There is one long term plan that may help.

We've got a proposal to start signing every build with a Fedora Buildsystem key, and not re-sign each build for each release. This would reduce the number of duplicate rpms where only the signature header has changed. It requires more work on the signing server side and some other infrastructure, but that's where we're going.

That said, we're growing our package set at a very rapid pace, and I'm not sure how much we can save by garbage collecting builds from older releases (FC6/F7/F8/F9), nor if we want to remove those from the buildsystem. A storage growth plan is probably in order, as well as some discussions about how long we actually keep these things in the buildsystem.

Well, lets look at the following sizes:

1. Size of each release to date (specifically how much we'd free up by deleting it)
2. Size of duplicate information. I'm not even sure how to go about doing that.
3. Size of non fedora builds (EPEL, olpc, others?)
4. Size of all kernel builds

dennis, any other suggestions to look at?

the rate and quantity of packages we ship is growing exponentially. we are getting diminishing returns from garbage collection. I think we need to start planning upgrades and migration. we could slow down the growth if we decided that we delete old releases entirely from /mnt/koji. It would need careful planning but could be done.

I've created a script to prune tags of all but their latest build. I had noticed that garbage collection will completely ignore tags with locks on them, thus we were missing out on a lot of garbage collection. Thus far I've untagged nearly 20K builds, which should be picked up in the garbage collection over the next few weeks and clear out some space. This should help with our rate of growth as well, if we prune the tags as they get retired.

Within the next few days we should see a significant increase to the number of packages in the 'trashcan' tag. Right now its 1300~ but that should grow by many thousands. Those builds will live in trashcan for 4 weeks and then get deleted, which should save some measurable amount of space.

We are back down to 82%. I'll open another ticket if we hit 85% again.

{{{

Filesystem Size Used Avail Use% Mounted on
nfs1.fedora.phx.redhat.com:/mnt/koji
9.9T 7.7T 1.8T 82% /mnt/koji

}}}