In "--fast" option for git annex get? it was indicated that git annex get --fast doesn't have any effect.
In an HPC context, users are frequently expected to use login (or dedicated data transfer) nodes for data transfer, and can get their sessions killed for excessive CPU use. For OpenNeuro, the high bandwidth between many HPC centers and S3 means that checksums can become the bottleneck in data transfer. I would like to be able to recommend something like:
git annex get -f s3-PUBLIC --fast --all
srun git annex fsck
Is this feasible?
Annex.fastsetting, but thegetcommand implementation is harder to grep for.Command/Get.hsWhat you can recommend, which works already, is:
As to adding this to --fast, I think some would be surprised if --fast allowed bad data to get into the repository. And commands like
git-annex copy --tothat do support --fast already use it to avoid round trip checks. It would not do to make --fast for those commands also avoid verification. Andgit-annex copyis very close togit-annex get, to the point thatgit-annex get --fromis the same asgit-annex copy --from.So, I think it's better to keep this a separate option, and the -c option I gave above works well enough I suppose.
With that said, you're the second person asking about this in an HPC context this week. I suspect maybe you and @mih were working on the same problem in asking about this? Anyway, since you both seemed to have difficulty finding the way to do this, maybe it would be worth making a dedicated option like
--no-verify.annex.verifyinman git-annex-config. Do those docs need refreshing?It's documented on the main git-annex man page with all other git configs. The git-annex-config man page is for configs that can be stored in the git-annex branch, which this one cannot be.