embedded Woodpecker)
/ci-rerun-gosdk-dist-wp // for ci-v2/go-sdk-dist-wp (Go SDK E2E, distributed + Woodpecker service)

If you have any questions or requests, please contact @zhikunyao.

sre-ci-robot · 2026-07-24T01:33:59Z

❌ CI Loop Results `cfb2fe7`

Stage	Result	Duration	Tests
✅ Build	SUCCESS	16.5min	-
❌ Code-Check	FAILURE	3.9min	-
❌ UT-Integration	SKIPPED	-	-
❌ UT-GO	SKIPPED	-	-
✅ UT-CPP-Cov	SUCCESS	66.7min	8243 total, 8243 passed, 0 failed

Total: 89min | Pipeline | Artifacts
Diff Coverage: CPP 100.0% (2 hit, 0 miss, 2 measurable lines, 18 unmeasured)
Total Patch Coverage: 100.0% (2/2 measurable lines, 18 unmeasured)

Failed Test Logs:

Code-Check: view log

Signed-off-by: premal <premal@6sense.com> Co-Authored-By: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com>

mergify · 2026-07-24T07:07:55Z

@premal Please associate the related issue to the body of your Pull Request. (eg. "issue: #")

…rmat); revert unrelated common_func import tweak Signed-off-by: premal <premal@6sense.com> Co-Authored-By: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com>

sre-ci-robot · 2026-07-24T09:56:27Z

✅ CI Loop Results `4493ff9`

Stage	Result	Duration	Tests
✅ Build	SUCCESS	11.2min	-
✅ Code-Check	SUCCESS	8.2min	-
✅ UT-Integration	SUCCESS	24.3min	-
✅ UT-GO	SUCCESS	22.6min	-
✅ UT-CPP-Cov	SUCCESS	46.1min	8243 total, 8243 passed, 0 failed

Total: 96min | Pipeline | Artifacts

Overall Coverage: 73.1%
Diff Coverage: CPP 100.0% (2 hit, 0 miss, 2 measurable lines, 18 unmeasured) | Go 45.5% (46 hit, 55 miss, 101 measurable lines, 115 unmeasured)
Diff Coverage HTML: view changed lines
Go Patch Warning: WARNING: Go patch coverage is partial; 115 changed lines were unmeasured.
Total Patch Coverage: 46.6% (48/103 measurable lines, 133 unmeasured)

sre-ci-robot · 2026-07-24T10:06:15Z

❌ CI Loop Results `6fd26d9`

Stage	Result	Duration	Tests
✅ Build	SUCCESS	16.1min	-
❌ Code-Check	FAILURE	4.5min	-
❌ UT-Integration	SKIPPED	-	-
❌ UT-GO	SKIPPED	-	-
✅ UT-CPP-Cov	SUCCESS	65.8min	8243 total, 8243 passed, 0 failed

Total: 104min | Pipeline | Artifacts
Diff Coverage: CPP 100.0% (2 hit, 0 miss, 2 measurable lines, 18 unmeasured)
Total Patch Coverage: 100.0% (2/2 measurable lines, 18 unmeasured)

Failed Test Logs:

Code-Check: view log

…fp16 re-encode) Adds the reconciliation Magnitude flagged as an open item: the int8 device layout (native ~1x vs fp16 re-encode ~1.7x) is decided at load by the deserialized storage class (HasInverseL2Norms), not the config metric. Explains why a querynode log showing upload_int8_dataset (1 B/dim, native) and an isolated cosine-subclass collection re-encoding to fp16 are both correct and consistent, and why the re-encode is not dead code. No code change. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> Signed-off-by: premal <premal@6sense.com> Co-Authored-By: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com>

sre-ci-robot · 2026-07-25T02:13:04Z

✅ CI Loop Results `2324c32`

Stage	Result	Duration	Tests
✅ Build	SUCCESS	10.7min	-
✅ Code-Check	SUCCESS	7.4min	-
✅ UT-Integration	SUCCESS	24.1min	-
✅ UT-GO	SUCCESS	22.6min	-
✅ UT-CPP-Cov	SUCCESS	45.7min	8243 total, 8243 passed, 0 failed

