From CommunityData
Jump to navigation Jump to search

Note Note: This page is intended to replace the main CommunityData:Hyak page in the near future. This is a part of our transition to the new Slurm-based job scheduler. Some of the sections may be incomplete, and the instructions may not work. Feel free to edit and fix the content that is incorrect/out-of-date.

To use Hyak, you must first have a UW NetID, access to Hyak, and a two factor authentication token. Details on getting set up with all three are available at CommunityData:Hyak setup.

There are a number of other sources of documentation:

Setting up SSH[edit]

When you connect to SSH, it will ask you for a key from your token. Typing this in every time you start a connection be a pain. One approach is to create an .ssh config file that will create a "tunnel" the first time you connect and send all subsequent connections to Hyak over that tunnel. Some details in the Hyak documentation.

I've added the following config to the file ~/.ssh/config on my laptop (you will want to change the username):

 Host hyak-mox
     User sdg1
     ControlPath ~/.ssh/master-%r@%h:%p
     ControlMaster auto
     ControlPersist yes
     Compression yes

Note Note: If your SSH connection becomes stale or disconnected (e.g., if you change networks) it may take some time for the connection to time out. Until that happens, any connections you make to hyak will silently hang. If your connections to ssh hyak are silently hanging but your Internet connection seems good, look for ssh processes running on your local machine with:

ps ax|grep hyak

If you find any, kill them with kill <PROCESSID>. Once that is done, you should have no problem connecting to Hyak.

X11 forwarding[edit]

You may also want to add these two lines to your Hyak .ssh/config:

ForwardX11 yes
ForwardX11Trusted yes

These lines will mean that if you have "checked out" an interactive machine, you can ssh from your computer to Hyak and then directly through an addition hop to the machine (like ssh n0652). Those ForwardX11 lines means if you graph things on this session, they will open on your local display.

Connecting to Hyak[edit]

To connect to Hyak, you now only need to do:

ssh hyak-mox

It will prompt you for your UWNetID's password. Once you type in your password, you will have to respond to a 2-factor authentication request.

Setting Up Hyak[edit]

For Mox, we have created a set of bash scripts which initialize a good data science environment.

We recommend that new users of hyak load this environment by adding

source /gscratch/comdata/env/cdsc_mox_bashrc

to the end of your ~/.bashrc file.

This does a number of useful things. It loads modern versions of R and Python and places Spark in your environment. It also provides a number of convenient commands for interacting with the SLURM HPC system for checking out nodes and monitoring jobs. Particularly important commands include


Which attempts to check out a supercomputing node.


Requests a node with 240GB of memory.


Checks out a build node which can access the internet and is intended to be used to install software.


Prints all the running jobs by people in the group.


Displays jobs by members of the group.

Read the files in /gscratch/comdata/env to see how these commands are created as well as other features not documented here.


We recently switched to using Anaconda to manage Python on Mox. Anaconda comes with the `conda` tool for managing python packages and versions. Multiple python environments can co-exist in a single Anaconda installation, this allows different projects to use different versions of Python or python packages, which can be useful for maintaining projects that use old versions.

By default, our shared setup loads a conda environment called `minimal_ds` that provides recent versions of python packages commonly used in data science workflows. This is probably a good setup for most use-cases, and allows everyone to use the same packages, but it can be even better to create different environments for each project. See the anaconda documentation for how to create an environment.

Moving files from ikt to mox[edit]

You can copy files at high speed without a password between the Hyak systems using commands like the ones below (instructions from the Hyak documentation).

From ikt to mox

   ikt1$ hyakbbcp myfile
   ikt1$ hyakbbcp -r mydirectory

From mox to ikt

   mox1$ hyakbbcp myfile
   mox1$ hyakbbcp -r mydirectory

SSH into compute nodes[edit]

The hyak wiki has instructions for how to enable ssh within hyak. Reproduced below:

You should be able to ssh from the login node to a compute node without giving a password. If it does not work then do below steps:

1) ssh-keygen

Press enter for each question. This will ensure default options.

2) cd .ssh

3) cat >> authorized_keys

Running Jobs on Hyak[edit]

When you first log in to Hyak, you will be on a "login node". These are nodes that have access to the Internet, and can be used to update code, move files around, etc. They should not be used for computationally intensive tasks. To actually run jobs, there are a few different options, described in detail in the Hyak User documentation. Following are basic instructions for some common use cases.

