Skip to content

feat(refine): support configurable flat storage types - #677

Open
luoxiaojian wants to merge 1 commit into
alibaba:mainfrom
luoxiaojian:codex/refine-flat-dtype
Open

feat(refine): support configurable flat storage types#677
luoxiaojian wants to merge 1 commit into
alibaba:mainfrom
luoxiaojian:codex/refine-flat-dtype

Conversation

@luoxiaojian

Copy link
Copy Markdown
Collaborator

No description provided.

@luoxiaojian
luoxiaojian force-pushed the codex/refine-flat-dtype branch 3 times, most recently from e0770ff to 6327e92 Compare August 14, 2026 11:28
Allow HNSW and Vamana refine Flat indexes to store FP32, FP16, or UINT8 vectors, with raw distance and conversion kernel dispatch.

Add an optional hugepage-backed contiguous Flat entity aligned with graph entity dispatch, while preserving mmap fallback for writes and buffer-pool storage.
@luoxiaojian
luoxiaojian force-pushed the codex/refine-flat-dtype branch from 6327e92 to bc33654 Compare August 14, 2026 12:57
int FlatContiguousStreamerEntity::add_vector_with_id(uint32_t id,
const void *query,
uint32_t element_size) {
degrade_to_mmap();

Copy link
Copy Markdown
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

这里需不需要考虑add和search的并发问题?比如正常用连续内存search的时候来了个add,清空了连续内存,search就会访问到空指针。

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants