refactor(diskann): decouple from libaio via dlopen by richyreachy · Pull Request #532 · alibaba/zvec

richyreachy · 2026-06-26T09:49:17Z

Currently, DiskAnn ships as a separate runtime-loaded .so (libzvec_diskann_plugin.so) file which requires a hard build-time dependency on libaio-dev.

This PR improves with a simpler, more robust design. Probing libaio.so file, a thread-safe singleton dlopens it and caches the syscall pointers. If libaio is absent, it will fall back to degrade to synchronous pread() with a warning.

The following shows the performance comparison on Cohere 1M between pread w/ and w/o AIO enabled. (Aliyun g9i, PL0, 10000 IOPS)

Pread w/ AIO Enabled

Result	FP32-100	FP32-300	FP32-500	FP16-100	FP16-300	FP16-500
Search Mem（4 thread）	503M	503M	503M	486M	486M	486M
Recall@1	96.1%	98.9%	99%	95.3%	97.8%	98.4%
Recall@10	96.14%	98.6%	98.94%	95.86%	98.1%	98.63%
Recall@50	93.394%	97.782%	98.598%	93.452%	97.354%	98.188%
QPS（1 thread）	32.6	11.2	6.7	32.3	11.1	6.7
QPS（2 thread）	65.2	22.3	13.5	64.5	22.4	13.4
QPS（4 thread）	106.4	36.6	22.1	106.4	36.6	22.1

Pread w/o AIO Enabled

Result	FP32-100	FP32-300	FP32-500	FP16-100	FP16-300	FP16-500
Search Mem（4 thread）	503M	503M	503M	486M	486M	486M
Recall@1	96.1%	98.9%	99%	95.3%	97.8%	98.4%
Recall@10	96.14%	98.6%	98.94%	95.86%	98.1%	98.63%
Recall@50	93.394%	97.782%	98.598%	93.452%	97.354%	98.188%
QPS（1 thread）	18.4	6.3	3.8	18.3	6.3	3.8
QPS（2 thread）	36.8	12.6	7.5	36.5	12.6	7.6
QPS（4 thread）	73.1	25.2	15.2	72.5	25.1	15.2

… into refactor/diskann_dlopen

iaojnh · 2026-07-01T11:51:19Z

可以把workflow里，打包相关的libaio-devel依赖，在这个PR里移除

feihongxu0824 · 2026-07-02T03:48:51Z

+        # so these exports are no longer needed for runtime plugin loading,
+        # but they are still required for the C API and for C++ consumers
+        # that link against the shared module.
        extern "C++" {


我建议删掉除了判断diskann with aio available的接口外，任何diskann plugin的接口。

站在用户侧，只需要关注当前runtime是否支持libaio的diskann就够了。

已经清理

feihongxu0824 · 2026-07-02T08:52:09Z

-// on these symbols; they may change or be removed in future releases. On
-// hosts missing libaio the bring-up fails cleanly and only DiskAnn becomes
-// unavailable — other index types (HNSW / IVF / Flat / Vamana) keep working.
+// on these symbols; they may change or be removed in future releases.


这个头文件还有必要留着吗？可能只需要放一个static标记位就可以了

diskann plugin命名歧义，已经去掉了改成diskann runtime，并只保留了load aio检测判断的必要代码。

feihongxu0824 · 2026-07-02T08:55:52Z

          sudo apt-get install -y clang libomp-dev
        shell: bash

-      - name: Install AIO 


ci移除了aio，如何验证w aio的情况？这里w aio和w/o aio的情况都要测

add a job to test w/ libaio, since main job is test on default env

feihongxu0824 · 2026-07-02T08:59:16Z

  transform(log_level.begin(), log_level.end(), log_level.begin(), ::tolower);
  if (LOG_LEVEL.find(log_level) != LOG_LEVEL.end()) {
    IndexLoggerBroker::SetLevel(LOG_LEVEL[log_level]);
+    zvec::ailego::LoggerBroker::SetLevel(LOG_LEVEL[log_level]);


这行是干啥的？

去掉了，之前的没有生效，改成了新的log库

feihongxu0824 · 2026-07-02T08:59:24Z

  transform(log_level.begin(), log_level.end(), log_level.begin(), ::tolower);
  if (LOG_LEVEL.find(log_level) != LOG_LEVEL.end()) {
    IndexLoggerBroker::SetLevel(LOG_LEVEL[log_level]);
+    zvec::ailego::LoggerBroker::SetLevel(LOG_LEVEL[log_level]);


