Local Foldseek searches are no longer working
JackKay404 opened this issue · comments
Recently cloned the repo again and re-built the PDB database, setting up for foldseek local server.
On docker compose up the server seems functional, however after submitting any job mmseqs-web-worker-1 errors out with the following:
mmseqs-web-worker-1 | Index version: fs1
mmseqs-web-worker-1 | Generated by: 7192be546fc4086c7176b86b630932b3994970b1
mmseqs-web-worker-1 | ScoreMatrix: 3di.out
mmseqs-web-worker-1 | Index version: fs1
mmseqs-web-worker-1 | Generated by: 7192be546fc4086c7176b86b630932b3994970b1
mmseqs-web-worker-1 | ScoreMatrix: 3di.out
mmseqs-web-worker-1 | Index version: fs1
mmseqs-web-worker-1 | Generated by: 7192be546fc4086c7176b86b630932b3994970b1
mmseqs-web-worker-1 | ScoreMatrix: 3di.out
mmseqs-web-worker-1 | Index version: fs1
mmseqs-web-worker-1 | Generated by: 7192be546fc4086c7176b86b630932b3994970b1
mmseqs-web-worker-1 | ScoreMatrix: 3di.out
mmseqs-web-worker-1 | [=================================================================Error: Convert Alignments died
mmseqs-web-worker-1 | Segmentation fault (core dumped)
mmseqs-web-worker-1 | 2024/04/29 16:02:53 Execution Error: Execution Error: exit status 1
Any clues on what changes could have caused this?
Thanks!
Which commit is this? I did a lot of refactoring today and might have broken something.
However, this does looks more like a foldseek issue and not server issue. What are you searching?
I think I currently have commit 00f89de checked out?
commit 00f89de8fa191bf3824d07aea24c86c80117f148 (HEAD -> master, origin/master, origin/HEAD)
Author: Milot Mirdita <milot@mirdita.de>
Date: Mon Apr 29 17:31:51 2024 +0900
Another try to fix chain names
I'm searching one of my structures (a small protein ~ 50 AA) against the PDB structure database as a test that my setup is functional. I've run this search many times with previous commits without any issues but I saw new updates on the foldseek server so re-cloned the repo to implement them locally
I tested the same sequence on the public server and this ran without error. Is the public server using your latest commits or a previous commit that I could roll back to?
Did you rebuilt the docker containers or are you using the containers downloaded from GHCR? Can you try to do a docker compose pull within the docker-compose directory to make sure you are also on the latest GHCR container?
(This is the exact setup that the server is using).
Apologies, I now think that this is relating to my setup. I'm running this inside a Multipass VM (Ubuntu 22.04) and observing the Segmentation fault (core dumped) error, however when I replicate in WSL2 the search runs perfectly. I will need to figure out why this happens in the VM but not WSL2... Any thoughts? I have allocated 12 Gb of memory to the VM and there is sufficient disk space
I have no idea about multipass, but I've seen issues with SIMD support and VMs before. Make sure you are using whatever "host" cpu mode multipass might offer.
It's likely that the VM claims to support AVX2 in the /proc/cpuinfo, however then doesn't correctly expose the instructions.