- added to the class `ISqlDatabaseBackend` the ability to delete a sql table
- also added `SshAccessedMysqlDb`: an implementation of `ISqlDatabaseBackend` that implements the sql database as a remote sql database accessed using ssh
- added unit tests to validate the backends SshAccessedMysqlDb and SqliteDb
work related to [https://bugzilla.ipr.univ-rennes.fr/show_bug.cgi?id=3958]
- added to the class `ISqlDatabaseBackend` the ability to create a sql table
- also added `SqliteDb`: an implementation of `ISqlDatabaseBackend` that implements the sql database as a sqlite3 database
work related to [https://bugzilla.ipr.univ-rennes.fr/show_bug.cgi?id=3958]
- all styling errors are now fixed, but there are still warnings and information
- most functions have been converted to snake case
- most functions now have type hinting
work related to [https://bugzilla.ipr.univ-rennes.fr/show_bug.cgi?id=3873]
when Job2 was introduced, it was actually representing a job, but the class Job (which actually represented a Task, and this was misleading) already existed
also:
- added type hinting to ease code understanding
- took this opportunity to fix styling issues
- this bug was found when working on Bug 3575 - la page physix status renvoit internal server error
- in order to ease the detection of such regressions, added a unittest that covers this code
- fixed bug : it was caused by the fact that the database has no information about ietr machines order. Fixed this by ignoring ietr machines because they're only part of ipr cluster temporarily and as such, they shouldn't appear in the stats.
also some improvements:
- improved the colors of the owners of the machines so that tey're visually close for the same organism (eg ipr owners have close colors). This change was made in june 2020 I think (for the questionnaire about it needs related to eskemmdata)
- added a graph to show the age of cores
- prevented overlapping of dates in the graph showing the evolution of gflops price
- improved the handling of colors. As a result, the number of owners is no longer restricted, as the color palette is now no longer restricted in size
- improved readability by putting the legend on the upper left part of the graphs, where it no longer hides the upper right part of the graph
- fixed clutter of years on the x axis which made it difficult to read the years. Now, there is a limit in the number of years displayed on the x axis
- improved the legend showng the owners, which now displays the lab in addition to the department
- made ClusterStats work on graffy-ws2 : made code compatible with python3 (still compatible with python2 though)
- added support for physix in addition to simpatix)
- fixed bug
Ticket n°215786 : installer le logiciel Ls Dyna sur le cluster de calcul de l'IPR
also fixes Bug 2812 - la mise à jour de physix-master a échoué (parsing de qconf -sconf global)
As a result, clusterconfigurator no longer pesses with hardcoded sys.path, we now use PYTHONPATH, which is somewhat cleaner.
As mypy doesn't support python2, I had to migrate code from python2 to python3. Because some libraries are shared between all python code, python2 code using these libraries will be broken while all python code is not upgraded to python3.
Bug 2701 - mettre en place un mécanisme pour éviter le blocage des gpu par des jobs cpu
- après avoir tenté de représenter le powerdiagram en deux modes simultanés (normal et worst_case), j'ai finalement opté pour une option qui permet de choisir un seul mode à la fois. Le mode worst case scenario est actif par défaut parce qu'il est le plus important : il permet de vérifier que le cablage est bien dimensionné en cas de coupure edf ou d'une alim défectueuse. Bien que le mode de calcul soit différent, on retrouve bien la même conso sur edf dans les deux modes.
- amélioration : ajout de la conso en W de chaque appareil (pratique pour vérifier s'il n'y a pas d'incohérence)
- désormais, les prises des appareils doivent être renseignés, sinon powerdiagram affiche les données qui lui manquent. Ca a notamment permis de détecter un oubli qui faisait qu'un des câbles était affiché plus gros que les autres (16A au lieu de 10A) sans raison apparente.
- amélioration du code pour que les alimentations redondantes soient gérées correctement. Il en ressort que plus de câbles sont en surcharge.
- j'en ai profité pour afficher les alims secourues en vert et les non secourues en bleu.
amélioration: désormais le diagramme fait bien ressortir les câblages qui méritent de l'attention :
- en rouge les cables en surcharge (dont l'ampérage max peut dépasser la capacité)
- en noir les cables qui n'ont pas de charge
Pour que le rouge soit réservé aux cables à problème, les conventions de couleurs pour les cables non surchargés ont été modifieés : c'est désormais en teintes de bleu.
Guillaume Raffy