# Triton v3 communication

Proposal: draft everything as blog posts, post, send email with a link to that saying "any further updates will go to this blog post instead of as a separate email"

OR: should it become a Triton issue instead?

## Initial email/post
Title: Triton v3: open for testing

*This post will be updated with additional info, as people ask questions to us.*

We are happy to announce that Triton v3 has begun testing. V3 will use all of the same hardware and data, but is running a new operating system. This should allow us to solve some of the common problems we have had lately and provide a solid base for future growth.

### What is happening now

As of 9 April 2024, about 20% of current Triton resources (cpu, gpu, etc) have moved to the v3 operating system. Adventurous users can begin testing and running jobs here (you might find there are lots of resources available; maximum job duration is 2 days, though). The rest will be migrated shortly, once we are sure things work. **For now you can keep using old or new** - we aren't forcing a change right now, but the day will eventually come.

### Data and paths

**All data and paths are identical.** If you modify data on one side, it's modified on the other side (the storage systems haven't changed and it's serving both v2 and v3 at the same time). While it may seem similar, other than data and accounts, the environments of v2 and v3 are fully separate.

### Login to v3

You can login to v3 now, but the default login methods will still take you to v2. Login is via
- SSH to `login4.triton.aalto.fi`
- https://ondemand.triton.aalto.fi (Open OnDemand web interface - note that this is a different address)
- Jupyter (there is no direct replacement in V3 - use Open OnDemand instead where you can find jupyter)

### Software

**All software has been fully reinstalled - this is the hardest point of the transition.** This means any modules will be different, our provided Python/R/Matlab/etc modules are different, and so on. Module names may be slightly different - you will need to check and update your workflows. We'll be adding transition guides for different types of workflows later.
- Python and JupyterHub: TODO
- Software you have compieled: you must re-compile, see :doc:`tritonv3-compiled`
- R: use new modules and re-compile any locally installed packages

Note that we have hierarchical modules (we've added this to v2 already, but it's worth pointing it out again): When you `module spider`, it will tell you what modules you must load *before* you can load other modules. For example, it will tell you to load the relevant compiler first, and then you can load a module.

The new system is based on a new version of linux: Red Hat Enterprise Linux 9.

Using the cluster should be mostly the same.

### GPUs
For gpu's the slurm notation has been updated. From now on, use `--gpus=N` and `--gpus=TYPE:N` instead of `--gres=gpu:N`). Check docs for details. We recommend on moving to the new notation asap because it works on v2 also.

We have now split our gpu-partition to multiple partitions. Names of these partitions are:

- `gpu-p100-16g`
- `gpu-v100-16g`
- `gpu-v100-32g`
- `gpu-a100-80g`

Each partition has the corresponding GPU types and thus in future you can choose which GPU to use by submitting your job to the corresponding partitions with e.g. `#SBATCH --partition gpu-v100-32g,gpu-a100-80g`.

Constraints will still work, so you can use those to limit yourself to certain GPU types.

### Reporting problems

If you notice problems, do let us know! It is probably either something that is wrong, or something we need to clarify. **First, check this post for updates.** Then, check the :ref:`issuetracker-general` for a similar problem, then report issues to that tracker:

- missing features or software
- something that is not working
- working differently than you expect

You can use :ref:`chat` for quick questions, but we may be busy so if you don't get a response, make an issue instead.

---

## Blog post: Python
Title: Triton v3: Python and Jupyter

This blog post describes how to migrate to Triton V3, if you use Python or Jupyter. Read more about the migration at :doc:`triton-v3-migration`. Our main documentation at scicomp.aalto.fi is in the progress of being updated (and may not be fully updated yet).

### Generic Python

The `anaconda` module is replaced with `scicomp-python-env` (the previous module wasn't default Anaconda anyway).

This module is a generic environment that has many common packages inside of it. We can install some more packages here, but for advanced use you should make your own conda environments.

### Conda environments

:ref:`ref-conda`

Old conda environments *might* still work, but in case of any problems reinstalling the environment is the recommended way of fixing problems. In general, we would recommend to export the environment and reinstall it to avoid unexpected side effects (especially if additional pip packages were installed)

You will also have to re set up your conda configuration (see [here](https://scicomp.aalto.fi/triton/apps/python-conda/#first-time-setup))
Instead of using `miniconda` one should load the module `mamba` for loading and creation of other environments. This is a minimal environment with just conda and mamba and in it:

$ module load mamba
$ mamba env create -f environment.yaml
$ source activate NAME

Remember, `mamba` is a drop-in replacement for `conda`, but has a much faster dependency resolution algorithm. They can be used together.

### Virtual environments

Virtual environments need to recreated, since the Python interpreter they depend on will change.

### JupyterHub

Triton JupyterHub won't continue. Instead, a similar environment is available in Open OnDemand. The v3 Open OnDemand is available at https://ondemand.triton.aalto.fi/ .

By using the "Jupyter" interactive app, you can start a Jupyter environment equivalent to the old JupyterHub (comes with pre-installed kernels, you can install your own kernels). In the future, it will be possible to run Jupyter in your own environments.

The biggest known issue is that Open OnDemand isn't yet available outside of Aalto networks (VPN must be used). If this is important to you, please let us know.

---

## Blog post: compiled software
Title: Triton v3: compiled software and toolchains

All complied software must be re-compiled. You can find compilers and libraries if you search for modules. Everything is re-installed, so some lesser-used things might still be missing (ask us in that case).

The blog post is created to have a place to update FAQs about the transition, that might be hard to find from scicomp.aalto.fi. Check back here if you have problems.