Rosetta
A container-centric science platform
for interactive workloads
Rosetta in a nutshell
- Web-based science platform supporting GUI-based, interactive computing
- Batteries included, runs entirely on Docker containers
- Executes workloads in Docker or Singularity containers on a number of computing elements:
- local CPUs;
- remote servers (including commercial Clouds);
- HPC Clusters (i.e. Slurm).
- Requires minimum effort to get Rosetta-compatible (just extend base containers)
- Allows users to setup and run their own containers, autonomously.
Example
- Task list and remote desktop container connection via automatic tunneling and Web VNC
******
A step back: what container actually are?
“Containers are a solution to the problem of how to get software to run
reliably when moved from one computing environment to another.”
[Link]
A step back: what container actually are?
“Containers are a solution to the problem of how to get software to run
reliably when moved from one computing environment to another.”
[Link]
A.K.A. the dependency hell problem
The “dependency hell” problem
Mike wants to use a new software.
Mike cannot find a precompiled version that works with his OS and/or libraries.
Mike asks/Google for help and gets some basic instructions - like “compile it”.
Mike starts downloading all the development environment, and soon realizes that
he needs to upgrade (or downgrade!) some parts of his main Operating Systems.
During this process, something goes wrong.
Mikes spends an afternoon fixing his own OS, and all the next day in trying to
compile the software. Which at the end turns out not to do what he wanted.
The “dependency hell” problem: solutions spectrum
Proper Virtual Statically
Containerization VMs VMs with
requirements environments linked
hardware
binaries
emulation
The “dependency hell” problem: solutions spectrum
Proper Virtual Statically
Containerization VMs VMs with
requirements environments linked
hardware
binaries
emulation
The “dependency hell” problem: solutions spectrum
Proper Virtual Statically
Containerization VMs VMs with
requirements environments linked
hardware
binaries
emulation
Singularity Docker
Singularity vs Docker
Singularity Docker
Scientific computing IT industry standard
Running container are seen as processes Running containers are seen as (micro)services
Build as root, run as user Need near-root access or proper orchestrators
Limited or no support for networking Extensive support for networking
Singularity vs Docker
Singularity Docker
Filesystem: only partially isolated,directories as $HOME, Filesystem: completely isolated by default, volume or
/tmp, /proc, /sys, and /dev are all binded by default. folder binds must be explicitly set
Environment: from the host Environment: from scratch
Network: from the host Network: dedicated subnet
More similar to an environment More similar to a virtual machine
Hard truth:
Environments-like containerisation solutions like Singularity do not not solve the dependency hell problem
and how to reliably move software around, which in turn does not guarantee reproducibility!!
Hard truth:
Environments-like containerisation solutions like Singularity do not not solve the dependency hell problem
and how to reliably move software around, which in turn does not guarantee reproducibility!!
Singularity in Rosetta
Singularity is still very interesting as it can be used autonomously by users without asking
much effort from sysadmins or to change the scheduling / workload management.
To overcome its environment-like limitations, in Rosetta Singularity is always forced to run in
a specific way which allow to achieve an almost container-like behaviors, as Docker.
This requires a few specific command line switched (set by Rosetta itself) and some extra
orchestration (for sandboxing)
mkdir -p /tmp/ee3f2776-1b9e-4dd4-a250-f80813dc9fe0_data/tmp && \
mkdir -p /tmp/ee3f2776-1b9e-4dd4-a250-f80813dc9fe0_data/home && \
chmod 700 /tmp/ee3f2776-1b9e-4dd4-a250-f80813dc9fe0_data && \
singularity run --pid --no-home --containall --cleanenv --writable-tmpfs
--workdir /tmp/ee3f2776-1b9e-4dd4-a250-f80813dc9fe0_data/tmp --home=/home/metauser \
-B/tmp/ee3f2776-1b9e-4dd4-a250-f80813dc9fe0_data/home:/home/metauser
Container interfaces & graphical applications
One missing bit in using containers (regardless of Docker or Singularity) is that the software
you use to interact with your container is a dependency.
Usually, this dependency is a command-line, or a terminal, which is stable from the 80’s.
However, with GUI applications, how to ensure the most common possible dependency?
Container interfaces & graphical applications
Possible options:
1. X11
2. VNC
3. Web-based VNC
Container interfaces & graphical applications
Possible options:
1. X11
2. VNC
3. Web-based VNC
This one. Why? ..because like it or not, web
browsers and Javascript are the most standard
technologies you can find in your usebase
The interface is the
web browser
..therefore:
Reproducible
scientific
stuff within
Singularity
Rosetta quickstart & demo (for users)
Rosetta quickstart & demo (for users)
Rosetta quickstart & demo (for users)
Rosetta quickstart & demo (for users)
Rosetta quickstart & demo (for users)
Rosetta quickstart & demo (for users)
Rosetta quickstart & demo (for users)
Rosetta quickstart & demo (for users)
Rosetta quickstart & demo (for users)
Rosetta quickstart & demo (for users)
Rosetta quickstart & demo (for users)
Rosetta quickstart & demo (for users)
Rosetta quickstart & demo (for users)
Rosetta quickstart & demo (for users)
Rosetta quickstart & demo (for users)
Rosetta quickstart & demo (for users)
Rosetta Architecture
External or internal
File Storages
External Singularity
Registry Rosetta Agent
Rosetta Platform
External or internal
Docker Registry Singularity Docker
Computing element
Registry Registry
Singularity Docker
Container Container Web App
Single node setup
Input files
User
Output files
Rosetta technology stack
1. Docker + docker compose for orchestrating Rosetta itself
2. Python + Django for the orchestration
3. HTML for the Webapp
4. REST APIs (in development)
5. Postgres (or SQLite) database
Rosetta quickstart (for devs)
Rosetta status
- Deployed on [Link], invite-based only for power-users beta tester
- A few user from other institutes using it autonomously on their own laptops for its
orchestration capabilities
→ (i.e. to easily access remote desktops when locked out from offices in the pandemic.)
- Supports Singularity containers with preliminary support for Docker.
- Ships “batteries included”, with a few containers preconfigured and a test Slurm cluster
instance up & running
That’s it :)
Questions?
Stefano Alberto Russo
[Link]@[Link]
Rosetta repository: [Link]
