Plan and migrate all model files into git-lfs, to make development and distribution more convenient.
Description
Status | Subtype | Assigned | Task | ||
---|---|---|---|---|---|
Duplicate | None | T197180 Try to increase ORES deployment parallelism | |||
Resolved | Ladsgroup | T197096 [Epic] Use LFS for large ORES files | |||
Resolved | Ladsgroup | T197097 Migrate models to LFS |
Event Timeline
I'm reshaping all of the repos to use git lfs. Took a back up https://analytics.wikimedia.org/datasets/archive/public-datasets/all/ores/20180820/ and also will keep another backup in my laptop
All of the repos with model files now are using git lfs. Deploying this to prod is blocked on T192042
Mentioned in SAL (#wikimedia-operations) [2018-08-22T10:30:25Z] <Amir1> deploying ores gerrit:454283 to beta (T197097)
Mentioned in SAL (#wikimedia-operations) [2018-08-22T20:05:55Z] <ladsgroup@deploy1001> Started deploy [ores/deploy@8ff4da1]: Update ORES to use git lfs for model T197097
Mentioned in SAL (#wikimedia-operations) [2018-08-22T20:37:37Z] <ladsgroup@deploy1001> Finished deploy [ores/deploy@8ff4da1]: Update ORES to use git lfs for model T197097 (duration: 31m 42s)
Something for the documentation—I got the same "smudge" error as you reported earlier, but on a different repo and object. A git submodule sync seems to have solved everything in my case (though I'm still waiting an hour plus or minus two hours to download the objects ;-)