Interactive nodes[edit]

Interactive nodes are systems where you get a bash shell from which you can run your code. This mode of operation is conceptually similar to running your code on your own computer, the difference being that you have access to much more CPU and memory. To check out an interactive node, run the big_machine or any_machine command from your login shell. Before running these commands, you will want to be in a tmux or screen session so that you can start your job, and log off without having to worry about your job getting terminated.

Note Note: At a given point of time, unless you are using the ckpt (formerly the bf) queue, you can have one instance of big_machine and three instances of any_machine running at the same time. You may need to coordinate over IRC if you need to use a specific node for any reason.

Killing jobs on compute nodes[edit]

The Slurm scheduler provides a command called scancel to terminate jobs. For example, you might run queue_state from a login node to figure out the ID number for your job (let's say it's 12345), then run scancel --signal=TERM 12345 to send a SIGTERM signal or scancel --signal=KILL 12345 to send a SIGKILL signal that will bring job 12345 to an end.

Parallel R[edit]

The nodes on Hyak have 28 CPU cores. These may help in speeding up your analysis significantly. If you are using R functions such as lapply, there are parallelized equivalents (e.g. mclappy) which can take advantage of all the cores and give you a 2800% boost! However, something to be aware of here is your code's memory requirement—if you are running 28 processes in parallel, your memory needs can also go up to 28x, which may be more than the ~200GB that the big_machine node will have. In such cases, you may want to dial down the number of CPU cores being used—a way to do that globally in your code is to run the following snippet of code before calling any of the parallelized functions.

options(mc.cores=20)  ## tell the mc* functions to use 20 cores unless otherwise specified

More information on parallelizing your R code can be found in the parallel package documentation.

Using the Checkpoint Queue[edit]

Hyak has a special way of scheduling jobs using the checkpoint queue. When you run jobs on the checkpoint queue, they run on someone else's hyak node that they aren't using right now. This is awesome as it gives us a huge amount of free (as in beer) computing. But using the checkpoint queue does take some effort, mainly because your jobs can get killed at any time if the owner of the node checks it out. So if you want to run a job for more than a few minutes on the checkpoint queue it will need to be able to "checkpoint" by saving it's state periodically and then restarting.

This would be a pain to do manually, fortunately, we have dmtcp which can automatically checkpoint and resume most programs.

Nate's working got dmtcp working for arbitrary scripts, and also with wikiq using parallel_sql.

dmtcp 3.0 is installed on Mox.

This will make more sense if you know that dmtcp works by starting a coordinator process which is responsible for pausing and saving the checkpointed process. A tutorial on dmtcp with slurm from USC has a bash function for starting the coordinator called start_dmtcp_coordinator. Nate added this function to the shared .bashrc. So it should be available in your environment on Mox.

Starting a checkpoint queue job[edit]

To start a checkpoint queue job we'll use sbatch instead of srun. See the documentation for a refresher starting hpc jobs using sbatch.

To request a job on the checkpoint queue put the following in the top of your sbatch script.

   #SBATCH --export=ALL
   #SBATCH --account=comdata-ckpt
   #SBATCH --partition=ckpt

You'll might have other stuff in your SBATCH script to request a certain number of cores or memory. Those will matter when we run wikiq below, but here they can be whatever they would be if you were running an sbatch job on one of our machines. The next thing you need to do specifically for a ckpt job is to run start_coordinator. This function takes care of making sure that we start a coordinator using the right set of ports and temporary files. We still need to pass in the interval that we want checkpoints. The bigger this interval the faster your job will run but the more work will be lost when it's interrupted.

   start_dmtcp_coordinator -i 600  #checkpoint every 10 minutes

Next you need to run your job in a special way so that it is managed by dmtcp and restarted if it gets interrupted.

   # The restart script is created by dmtcp_launch after initialization
   if [ -x ]; then
       # On first pass, run program under DMTCP
       dmtcp_launch --rm [YOUR SCRIPT]	# must run interpreter for scripts

This works because is created when you launch your job using dmtcp_launch. If that script exists your job should run it instead of your job.

There are options that you can pass to dmtcp_launch that can be important. In particular --checkpoint-open-files and --allow-file-overwrite modify how IO is checkpointed.

