- Oct 13, 2022
-
-
SIMONIN Matthieu authored
-
SIMONIN Matthieu authored
-
SIMONIN Matthieu authored
-
SIMONIN Matthieu authored
-
SIMONIN Matthieu authored
-
SIMONIN Matthieu authored
-
SIMONIN Matthieu authored
Examplify with g5k
-
SIMONIN Matthieu authored
-
SIMONIN Matthieu authored
-
SIMONIN Matthieu authored
-
SIMONIN Matthieu authored
+ installation note
-
SIMONIN Matthieu authored
-
SIMONIN Matthieu authored
-
SIMONIN Matthieu authored
- remove unecessary `finalize` calls - remove unecessary prod network declaration
-
SIMONIN Matthieu authored
-
SIMONIN Matthieu authored
Fix typo in docs See merge request discovery/enoslib!133
-
-
- Oct 12, 2022
-
-
SIMONIN Matthieu authored
CI: Add pylint for better static analysis See merge request discovery/enoslib!130
-
JONGLEZ Baptiste authored
-
JONGLEZ Baptiste authored
-
JONGLEZ Baptiste authored
UserList has long been deprecated, because it's now possible to inherit directly from "list". Inheriting from "list" clearly helps pylint figuring out what is the type of the elements of the custom list.
-
JONGLEZ Baptiste authored
I choose not to add pylint as a pre-commit hook because: - it requires dependencies to be installed, so it will fail locally when some optional dependencies are not installed (e.g. iotlab, chameleon) - it's much slower than the existing pre-commit hooks
-
JONGLEZ Baptiste authored
-
JONGLEZ Baptiste authored
-
JONGLEZ Baptiste authored
-
SIMONIN Matthieu authored
Fix check typing See merge request discovery/enoslib!131
-
-
- Oct 07, 2022
-
-
SIMONIN Matthieu authored
G5k/multisite: migrate to Providers. See merge request discovery/enoslib!127
-
- Oct 06, 2022
-
-
SIMONIN Matthieu authored
The idea is to scope our G5k provider. Scoping means that the actions on remote resources (e.g jobs) are restricted to a white list of sites. Another key thing is that working on a single site or working in multisite follows a similar workflow: ``` reserve -> get resources -> initial provisionning (deploy | grant_root_access) ``` Note that the `initial provisioning` takes as input the resources (nodes, networks) got from the `get_resources` function and already works with resources from different sites. In conclusion we only need to make sure the `reserve` and the `get_resources` work in a multisite setting. The generic workflow is implemented in a `G5KBase` class and we introduce the inherited class G5k. It will overwrite the reserve to work with the Providers (plural) abstraction and reload the resources from all sites (unscoped provider).
-
SIMONIN Matthieu authored
-
SIMONIN Matthieu authored
-
SIMONIN Matthieu authored
Python refactoring and typo error fixes See merge request discovery/enoslib!129
-
-
SIMONIN Matthieu authored
g5k: Don't deploy by default, and require an env_name when deploying See merge request discovery/enoslib!125
-
JONGLEZ Baptiste authored
-
JONGLEZ Baptiste authored
-
JONGLEZ Baptiste authored
-
JONGLEZ Baptiste authored
This is just cosmetic
-
JONGLEZ Baptiste authored
-
JONGLEZ Baptiste authored
The idea is to make things more explicit and reproducible: - by default (no `job_type`), we now use the standard Grid'5000 env with sudo-g5k instead of deploying. This is much faster to provision, and matches the Grid'5000 oarsub default when no job type is specified. - when using the `deploy` job type, we now also require the `env_name` parameter instead of deploying a default env (currently `debian11-nfs`). This makes sure that the user explicitly chooses an environment to deploy, which is good for reproducibility: the default env chosen by Enoslib tended to evolve over time as new stable Debian releases become available on Grid'5000. For users that don't specify any job type, they will now end up with a `debian11-std` environment instead of `debian11-nfs`. This is very similar, except that many more software packages are installed ("std" is equivalent to "big" in [1]). For backwards compatibility, we automatically add the 'deploy' job type if the user specifies an `env_name` (to account for the previous default of deploying). However, user code should be updated. Also for backwards compatibility, if a Kavlan network is used, we automatically add the 'deploy' job type and set `env_name` to "debian11-nfs" if missing. But again, user code should be updated to be more explicit and choose an appropriate environment to deploy. Note that "allow_classic_ssh" is still supported as job type but has no effect on Grid'5000 and is deprecated (see #130). You can get the same behaviour by just using `job_type=[]` or omitting the job type altogether. This is because of a recent change in the Grid'5000 platform itself that made allow_classic_ssh the default behaviour. [1] https://www.grid5000.fr/w/Getting_Started#On_Grid.275000_reference_environments
-