CommunityData:Hyak

From CommunityData

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 Feel free to edit and fix the content that is incorrect/out-of-date.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

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 mox2.hyak.uw.edu
     User sdg1
     HostName mox2.hyak.uw.edu
     ControlPath ~/.ssh/master-%r@%h:%p
     ControlMaster auto
     ControlPersist yes
     Compression yes


ONE WARNING: 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.

Connecting to Hyak

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 a [2-factor authentication request].

Setting Up Hyak

When setting up Hyak, you must first add this stanza to the very bottom of your ~/.bashrc file. Generally, you can simply edit the following file on Hyak: ~/.bashrc

## START hyak-cdsc specific options -- BOTTOM OF FILE

source /etc/profile.d/modules.sh
module load contrib/r/3.4.3

alias big_machine='srun -p mako -A mako --mem=248G --time=500:00:00 --pty bash -l'
alias any_machine='srun -p mako -A mako --mem=124G --time=500:00:00 --pty bash -l'
alias build_machine='srun -p build --mem=100G --time=2:00:00 --pty bash -l'
alias queue_state='squeue -p mako'

umask 007

## END hyak-cdsc specific options -- BOTTOM OF FILE

The final line is particularly important. If you do not do this, the files you create on Hyak will be able to be read or written by others in the group!

Once you do this, you will need to restart bash. This can be done simply by logging out and then logging back in or by restarting bash with the command exec bash.

I also add these two lines to my Hyak .ssh/config:

ForwardX11 yes
ForwardX11Trusted yes

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

Python Packages

If you need python libraries that are not installed in the shared environment:

$ pip3 install --user YOURLIBHERE

...replacing YOURLIBHERE with the name of the library you need, e.g. 'pandas'. The --user option will install it for just you.

If you have a lot of dependencies for a specific project, consider using Python Virtual Environments

Parallel R

The hyak machines have 16 cpu cores. The Mox machines will have 28! Running your program on all the cores can speed things up a lot! We make heavy use of R for building datasets and for fitting models. Like most programming languages, R uses only one cpu by default. However, for typical computation-heavy data science tasks it is pretty easy to make R use all the cores.

For fitting models, the R installed in Gentoo should use all cores automatically. This is thanks to OpenBlas, which is a numerical library that implements and parallelizes linear algebra routines like matrix factorization, matrix inversion, and other operations that bottleneck model fitting.

However, for building datasets, you need to do a little extra work. One common strategy is to break up the data into independent chunks (for example, when building wikia datasets there is one input file for each wiki) and then use mcapply from library(parallel) to build variables from each chunk. Here is an example:

   library(parallel)
   options(mc.cores=detectCores())  ## tell R to use all the cores
   
   mcaffinity(1:detectCores()) ## required and explained below 
  
   library(data.table) ## for rbindlist, which concatenates a list of data.tables into a single data.table
   
   ## imagine defining a list of wikis to analyze
   ## and a function to build variables for each wiki
   source("wikilist_and_buildvars")
   
   dataset <- rbindlist(mclapply(wikilist,buildvars))
   
   mcaffinity(rep(1,detectCores())) ## return processor affinities to the status preferred by OpenBlas

A working example can be found in the Message Walls git repository.

mcaffinity(1:detectCores()) is required for the gentoo R library(parallel) to use multiple cores. The reason is technical and has to do with OpenBlas. Essentially, OpenBlas changes settings that govern how R assigns processes to cores. OpenBlas wants all processes assigned to the same core, so that the other cores do not interfere with it's fancy multicore linear algebra. However, when building datasets, the linear algebra is not typically the bottleneck. The bottleneck is instead operations like sorting and merging that OpenBlas does not parallelize.

The important thing to know is that if you want to use mclapply, you need to do mcaffinity(1:detectCores()). If you want to then fit models you should do mcaffinity(rep(1,detectCores()) so that OpenBlas can do its magic.

