Include guidance on setting up cleaning rules for the blob store

[mary-ext]

the PDS temporarily stores blobs in a holding directory while it waits for a record that it can be linked to

this is fine, but both the S3 and disk blob store implementation does not have a periodic cleanup task that'd purge blobs that has been uploaded but never used, the holding directory would continue to grow and grow consuming a lot of disk space

I think we should point this out in the README, S3 rules should be easy to set up but I think we might need an example systemd timer for disk blob store?

[ThisIsMissEm]

It should be cleaning them up after each repo write: https://github.com/bluesky-social/atproto/blob/main/packages/pds/src/actor-store/blob/transactor.ts#L187

[mary-ext]

deleteDereferencedBlobs only cleans up blobs that were linked to a record, it doesn't address blobs that were never linked to a record to begin with (say the user tried to post, the blob upload made it through but the user cancelled the post for whatever reason, now that blob resides forever in holding)

[ThisIsMissEm]

That sounds like a bug then.

[mary-ext]

uploadBlob does insert the temp blob into the blob table, I think this is intentionally made to be as opaque as possible though as there's nothing in the blob store for listing, removing and checking whether a temporary blob exists

it does seem that the intended solution here would've been to use S3's cleaning rules, but we would still need an equivalent for PDS set up with disk blob store

[mary-ext]

for R2 I was able to configure it like so

[axfelix]

uploadBlob does insert the temp blob into the blob table, I think this is intentionally made to be as opaque as possible though as there's nothing in the blob store for listing, removing and checking whether a temporary blob exists

it does seem that the intended solution here would've been to use S3's cleaning rules, but we would still need an equivalent for PDS set up with disk blob store

I believe this is correct — could use an example cron or systemd here.

[fatfingers23]

Running full backups first. But I'm working on one that uses rclone and will test and report back. With rclone would work no matter the backend once setup. Does save selfhosted almost 5 gigs

[fatfingers23]

rclone should work with either backend, but is an install and may just want a bash script for local file system. But what I did

1. Install rclone
2. Set up S3 config if needed. I have some directions here or rclone has some here. Local file system doesn't need anything else
3. rclone size rclone_remote:bucket_name/tmp to get the size of the tmp folder in S3. rclone size /pds/blocks/tempt
4. Same format as above to delete everything over a day old rclone delete rclone_remote:bucket_name/tmp --min-age 1d -n the -n is for a dry run, does not delete anything just shows what it would. Can remove to delete files. Can also do --min-age 15m to delete all older than 15mins than a day, but a day was fine for me. Also looks like for the local file system the folder is tempt not tmp

may not be the end all solution, but hope it helps!