NVIDIA / NVIDIA/cuvs

[BUG] Iterative CAGRA graph build fails while benchmarking with override_kv=dataset_memory_type:"mmap"

Open
#1,637 0 comments 0 reactions 0 assignees View on GitHub

Nobody has claimed this yet.

bug
Dominant language
Cuda
Stars
854
Forks
236
Avg merge
3d 3h
Merged PRs (30d)
62

Description

Iterative CAGRA graph build fails with an exception
CUDA call='cudaFreeAsync(ptr, stream)' at file=/workspace/repos/cuvs/cpp/src/neighbors/detail/cagra/compute_distance.hpp line=248 failed with misaligned address

override_kv=dataset_memory_type:"device" and override_kv=dataset_memory_type:"host" work

** To reproduce **
Run

CUVS_CAGRA_ANN_BENCH --build --force --data_prefix=/workspace/datasets/ --benchmark_out_format=csv --benchmark_out=res_build.csv --benchmark_counters_tabular=true --override_kv=dataset_memory_type:"mmap" /workspace/repos/vdb_bench/laion_1M_cagra_iterative.json

** laion_1M_cagra_iterative.json **

{
  "dataset": {
    "name": "laion_1M",
    "base_file": "laion_1M/base.1M.fbin",
    "subset_size": 1000000,
    "query_file": "laion_1M/queries.fbin",
    "groundtruth_neighbors_file": "laion_1M/groundtruth.1M.neighbors.ibin",
    "distance": "euclidean"
  },
  "search_basic_param": {
    "batch_size": 10000,
    "k": 10
  },
  "index": [
  
    {
      "name": "cuvs_cagra_iterative",
      "algo": "cuvs_cagra",
      "build_param": { 
        "graph_degree": 64,
        "intermediate_graph_degree": 128,
        "search_width": 1
      },
      "file": "laion_1M/cagra/q_coarse.ibin",
      "search_params": [
        {"itopk": 64, "search_width": 2, "max_iterations": 8},
        {"itopk": 128, "search_width": 1, "max_iterations": 12},
        {"itopk": 128, "search_width": 1, "max_iterations": 16},
        {"itopk": 256, "search_width": 1, "max_iterations": 16},
        {"itopk": 512, "search_width": 2, "max_iterations": 10},
        {"itopk": 256, "search_width": 2, "max_iterations": 12},
        {"itopk": 32, "search_width": 1, "max_iterations": 32},
        {"itopk": 32, "search_width": 1, "max_iterations": 64},
        {"itopk": 192, "search_width": 4, "max_iterations": 12},
        {"itopk": 256, "search_width": 4, "max_iterations": 12},
        {"itopk": 256, "search_width": 16, "max_iterations": 10},
        {"itopk": 512, "search_width": 16, "max_iterations": 32}
      ]
    }
  ]
}

Contributor guide

Open the contributing guide

First steps

  1. Read the whole issue, then the project's contributing guide.
  2. Comment on the issue to say you are picking it up — it saves two people doing the same work.
  3. Fork the repository and make your change on a branch.
  4. Open a pull request that references the issue number.

Research direction

Reproduce the failure with CUVS_CAGRA_ANN_BENCH and laion_1M_cagra_iterative.json using dataset_memory_type:"mmap". Inspect cpp/src/neighbors/detail/cagra/compute_distance.hpp around line 248 and compare the mmap path with the working device and host paths; done means iterative graph building completes without the misaligned-address exception.

Written by the indexing model from the issue text.

Assessment

Tech stack
cpp
Domain
machine-learning
Issue type
Bug
Difficulty
4/5
Estimated time
3-5 days
Activity status
Stale
Clarity
Mostly clear
Newbie friendliness
35/100

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.