Total: 73min | Pipeline | Artifacts

Overall Coverage: 73.1%
Diff Coverage: CPP 100.0% (2 hit, 0 miss, 2 measurable lines, 18 unmeasured) | Go 45.5% (46 hit, 55 miss, 101 measurable lines, 115 unmeasured)
Diff Coverage HTML: view changed lines
Go Patch Warning: WARNING: Go patch coverage is partial; 115 changed lines were unmeasured.
Total Patch Coverage: 46.6% (48/103 measurable lines, 133 unmeasured)

…ive check) Live-cluster probe (v7) confirms mpd_v2 is IndexHNSWSQCosine (is_cosine=1 on all sampled segments, upload_int8_dataset absent) and takes the fp16 re-encode branch (~1.7x resident / 2.0x estimator peak), matching vram_probe_int8_cos. The native upload_int8_dataset (1 B/dim) logs come from non-cosine int8 collections, not mpd_v2. No code change. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> Signed-off-by: premal <premal@6sense.com> Co-Authored-By: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com>

sre-ci-robot · 2026-07-25T10:08:45Z

✅ CI Loop Results `459c812`

Stage	Result	Duration	Tests
✅ Build	SUCCESS	10.5min	-
✅ Code-Check	SUCCESS	6.9min	-
✅ UT-Integration	SUCCESS	23.9min	-
✅ UT-GO	SUCCESS	22.6min	-
✅ UT-CPP-Cov	SUCCESS	45.6min	8243 total, 8243 passed, 0 failed

Total: 73min | Pipeline | Artifacts

Overall Coverage: 73.1%
Diff Coverage: CPP 100.0% (2 hit, 0 miss, 2 measurable lines, 18 unmeasured) | Go 45.5% (46 hit, 55 miss, 101 measurable lines, 115 unmeasured)
Diff Coverage HTML: view changed lines
Go Patch Warning: WARNING: Go patch coverage is partial; 115 changed lines were unmeasured.
Total Patch Coverage: 46.6% (48/103 measurable lines, 133 unmeasured)

Consolidate the three duplicated load-stage override loops (QueryNode LoadSegments handler, segmentLoader.Load, segmentLoader.ReopenSegments) into one exported, unit-tested helper ApplyLoadStageOverrides, and extract the numRows fallback (resolveIndexNumRows) and GPU VRAM admission fallback (gpuAdmissionCost) into small pure helpers so the previously-uncovered GPU_HNSW glue is directly testable. Tests added: - querycoordv2/task: getLoadInfo applies (and, without config, skips) the HNSW -> GPU_HNSW load-stage override. - querynodev2/segments: ApplyLoadStageOverrides override/no-override/nil; resolveIndexNumRows fallback (NumRows==0/negative -> segment rows, the OOM-admission regression); gpuAdmissionCost device-vs-host fallback. No behavior change: helpers preserve the exact prior semantics. Signed-off-by: premal <premal@6sense.com> Co-Authored-By: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com>

sre-ci-robot · 2026-07-25T16:06:04Z

✅ CI Loop Results `2d7e280`

Stage	Result	Duration	Tests
✅ Build	SUCCESS	10.2min	-
✅ Code-Check	SUCCESS	6.0min	-
✅ UT-Integration	SUCCESS	23.5min	-
✅ UT-GO	SUCCESS	17.1min	-
✅ UT-CPP-Cov	SUCCESS	45.1min	8243 total, 8243 passed, 0 failed

Total: 65min | Pipeline | Artifacts

Overall Coverage: 73.1%
Diff Coverage: CPP 100.0% (2 hit, 0 miss, 2 measurable lines, 18 unmeasured) | Go 55.1% (54 hit, 44 miss, 98 measurable lines, 136 unmeasured)
Diff Coverage HTML: view changed lines
Go Patch Warning: WARNING: Go patch coverage is partial; 136 changed lines were unmeasured.
Total Patch Coverage: 56.0% (56/100 measurable lines, 154 unmeasured)

sre-ci-robot · 2026-08-13T18:14:53Z

[INFO] PR Label Summary by Default
[INFO] Dependent PR check skipped - not required for branch 'master'

[WARNING] Milestone not set

PR: feat: GPU HNSW #51786
Title: feat: GPU HNSW
Please set a milestone for better release tracking

You can set milestone by commenting:
/set-milestone
Example:
/set-milestone 2.5.0

