On 06/22/2016 04:39 PM, Sam Varshavchik wrote:
Rick Stevens writes:
> On 06/22/2016 03:46 PM, Sam Varshavchik wrote:
>
> > Last time I checked, I was told that the full repo weighed in somewhere
> > north of 20 gigabytes.
>
> You have to have the content SOMEWHERE local, don't you? You don't have
> to mirror the whole shooting match (all arches, the baseline OS, etc.),
> just the x86_64 updates repos you're interested in. And with 1TB drives
I am not talking about the update repos. For system-upgrade I need to go
to the full repo.
> costing $80USD (and you only need one on your local repo server), this
> is an issue?
Disk space is not an issue. The issue is piss poor bandwidth for a
typical US broadband.
It took just a bit less than half hour to download the packages needed
for a full upgrade to F24. But multiply that by the number of machines
to upgrade to F24, and this adds up quickly.
The issue is not regular daily updates. I have that automated and
covered. A daily rsync of the updates directory to a local repo, with
all machines pointing to it, and the regular updates repo turned off,
does the trick.
The issue is upgrading to a new release. There is no good way to
optimize the downloads in the same manner. rsyncing the entire 20 gig
full Fedora release (if it's still about 20 gigs), would take me about
ten hours.
> Downloading once to a local machine and having the other machines on the
> LAN use it as their repo or setting up a caching proxy like squid and
That's one option, sure. I don't normally need squid, for my regular
daily needs.
But I'll try the trick of rsyncing /var/lib/dnf/system-upgrade, first.
This is apparently where dnf system-upgrade drops all of the downloaded
packages.
If that's going to be sufficient, this will be fine for something that
needs to be done twice a year. If not, I'll probably find the time to
get squid up and running, in the next six months.
> runs a minimal Fedora server 23 (at the moment). It is a full repo for
> Fedora 21-23 (32- and 64-bit), CentOS 6 and 7 (both 32- and 64-bit) and
> serves over 300 client machines without even breaking a sweat. Hardware
> total: about $200USD. Took less than a day to set up. Polls the repos
> once a day to pick up updates. Simple.
Daily updates is not the issue. The "dnf system-upgrade" reference in
the subject line does not refer to daily updates.
Ah, OK, yes, I missed that bit. But, as you said, it's only twice a
year and so setting up something to rsync the whole repo down in the
background when you're deciding to upgrade a batch of machines may not
be such an onerous thing after all. After all, 30 minutes/machine times
20 machines = 10 hours. If you have <=20 machines to upgrade, do it the
way you're doing it. If it's >20 machines, then pulling the entire
repo down would be easier.
I have multiple 10Gbps pipes available to me at our data center, so I
don't think about bandwidth issues per se very often. Sorry if I seemed
callous about it.
----------------------------------------------------------------------
- Rick Stevens, Systems Engineer, AllDigital ricks(a)alldigital.com -
- AIM/Skype: therps2 ICQ: 226437340 Yahoo: origrps2 -
- -
- After a shooting spree, they always want to take the guns away -
- from the people who didn't do it. -
- -- William S.Burroughs -
----------------------------------------------------------------------