这行是干啥的？

去掉了，之前的没有生效，改成了新的log库

feihongxu0824 · 2026-07-02T09:00:01Z


 set(CONFIGURE_ENV_LIST "")
 if(USE_OSS_MIRROR)
+        set(_OSS_BASE "https://zvec-bj.oss-cn-beijing.aliyuncs.com/thirdparty")


… into refactor/diskann_dlopen

feihongxu0824 · 2026-07-06T03:01:25Z

+            --no-build-isolation \
+            --config-settings='cmake.define.BUILD_TOOLS=ON' \
+            --config-settings='cmake.define.ENABLE_WERROR=ON' \
+            --config-settings='cmake.define.CMAKE_C_COMPILER_LAUNCHER=ccache' \


为啥要新开一个yml文件，理论上编译过程和03-macos-linux-build完全一样吧？没必要重复编译吧

放在一个里面，跑两次ut？第二次之前安装libaio库吧？

feihongxu0824 · 2026-07-06T03:02:18Z

          sudo apt-get update
          sudo apt-get install -y --no-install-recommends \
-            lcov libaio-dev
+            lcov


coverage可以不用去掉吧？不然也测不到diskann代码的覆盖度了？

feihongxu0824 · 2026-07-06T03:05:45Z

-# libaio-devel lives in BaseOS; disable EPEL so a flaky EPEL mirror cannot
-# break metadata refresh.
-before-all = "dnf install -y --disablerepo=epel libaio-devel"
+# libaio is loaded at runtime via dlopen (no build-time dependency).


这句注释也去掉好了，没有上下文看到这句话也挺让人疑惑的，或者把注释写完整一些也行

feihongxu0824 · 2026-07-06T03:33:56Z

+namespace zvec {
+
+// Return codes for InitDiskAnnRuntime().
+enum DiskAnnRuntimeStatus {


这里的接口如何考虑后续给diskann引入io uring?

已经把runtime去掉。改成在diskannxxx几个类初始化的时候，进行判断。

feihongxu0824 · 2026-07-06T03:38:52Z

+//
+// The ``path`` parameter is retained for API compatibility but is ignored.
+int InitDiskAnnRuntime(const std::string &path = "");
+


这里要放在core层的interface吗？作为db层需要用到这个吗？

c api是不是没调整？

这个是在构建索引的时候给用户提供warning信息。去掉了，改成在diskannxxx几个类初始化的时候，进行判断。

richyreachy added 6 commits June 24, 2026 21:19

refactor: use dlopen

74da77a

refactor: use dlopen to load libaio

17a7fd5

fix: log out the warning only once

8ef5d2d

fix: add log

50c6c8d

fix: add log

ceee226

Merge branch 'refactor/diskann_dlopen' of github.com:richyreachy/zvec…

5e77ee3

… into refactor/diskann_dlopen

richyreachy requested review from Cuiyus, JalinWang, chinaux, feihongxu0824, iaojnh and zhourrr as code owners June 26, 2026 09:49

github-actions Bot assigned richyreachy Jun 26, 2026

richyreachy added 2 commits June 26, 2026 18:29

refactor: merge main

6d774ed

Merge branch 'main' into refactor/diskann_dlopen

e2a2ede

richyreachy requested a review from egolearner June 30, 2026 08:54

richyreachy added 2 commits July 1, 2026 10:53

fix: fix windows build

a2524eb

Merge branch 'refactor/diskann_dlopen' of github.com:richyreachy/zvec…

ac109ae

… into refactor/diskann_dlopen

fix: remove libaio dependency

6688e1a

JalinWang reviewed Jul 2, 2026

View reviewed changes

Comment thread python/tests/test_collection_diskann.py Outdated

feihongxu0824 reviewed Jul 2, 2026

View reviewed changes

richyreachy added 7 commits July 2, 2026 19:11

Merge branch 'main' into refactor/diskann_dlopen

760fb7a

fix: remove no use code

f2559b5

fix: fix makefile

d753080

Merge branch 'refactor/diskann_dlopen' of github.com:richyreachy/zvec…

bc7b843

… into refactor/diskann_dlopen

fix: fix ut

292322c

fix: remove export map

a0a9ce9

fix: clean code

69b1e23

fix: add ci job to test libaio

081aee3

feihongxu0824 reviewed Jul 6, 2026

View reviewed changes

richyreachy added 2 commits July 6, 2026 16:56

fix: fix on comment

ab8901a

fix: remove diskann runtime

63f5ff7

Uh oh!

Conversation

richyreachy commented Jun 26, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

iaojnh commented Jul 1, 2026

Uh oh!

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

4 participants

richyreachy commented Jun 26, 2026 •

edited

Loading