Use /refresh-label to update related check and label manually

premal and others added 8 commits July 22, 2026 05:43

Merge branch 'master' into gpu-hnsw

ef02535

sre-ci-robot requested review from aoiasd and czs007 July 23, 2026 23:59

sre-ci-robot added area/compilation size/XXL Denotes a PR that changes 1000+ lines. area/test sig/testing labels Jul 23, 2026

mergify Bot added the dco-passed DCO check passed. label Jul 23, 2026

mergify Bot added the do-not-merge/invalid-pr-format label Jul 23, 2026

premal and others added 2 commits July 24, 2026 07:02

fix(gpu_hnsw): clang-format-15 resource_c.h gpu_memory_cost comment

6fd26d9

Signed-off-by: premal <premal@6sense.com> Co-Authored-By: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com>

fix(gpu_hnsw): gofmt -s comment alignment in gpu_hnsw_checker_test.go

0e86d3a

Signed-off-by: premal <premal@6sense.com> Co-Authored-By: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com>

premal changed the title ~~GPU HNSW~~ feat: GPU HNSW Jul 24, 2026

mergify Bot added kind/feature Issues related to feature request from users and removed do-not-merge/invalid-pr-format labels Jul 24, 2026

mergify Bot added the do-not-merge/missing-related-issue label Jul 24, 2026

test(gpu_hnsw): ruff-clean python tests (explicit DataType import, fo…

4493ff9

…rmat); revert unrelated common_func import tweak Signed-off-by: premal <premal@6sense.com> Co-Authored-By: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com>

mergify Bot removed the do-not-merge/missing-related-issue label Jul 24, 2026

sre-ci-robot added the do-not-merge/need-milestone generate by v2-label-manager label Aug 13, 2026

sre-ci-robot added the do-not-merge/doc-need-two-approve Gray observation: docs PR needs two distinct approve commenters label Aug 14, 2026

zhikunyao removed the do-not-merge/need-milestone generate by v2-label-manager label Aug 14, 2026

Navigation Menu

Uh oh!

feat: GPU HNSW - #51786

feat: GPU HNSW#51786
premal wants to merge 14 commits into
milvus-io:masterfrom
6si:gpu-hnsw

premal commented Jul 23, 2026 •

edited

Loading

Uh oh!

sre-ci-robot commented Jul 23, 2026

Uh oh!

mergify Bot commented Jul 23, 2026

Uh oh!

sre-ci-robot commented Jul 24, 2026

Uh oh!

sre-ci-robot commented Jul 24, 2026

Uh oh!

mergify Bot commented Jul 24, 2026

Uh oh!

sre-ci-robot commented Jul 24, 2026

Uh oh!

sre-ci-robot commented Jul 24, 2026

Uh oh!

sre-ci-robot commented Jul 25, 2026

Uh oh!

sre-ci-robot commented Jul 25, 2026

Uh oh!

sre-ci-robot commented Jul 25, 2026

Uh oh!

sre-ci-robot commented Aug 13, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

Uh oh!

Conversation

premal commented Jul 23, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Summary

What's in this PR

Index registration & param validation

Load-time GPU override

GPU VRAM admission (the key resource change)

knowhere pin

Tests

Design notes / docs

Hardware / build

Related Issue

Uh oh!

sre-ci-robot commented Jul 23, 2026

Uh oh!

mergify Bot commented Jul 23, 2026

Uh oh!

sre-ci-robot commented Jul 24, 2026

Uh oh!

sre-ci-robot commented Jul 24, 2026

❌ CI Loop Results cfb2fe7

Uh oh!

mergify Bot commented Jul 24, 2026

Uh oh!

sre-ci-robot commented Jul 24, 2026

✅ CI Loop Results 4493ff9

Uh oh!

sre-ci-robot commented Jul 24, 2026

❌ CI Loop Results 6fd26d9

Uh oh!

sre-ci-robot commented Jul 25, 2026

✅ CI Loop Results 2324c32

Uh oh!

sre-ci-robot commented Jul 25, 2026

✅ CI Loop Results 459c812

Uh oh!

sre-ci-robot commented Jul 25, 2026

✅ CI Loop Results 2d7e280

Uh oh!

sre-ci-robot commented Aug 13, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

premal commented Jul 23, 2026 •

edited

Loading

❌ CI Loop Results `cfb2fe7`

✅ CI Loop Results `4493ff9`

❌ CI Loop Results `6fd26d9`

✅ CI Loop Results `2324c32`

✅ CI Loop Results `459c812`

✅ CI Loop Results `2d7e280`