Hi,
Thank you for this work > Considering that the standard public runner on Github runs on a 4 CPU + 16 GB > RAM machine intel arch, the rough conclusion is that we would > need 4.5 Isn't it 3.5 instead ? That's the number you get in the table Assuming that we would have a better control on these machine, maybe we could have more disk space and so maybe more build cache that would speed up the build time. It could also reduce the time to pull some resource elsewhere (docker, oracle/hana binary...). It's highly hypothetical, I'm just wondering. Regards, Julien > Hi, > > During the last PSC meeting we talked briefly about how to solve the problem > that we have with the Github CI limitations, one of the possible > solutions that we discussed was to start migrating part of the CI to > self-hosted runners. > > I've just made an attempt to understand the hardware requirements that we > would need and I have collected some statistics from our Github > account, summarized here for the period of the last 30 days: > > https://docs.google.com/spreadsheets/d/16-tiSLndm-ISxRFgZcE-Ewytr8cwLj00gdYs1iBsz58/edit?usp=sharing > > Considering that the standard public runner on Github runs on a 4 CPU + 16 GB > RAM machine intel arch, the rough conclusion is that we would > need 4.5 of these machines to handle the actual workload, please note that > this a very rough estimation and does not take into account that we > probably have peaking hours and we'd need more power if we don't want the > jobs to sit in a queue for too long. > > Anyway, it's a start. > > Another thing to consider is that we could possibly cut some CI workflows > (e.g. mingw64, is that useful?) or move some to a daily cronjob > (ogc?). > > Any thoughts? -- Julien Cabieces Senior Developer at Oslandia [email protected] _______________________________________________ QGIS-Developer mailing list [email protected] List info: https://lists.osgeo.org/mailman/listinfo/qgis-developer Unsubscribe: https://lists.osgeo.org/mailman/listinfo/qgis-developer
