while assessing what consumes space I spotted that some zarrs have huge number of loose objects, which seems to be packed or cleaned up with git gc call. But it seems that we do call git gc after syncing zarr: https://github.com/dandi/dandisets/blob/306e1ae3732e8f44fbf79ad4b44e9bb7bf07dc2b/tools/backups2datalad/zarr.py#L577
so it is not clear why we still have the case (may be failed syncs which never got git gc ran, only me doing git reset --hard?). I am still collecting stats across all zarrs under /tmp/dandizarrs-inodes.out but here is an example of one and effect of running git gc now:
dandi@drogon:/mnt/backup/dandi/dandizarrs/0cf3c734-4bb2-47d1-8e6a-528a3739af01$ du -scm --inodes .git/objects/; git gc; du -scm --inodes .git/objects/
232704 .git/objects/
232704 total
Enumerating objects: 541506, done.
Counting objects: 100% (541506/541506), done.
Delta compression using up to 4 threads
Compressing objects: 100% (469070/469070), done.
Writing objects: 100% (541506/541506), done.
Total 541506 (delta 224707), reused 156961 (delta 72419), pack-reused 0
Removing duplicate objects: 100% (256/256), done.
7 .git/objects/
7 total
while assessing what consumes space I spotted that some zarrs have huge number of loose objects, which seems to be packed or cleaned up with
git gccall. But it seems that we do callgit gcafter syncing zarr: https://github.com/dandi/dandisets/blob/306e1ae3732e8f44fbf79ad4b44e9bb7bf07dc2b/tools/backups2datalad/zarr.py#L577 so it is not clear why we still have the case (may be failed syncs which never gotgit gcran, only me doinggit reset --hard?). I am still collecting stats across all zarrs under/tmp/dandizarrs-inodes.outbut here is an example of one and effect of runninggit gcnow:so we went from 230k to 7 inodes in .git/objects.