On 12/15/21 23:08, Chris Murphy wrote:
On Wed, Dec 15, 2021 at 4:09 AM Panu Matilainen <pmati...@redhat.com> wrote:
On 12/14/21 22:59, Colin Walters wrote:
On Thu, Dec 9, 2021, at 10:11 AM, Chris Murphy wrote:
The change is not so simple. It is not only the movement of files from one
location to another one. We store more types of data in that location - history
database (sqlite), module failsafe data (yamls). In future we will store system
state data (toml). Data is not only modified after RPM transactions but also by
module and mark commands. What I want to say is that the change will be painful
but in the proposal there are limited benefits.
When the rpmdb was bdb and not extensible, I understand the idea of dnf having
its own separate database. But I don't understand why there are two separate
sqlite databases now.
The rpmdb is not hard-wired to sqlite. Also, whatever the format used
it's strictly private to rpm except what's exposed through librpm APIs.
So it's certainly NOT extensible in the sense that anybody wanting to
put something package related there is free to do so.
Anyways, a lot going on here and rather than point by point I'll say basically:
It'd be great from the rpm-ostree PoV to have a shared consistent place for the
rpmdb. Last time this came up I think Panu really wanted to get the sqlite
change out first, which - fair. But now that's done, so are there any other
blockers?
Just the pain and misery of dealing with a fundamental change that has
zero benefit to rpm itself. It'll be a long, long, long time before
/var/lib/rpm is phased out of existence for good, and in the meanwhile
us rpm f get to answer all the questions and complaints about it. I'm
not looking forward to that.
The alternative to the proposal to switch rpmdb from /var to /usr, is
keeping it where it is. And provide matching between /usr and
/var/lib/rpm generations (or versions) by separate subvolume on Btrfs,
e.g. "varlibrpm" mounted at /var/lib/rpm. The separation is very cheap
on Btrfs, perhaps 200 KiB, because it's just a file btree. The cost is
a little more than a directory. And the snapshots only consume space
representing the difference between the snapshots.
However, for other implementations, such as and separate file system
volumes for LVM+ext4 LVM+XFS or Stratis, it's not as cheap. Each
snapshot includes the file system journal, which varies in size.
Storage cost could be anywhere from 8 - 800 MiB per snapshot.
I'm not sure if the alternative is that great either, considering
movement is in progress to put rpmdb in /usr already.
My grumbles aren't so much about rpmdb in /usr as a concept, it's the
move of a very very long standing "well known location". Oh, Colin
actually asked for blockers and I didn't mean this is a blocker, just an
unpleasantness left to deal with. Sorry about that.
It'd minimize the pain somewhat if the db location was left at
/var/lib/rpm on rpm level but /var/lib/rpm being a symlink pointing
inside /usr until all active Fedora versions had this, and at that point
changing the rpm configuration would make no difference to anybody
(within Fedora). But such a construct almost certainly wont survive
rpmdb rebuilds so it's probably not practical.
- Panu -
--
Chris Murphy
_______________________________________________
Rpm-ecosystem mailing list
Rpm-ecosystem@lists.rpm.org
http://lists.rpm.org/mailman/listinfo/rpm-ecosystem