Jupyter Notebook on Hyak

1. Choose a number you are going to use as a port. We should each use a different port and the number should be between 1000 and 65000. It doesn't matter what it is but it needs to be unique. Pick something unique. In the following instructions, replace $PORT with your number below.

2. Connect to Hyak and forward the the port from you local machine to the new one:

ssh -L localhost:$PORT:localhost:$PORT username@hyak.washington.edu

You can also add the following line to the Hyak section on your local .ssh/config file on your laptop:

    LocalForward $PORT localhost:$PORT

3. We're going to need to connect to one of the compute servers twice. As a result, we'll use a program called tmux. Tmux is very similar (but a little easier to learn) than a program called screen. If you know screen, just use that. Otherwise, run tmux like:

tmux

You can tell you're in tmux because of the green line at the bottom of the screen.

4. "Check out" a compute node

any_machine

5.

Keep track of which machine you are on. It should be something like n0650 and it should be displayed on the prompt. We'll refer to it as $HOST below.

6. Start jupyter on the compute node:

jupyter-notebook --no-browser --port=$PORT

You'll see that jupyter just keeps running in the background. This can be useful because when there are errors, they will sometimes be displayed in this terminal. Generally, you can just ignore this though.

6. Create a new window in tmux/screen

At this point, you have jupyter running on the compute node on $PORT. You also will have forwarded the port from your laptop to the login node. We're really only missing one thing which is the tunnel from the login node to the compute node within hyak. To do this, we'll create a new window inside tmux with the keystroke Ctrl-b c.

If you're not familiar with it, you'll want to read the CommunityData:tmux which includes a quick cheatsheet. To switch back to the original window running jupyter, you should type: Ctrl-b 0. If you switch though, be sure to switch back to the new window with Ctrl-b 1.

Because you originally ran tmux on the login node, the new window/terminal will be opened within tmux on the login node.

7. Open a tunnel from the login node to the compute node.

 ssh -L localhost:$PORT:localhost:$PORT $HOST

8. In your local browser, localhost:$PORT

Set up a password for Jupyter Notebook on Hyak

Once you have IPython/Jupyter up and running on Hyak and have set up all the port forwarding stuff described above, you might consider adding a password to secure your Jupyter session. Why bother? Anyone with access to Hyak can see that you're forwarding something via the login node. While unlikely, they may do something to interrupt or otherwise mess with your session. It should work. Keep in mind that anyone with access to your jupyter session can do anything you can do on the command line including access all your data, delete files, etc.

Instructions for setting up a password on your Jupyter sessions are available on the Hyak wiki (UW login required).

Note that you can/should skip the first command that loads the canopy module.

Running Jobs on Hyak

Screencast Examples:

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 itSigs documentation. Following are basic instructions for some common use cases.

Interactive nodes

For simple tasks, e.g. running R on a dataset, testing that code is working, etc. it is easiest to run it in an interactive node. This is a compute node that you interact with through the terminal. All of your disk storage is accessible just as though you were on the login node.

Parallel SQL

For big jobs you will want to use multiple nodes. Hyak has a very cool tool that makes this very easy, called Parallel SQL. Detailed instructions are in the itsigs parallel-sql documentation. There is also a full walkthrough example with instructions.

The basic workflow is:

1. Prepare the code, and test it with a single file (either on your computer, or on an interactive node). 2. Write a job_script file. This tells the node what job to run. There is an example on the Parallel SQL wiki page (linked above), and an example in the wikiresearch/hyak_example directory. 3. Create a task_list file. This is a list of commands that should be run, with one line per file that the command should operate on. An example file might look something like:

python analysis_script.py -i ./input/wiki_1.tsv -o ./output/wiki_1_analysis.tsv
python analysis_script.py -i ./input/wiki_2.tsv -o ./output/wiki_2_analysis.tsv
...

The README in the hyak_example directory has some example bash commands that you might use to generate this file.