Running wikiq with dmtcp and parallel_sql[edit]

To run wikiq with parallelsql the following need to be arranged:

  1. A shell script for each dumpfile that makes a workspace for dmtcp to keep it's data and restart script.
  2. These shell scripts loaded in parallel sql.
  3. A sbatch script that gets a checkpoint node and starts running jobs from parallel_sql.
  4. You need to restart jobs that get interrupted using parallel sql.

Nate made a python script that generates the scripts and makes a file with all the scripts. Notice that each dumpfile gets a script, it's own checkpoint directory, and a line in

#!/usr/bin/env python3
from os import path
import os
import stat
import glob

archives = glob.glob("/gscratch/comdata/raw_data/wikia_dumps/2010-04-mako/*.xml.7z")

scripts_dir = '/gscratch/comdata/users/nathante/wikiq_parallel_scripts'
output_dir =  '/gscratch/comdata/users/nathante/wikiq_output'
checkpoint_dir = '/gscratch/comdata/users/nathante/wikiq_checkpoint'

if not path.isdir(scripts_dir):

if not path.isdir(output_dir):

script ="""#!/bin/bash
mkdir -p {0}
cd {0}
start_dmtcp_coordinator -i 60  #checkpoint every 20 minutes

if [ -x ]; then
    # On first pass, run program under DMTCP
    dmtcp_launch --rm {1}

with open("",'w') as calls:
    for dumpfile in archives:
        wikiq_base_call = f"wikiq -u -o {output_dir} {dumpfile}"
        wikiq_call = wikiq_base_call
        wiki = path.split(dumpfile)[1]
        wikiq_script = script.format( path.join(checkpoint_dir,wiki), wikiq_call)

        script_file = path.join(scripts_dir, wiki + '.sh')
        with open(script_file,'w') as of:
        os.chmod(script_file,os.stat(script_file).st_mode | stat.S_IEXEC)


We also need an sbatch script as

#SBATCH --job-name=wikiq_dmtcp
## Allocation Definition
#SBATCH --account=comdata-ckpt
#SBATCH --partition=ckpt
## Resources
## Nodes. This should always be 1 for parallel-sql.
#SBATCH --nodes=1    
## Walltime (12 hours)
#SBATCH --time=12:00:00
## Memory per node
#SBATCH --mem=100G

module load parallel_sql

#Put here commands to load other modules (e.g. matlab etc.)
#Below command means that parallel_sql will get tasks from the database
#and run them on the node (in parallel). So a 16 core node will have
#16 tasks running at one time.
parallel-sql --sql -a parallel --exit-on-term

Next load the scripts into parallel_sql

 module load parallel_sql
 cat | psu --load

We can now fire up a whole bunch of checkpoint nodes. The limit is technically 2000! But let's just ask for 10 nodes :)

  for job in $(seq 1 10); do sbatch; done

If our jobs get interrupted we'll need to run psu --reset-slurm to set them back into avail state. We can run a little script running on a login node to do this automatically every minute or so.

#!/usr/bin/env python3
import time
import subprocess

running =["psu", "--show-running"],  universal_newlines=True, stdout=subprocess.PIPE, stderr=subprocess.PIPE)
while hasattr(running, 'stdout') and len(running.stdout) > 0:["psu","--reset-slurm"])
    running =["psu", "--show-running"],  stdout=subprocess.PIPE)

That's it! Unleash the power of the checkpoint queue! Reach out to Nate if you try this and have problems or if you have any questions!

Jupyter Notebook on Hyak[edit]

Set up a password for Jupyter Notebook on Hyak[edit]

Working on Hyak from a local emacs client[edit]

Custom software in Hyak[edit]

R packages[edit]

To install a R package that's not available globally, you can check out a build node, and install the package locally. Here's how to do it:

$ build_machine
$ R

This will start R, where you can install a package in the usual way. The build node has access to the Internet, so it will be able to download the required source packages, etc.

> install.packages('lme4')

Python Packages[edit]

DO NOT TRUST THIS SECTION. Intel python appears to have some issues.

The recommended python to use on hyak is the intel-python. This is a customized anaconda distribution with a magical optimization of python that really increases the performance of numpy.

Using an anaconda python distribution has important implications for how you install packages. While in normal python, you would install python packages using `pip`, when you use an anaconda distribution you should use `conda` to install packages. Conda also has some fancy features like virtual environments for using different versions of python or different versions of packages in different projects. The problem with using conda is that it does not include all the packages you might want to use. If you want to install a python package that is missing from conda, you can use pip.

