PlushZ

ZRAM to prevent OOM kills

When training ML models, you can observe that Linux out-of-memory (OOM) killer terminates the process in the middle of training, even on high-RAM machines (e.g., 64 GiB RAM). This can be caused by the system running out of RAM.

To solve this, consider enabling ZRAM, a Linux kernel feature that provides compressed swap space in RAM, giving fast, efficient memory overflow handling without relying on slow disk-based swap.

Usage

Install ZRAM tools

natefoo

Interactive Tools setup notes

1. Can't start uWSGI proxy before proxy SQLite map exists
2. interactivetools_map needs to go in the galaxy section for the Galaxy side
3. Docker has to be installed on the Galaxy server even though it won't run there (ok)
4. Nodes can't have the job dir root squashed
5. Set <param id="docker_set_user"></param> on destination to run container as root
6. Not implemented for DRMAA runner I added the call to get the ports, but container stopping doesn't work. I consider this a container resolver issue and not a runner issue and am using an Epilog script to deal with it for the moment.
7. Wildcard certs are only valid for a single level, so you need a cert for *.interactivetoolentrypoint.interactivetool.example.org where example.org is your Galaxy server.
8. uWSGI as a proxy seems to have the same problems I encountered when originally trying to set it up as a proxy (enabling offload-threads causes connnections to fail after 3 connections are made, disabling `offlo

TorHou

<?xml version="1.0"?>
<!--
    If the value of 'URL_method' is 'get', the request will consist of the value of 'URL' coming back in
    the initial response.  If value of 'URL_method' is 'post', any additional params coming back in the
    initial response ( in addition to 'URL' ) will be encoded and appended to URL and a post will be performed.
-->
<tool name="Cherry Py Synchronous" id="cherrypy" tool_type="data_source">
    <description>test</description>
    <command interpreter="python">data_source.py $output $__app__.config.output_size_limit</command>
    <inputs action="http://localhost:8090/getdata" check_values="false" method="post">

hexylena

Shell script to make producing tool_dependencies easier. It's simple to use, behaves a lot like a venv, and then produces a tool_dependencies.xml in the directory where you started when you deactivate at the end.

Pretty to activate:

. activate.sh

Example of use:

chriddyp