#7008 Old deltarpms are being thrown away on each new compose
Closed: Upstream 5 years ago Opened 5 years ago by jdieter.

It seems that all old deltarpms are being thrown away on each new compose rather than just the obsolete ones. This means that, unless users update after every compose, they don't get the benefit of the deltarpms.

I suspect this is a simple configuration file mistake, possibly after adding the Everything directory to updates in F28.


This seems to me to be a pungi behavior/bug. Would you be willing to file upstream with pungi?

https://pagure.io/pungi/issues

I think the issue is that a pungi updates compose doesn't know anything about other composes currently. There would have to be some way to copy the drpms from the previous compose (with some sort of time limit), but pungi developers would know more...

I generally do update after every compose, but even so deltas typically don't save more than 10-20% at best anymore, which means they're probably not worth the trouble to create - anyone who can manage with them, can manage without. They used to save around 75% on average. Something else may be going on as well. It seems that most packages don't have deltas at all.

I think the issue is that a pungi updates compose doesn't know anything about other composes currently. There would have to be some way to copy the drpms from the previous compose (with some sort of time limit), but pungi developers would know more...

Correct, it looks for the previous compose but only to generate deltas against it, not to copy anything from there. I would be mildly opposed to doing that in Pungi, since that is really more into shipping things than creating a compose. I don't know what exactly Bodhi is doing once the compose finishes, but it could rsync the old deltas over.

It seems that most packages don't have deltas at all.

I think this is a consequence of the delta generation only against latest updates compose: it only contains updated packages, so a delta will be generated only if a package is updated in two consecutive updates composes. This would definitely be worth fixing.

I think the issue is that a pungi updates compose doesn't know anything about other composes currently. There would have to be some way to copy the drpms from the previous compose (with some sort of time limit), but pungi developers would know more...

Correct, it looks for the previous compose but only to generate deltas against it, not to copy anything from there. I would be mildly opposed to doing that in Pungi, since that is really more into shipping things than creating a compose. I don't know what exactly Bodhi is doing once the compose finishes, but it could rsync the old deltas over.

So where would you like me to open a ticket, against Pungi or Bodhi?

So where would you like me to open a ticket, against Pungi or Bodhi?

I would prefer starting in Bodhi to see what could be done there.

I think this is a consequence of the delta generation only against latest updates compose: it only contains updated packages, so a delta will be generated only if a package is updated in two consecutive updates composes. This would definitely be worth fixing.

This is actually wrong, the fXY-updates tag has has all packages that were ever updated, not just the new updates. What I wrote applies only to testing.

Just to make sure we're all on the same page, this did actually used to work. The last Fedora 24 updates (dated 2017-08-07) have the full collection of deltarpms, while the last Fedora 25 updates (roughly six months later) only have the last compose, which means something broke sometime between a year ago and six months ago.

I think it was mash that used to actually copy the deltarpms across. If we stopped using mash in that time period, I'll happily file a bug with Bodhi, but, if not, it might be worth figuring out why this used to work and see if we can easily get it to work again.

It seems that most packages don't have deltas at all.

In addition to the not saving old deltarpms, it seems we're only delta-ing against f28-updates and not ga.

I think it was mash that used to actually copy the deltarpms across. If we stopped using mash in that time period, I'll happily file a bug with Bodhi, but, if not, it might be worth figuring out why this used to work and see if we can easily get it to work again.

Yup, mash did that. And indeed we did stop using mash and replaced it with Pungi.

Yup, mash did that. And indeed we did stop using mash and replaced it with Pungi.

Ok, I've filed ticket 2406 with bodhi

Metadata Update from @kevin:
- Issue close_status updated to: Upstream
- Issue status updated to: Closed (was: Open)

5 years ago

Login to comment on this ticket.

Metadata