Importantly, when using intel-python, you should prefer to install software using conda over pip.

Conda Documentation Pip Documentation

The first time you use intel-python you need to create a custom environment for installing software:

   conda create -n my_root

Then add the following to your .bashrc to use this environment.

   if [ -z $(conda info --env | grep my_root | grep \*) ]; then
       source activate my_root

Conda doesn't like it when you try to activate an environment that is already active. T

Conda modifies your prompt in a possibly annoying way. To disable this behavior run the command:

   $ conda config --set changeps1 False

Custom modules[edit]

Software on Hyak can be outdated, or in some cases, not available at all. In some of these situations, it may be possible to use environment modules to install and run software without necessitating administrative (root) privileges. For example, it is possible to have and run the newest version of R that is installed in a central, shared directory, and it is even possible to have multiple versions of R available in parallel. The following subsection shows how to do this. Ordinarily, this should not be necessary on a day-to-day basis.

Installing and making available a custom module[edit]

Note Note: If you are using screen to run and manage your builds, keep in mind that screen drops a few environment variables such as LD_LIBRARY_PATH, which may mess up your build process. You should check that all the relevant environment variables are set before starting your build.

The first step toward installing and making available a custom module (in this case, R 3.5.0) is to spin up the build node, download R, compile it with a specific prefix, and install it.

$ build_machine
$ module load contrib/texlive/2017  # loads the texlive module that is helpful for generating R documentation
$ module load contrib/openblas/0.2.20  # loads the openblas library, which speeds up some R operations significantly
$ wget
$ tar xzvf R-3.5.0.tar.gz
$ cd R-3.5.0
$ ./configure --prefix=/gscratch/comdata/modules/sw/R/3.5.0  --without-x --enable-R-shlib --with-lapack --with-blas="-L/sw/contrib/openblas/0.2.20/lib -lopenblas"
$ make
$ make install

The --prefix option to ./configure tells the build scripts that R is going to be installed in /gscratch/mako/modules/sw/R/3.5.0. This follows a convention that we picked—software in modules should go into /gscratch/mako/modules/sw/{SOFTWARE_NAME}/{SOFTWARE_VERSION}. The --prefix option is the most important flag for ./configure—any other flag or option will be specific to the software being installed.

The second step is to write a modulefile. This contains the metadata about our module. Edit the file /gscratch/mako/modules/modulefiles/R/3.5.0 to contain the following

proc ModulesHelp { } {
        puts stderr "\tModule providing R 3.5.0."

module-whatis "Module providing R 3.5.0."

module load contrib/openblas/0.2.20
prepend-path    PATH            /gscratch/mako/modules/sw/R/3.5.0/bin
prepend-path    MANPATH         /gscratch/mako/modules/sw/R/3.5.0/share/man

# The following line prevents everyone from installing libraries in the global namespace
file mkdir ~/R/x86_64-pc-linux-gnu-library/3.5

Note that the filename follows a similar convention as --prefix earlier (/gscratch/mako/modules/modulefiles/{SOFTWARE_NAME}/{SOFTWARE_VERSION}). This file sets up the PATH and MANPATH environment variables appropriately so that the specified version of R can be accessed and run as needed. There are many more directives that can go into the modulefile—see man modulefile for details on those directives.

Once this file is written out, the module avail command should list R/3.5.0 as an available module. This is because the module system is set up to look inside /gscratch/mako/modules/modulefiles for module files, thanks to the MODULEPATH variable that is set through .bashrc. The command module load R/3.5.0 should make R available and ready for use. To avoid running module load R/3.5.0 whenever you log in, you can add the command at the end of your .bashrc file (after the section that sets MODULEPATH).


To use spack to manage software on hyak, add the following to your .bashrc.

## we need to load these modules to use proprietary hyak compilers to get faster code.                                                                                                                                        
module load icc_18-impi_2018
module load icc_18
export LD_LIBRARY_PATH = /sw/intel-2018/lib/intel64:$LD_LIBRARY_PATH
# For bash/zsh users
export SPACK_ROOT=/gscratch/mako/spack/
. $SPACK_ROOT/share/spack/


For directions on working with spack, see the spack documentation.