4. Load the task_list into Parallel SQL.

$ module load parallel_sql
$ cat task_list | psu --load

5. Run the job_script on as many nodes as you need. When each task is finished, the node will get the next task from Parallel SQL.

$ for job in $(seq 1 N); do qsub job_script; done 
# N is the number of nodes

You can also use the -t flag, which makes jobs using multiple nodes easier to kill, but is not recommended by "the HYAK people".

$ qsub job_script -t 0-N
# N is the number of nodes

R Markdown

R markdown is a useful way of writing up your analysis as a mix of explanatory text and code. You can, for example, create fancy Tufte-style handouts with code and explanatory text in the same file! In order to use R markdown, in a compute node, run the following command

$ Rscript -e "rmarkdown::render('analysis.Rmd')"

Python Virtual Environments

Python virtual environments are a great way to manage project dependencies, and they seem to work on HYAK in the same way that they do on local machines. First install virtualenv using pip (this only needs to be done once).

$ pip install virtualenv --user

Initialize a new virtual environment in the current directory. Many people create a new virtual environment for each project.

$ # this virtual environment will use python 3
$ virtualenv venv -p python3

To activate the virtual environment from a login node or an interactive compute node

$ source <path_to_venv_parent_dir>/venv/bin/activate

To load a virtual environment in parallel sql, add the following to your PBS bash script

source <path_to_venv_parent_dir>/venv/bin/activate

Killing jobs on compute nodes

Torque documentation suggests that you should do this with qdel. That might work, but apparently our system runs moab on top of torque and the recommended (by Hyak admins) way to kill a job is to use the mjobctl command.

For example, you might run nodestate from a login node to figure out the ID number for your job (let's say it's 12345), then run mjobctl -c 12345 to send a SIGTERM signal or mjobctl -F 12345 to send a SIGKILL signal that will bring job 12345 to an end.

Note that only four user accounts at a time can have the bits necessary to kill other people's jobs, so while you can do this on your own jobs, you'll need to bother the IRC channel to find help cancelling other's jobs (we think that Jeremy, Nate, Aaron, and Mako currently have the bits). Also, check out the documentation for mjobctl for more info.

Working on Hyak from a local emacs client

Some of us (like Nate) rely heavily on the Emacs text editor. Emacs speaks statistics is a powerful emacs mode for programming in R and doing data analysis. There are a few options for using Emacs on hyak. If you open emacs on an interactive node with X-forwarding enabled then you will get a nice graphical emacs window and plots you make will be displayed on your screen. But if you disconnect from Hyak you will lose your R session. This makes running emacs the normal way on an interactive node unsuitable for fitting models. Another disadvantage is that your will be working with an x-forwarded emacs and so will not look as nice or be as responsive as your local emacs.

Alternatively, you might run emacs in console mode in tmux. Then Hyak will keep running your R process even when you log out. The downsides here is that you can't view plots on your display (you could save them as a pdf, and then open the pdf on your local machine) and that some emacs key chords will collide with tmux key bindings and configuring tmux to fix this is a pain.

A better way is to run emacs server on a compute node on hyak and then open a local emacs client that connects to that server.

Instructions For ESS

Unfortunately, this requires running emacsserver on a login node and viewing plots does not work. These problems should go away if hyak let us forward X from a compute node and tunnel it through a login node. This doesn't seem to work as ssh -X n0649 doesn't seem to forward X.

1. Open tmux on a login node and start emacsserver.

 $ tmux
 $ emacs --daemon

2. Still in tmux, start an interactive session

 $ any_machine

3. In a new terminal (not tmux) ssh into the login node and start an emacs client (-c means in a new window).

 $ emacsclient -c

4. In this emacsclient open a shell, ssh to the compute node, and start an R process.

 M-x shell
 $ ssh n0649

5. With focus on the R process buffer in emacs, connect ESS to the R process.

  M-x ess-remote