ann-benchmarks
ann-benchmarks copied to clipboard
erikbern
Unable to establish a connection with Milvus without relevant useful information.
Milvus has been installed via Docker, but I am unable to connect to it using ann-benchmark, and there is no useful error message provided.
Milvus is benchmarked using a standalone, non-client/server version (https://github.com/zilliztech/Knowhere). You would have to write a wrapper yourself if you want to test their official release.
Sorry, this was outdated. It seems that @cococo2000 added a lot of new functionality to milvus, and it connects to the db. It should be run using --algorithm milvus as far as I can see.
The names of the latest algorithms in milvus can be found in the ann_benchmarks/algorithms/milvus/config.yml file, including milvus-flat, milvus-ivfflat, milvus-hnsw, milvus-scann, milvus-ivfpq, and milvus-ivfsq8.
To run a command using one of these algorithms, you could use:
python run.py --algorithm milvus-flat --runs 1 --dataset mnist-784-euclidean
@Kinkisong
Thanks @cococo2000. Is there any reason to not have them tested in github actions? They don't seem to appear in
https://github.com/erikbern/ann-benchmarks/blob/main/.github/workflows/benchmarks.yml#L44-L49
Thanks @cococo2000. Is there any reason to not have them tested in github actions? They don't seem to appear in
https://github.com/erikbern/ann-benchmarks/blob/main/.github/workflows/benchmarks.yml#L44-L49
The Docker tag for all these algorithms is ann-benchmarks-milvus, which is why the library includes only the milvus parameter. Additionally, with --max-n-algorithms 3, the GitHub Action will run tests on three of these algorithms.
@cococo2000 I still cannot connect to milvus with new command. Is here any wrong with my command? Thanks for your help!
@cococo2000 I still cannot connect to milvus with new command. Is here any wrong with my command? Thanks for your help!
It seems that the Docker image ann-benchmarks-milvus was not built using python install.py ..... before running the benchmark. Running the following command should resolve the issue:
python install.py --algorithm milvus
@GreateFang Let me know if you need any further adjustments!
@cococo2000 I still cannot connect to milvus with new command. Is here any wrong with my command? Thanks for your help!
It seems that the Docker image ann-benchmarks-milvus was not built using
python install.py .....before running the benchmark. Running the following command should resolve the issue:python install.py --algorithm milvus@GreateFang Let me know if you need any further adjustments!
I sorry that I have keep tring for a long time to install milvus but I fail. Could you pls provide a docker image that generated by ann to let me test it? thanks a lot @cococo2000 so depressed, working offline limitation make this so hardTT hopefully could get your help! My Email is [email protected]
@GreateFang I have sent the image to you via email, I hope it can help you.
@GreateFang I have sent the image to you via email, I hope it can help you.
Thank you sooooooooooooooooo much!
Sorry for keeping bothering, Thanks a lot for your patiently reply.
Hi coco @cococo2000 ,I have used your image and it works!Thanks a lot again! Could you help to check If this output is normal? It have stopped outputting for a while. should I just wait for result? And how long will it usually take( I use angular-25 dataset)? Sorry for keeping bothering, Thanks a lot for your patiently reply.
@GreateFang It appears that something is not functioning correctly. The program is unable to connect to the Milvus server properly, likely due to a pthread_create failed: Operation not permitted error. I did not encounter this issue when testing on Ubuntu, so I suspect the problem might be related to using the CentOS system. Additionally, for the Operation not permitted error, you might want to try running the program as root.
I have solved this problem with upgrading docker version and docker-compose version. Now it works perfectly, thank you for your help! @cococo2000