Re: Signing server? (Re: Updates testing for F-11)
Am Freitag, den 12.06.2009, 08:48 -0700 schrieb Toshio Kuratomi: On 06/12/2009 08:14 AM, Christoph Wickert wrote: Am Freitag, den 12.06.2009, 05:34 +0200 schrieb Kevin Kofler: I don't see what it buys our users if they get one big update over 2 small ones. In most cases the biggest part (consuming time and cpu cycles) of the updates is not installing them but everything else like checking for new packages, downloading the metadata, This portion of the list is saved. This part happens in the background without user's notice, so this portion irrelevant. calculating dependencies, downloading the packages and running the transaction test. Especially for small updates this takes much more time than the actual rpm -U part. But this portion of your list is dependent on the size of the transaction so it isn't going to halve the time to go from two small updates to a single large update here. But this is the portion that requires user interaction, what happens between the first update notice and the message that all updates were installed. So from a users POV the time is reduced dramatically. I think it could even more than halve because scriptlets are only running once instead of several times. -Toshio Regards, Christoph -- fedora-devel-list mailing list fedora-devel-list@redhat.com https://www.redhat.com/mailman/listinfo/fedora-devel-list
Re: Signing server? (Re: Updates testing for F-11)
On Thu, Jun 11, 2009 at 08:54:19PM -0400, Josh Boyer wrote: On Thu, Jun 11, 2009 at 8:39 PM, Christoph Wickertchristoph.wick...@googlemail.com wrote: need it because things need to be predictable for package maintainers. Some updates are processed after a day, others not for two weeks. I'm a bit confused where your date is coming from. 2 weeks seems wrong lately. In fact, since I took over the push stuff, it's normally done daily or as often as the composes allow. Right now, the compose for f11-updates alone is 7-8 hours, so doing it daily often just doesn't work out. But 2 weeks seems wrong. Actually, mashing f11-updates last week took 11 hours. The entire push took about 22. luke -- fedora-devel-list mailing list fedora-devel-list@redhat.com https://www.redhat.com/mailman/listinfo/fedora-devel-list
Re: Signing server? (Re: Updates testing for F-11)
On Fri, 12 Jun 2009, Luke Macken wrote: On Thu, Jun 11, 2009 at 08:54:19PM -0400, Josh Boyer wrote: On Thu, Jun 11, 2009 at 8:39 PM, Christoph Wickertchristoph.wick...@googlemail.com wrote: need it because things need to be predictable for package maintainers. Some updates are processed after a day, others not for two weeks. I'm a bit confused where your date is coming from. 2 weeks seems wrong lately. In fact, since I took over the push stuff, it's normally done daily or as often as the composes allow. Right now, the compose for f11-updates alone is 7-8 hours, so doing it daily often just doesn't work out. But 2 weeks seems wrong. Actually, mashing f11-updates last week took 11 hours. The entire push took about 22. seems like that is a creating prestodelta.xml bug that we're working on now. -sv -- fedora-devel-list mailing list fedora-devel-list@redhat.com https://www.redhat.com/mailman/listinfo/fedora-devel-list
Re: Signing server? (Re: Updates testing for F-11)
Josh Boyer wrote: No. It simply is not possible. See my (and Luke's) email on how long a single push takes. Seth says the 22-hour run is a bug. If a run can be done in ~8 hours, that means an automated update procedure could do about 3 per day. But of course, if it takes one day, then let's do one per day. I'm not asking for the impossible. ;-) Kevin Kofler -- fedora-devel-list mailing list fedora-devel-list@redhat.com https://www.redhat.com/mailman/listinfo/fedora-devel-list
Re: Signing server? (Re: Updates testing for F-11)
On Fri, Jun 12, 2009 at 9:14 AM, Kevin Koflerkevin.kof...@chello.at wrote: Josh Boyer wrote: No. It simply is not possible. See my (and Luke's) email on how long a single push takes. Seth says the 22-hour run is a bug. If a run can be done in ~8 hours, that means an automated update procedure could do about 3 per day. Theoretically, yes. In reality, probably not. At least not until we get some of the other items in place that would even make this remotely possible. Also, we need to fix the failure paths, which make things indeterminably slower. But of course, if it takes one day, then let's do one per day. I'm not asking for the impossible. ;-) I WAS doing one a day. In fact, until a week ago I was doing them as often as possible. I plan to pick that back up on Monday, as I'm only able to be online for about 5 min at a time until then. josh -- fedora-devel-list mailing list fedora-devel-list@redhat.com https://www.redhat.com/mailman/listinfo/fedora-devel-list
Re: Signing server? (Re: Updates testing for F-11)
Am Freitag, den 12.06.2009, 05:34 +0200 schrieb Kevin Kofler: Christoph Wickert wrote: IMO this is something we should discuss on this list. We need to find a fine balance between pushing updates in time to make maintainers happy and not too many updates for the users. Maybe something like security/urgent updates daily, everything else once or twice a week. But this needs further discussion. I don't see what it buys our users if they get one big update over 2 small ones. In most cases the biggest part (consuming time and cpu cycles) of the updates is not installing them but everything else like checking for new packages, downloading the metadata, calculating dependencies, downloading the packages and running the transaction test. Especially for small updates this takes much more time than the actual rpm -U part. Plus, it'd require us to distinguish urgent vs. not urgent updates, and causes big issues with urgent updates accidentally depending on non-urgent ones. Good point. I did not think of that because my updates usually are at the end of a dependency chain and if not, I put all packages that require each other into one big update. Maintainers should be smart enough to do it that way. Of course it would cause problems for people waiting for other packager's updates, but IMO this is no difference to the current situation: If you don't ask rel-eng for a build root overwrite, you have to wait until the dependencies are pushed before you can build you packages. Of course it would require some automatic dependency check from bodhi, but this is something we should look at anyway as the recent vte update shows. Regards, Christoph -- fedora-devel-list mailing list fedora-devel-list@redhat.com https://www.redhat.com/mailman/listinfo/fedora-devel-list
Re: Signing server? (Re: Updates testing for F-11)
Christoph Wickert wrote: In most cases the biggest part (consuming time and cpu cycles) of the updates is not installing them but everything else like checking for new packages, downloading the metadata, calculating dependencies, downloading the packages and running the transaction test. Especially for small updates this takes much more time than the actual rpm -U part. If you include just the urgent stuff in daily updates, those will still be the same. Kevin Kofler -- fedora-devel-list mailing list fedora-devel-list@redhat.com https://www.redhat.com/mailman/listinfo/fedora-devel-list
Re: Signing server? (Re: Updates testing for F-11)
On 06/12/2009 08:14 AM, Christoph Wickert wrote: Am Freitag, den 12.06.2009, 05:34 +0200 schrieb Kevin Kofler: I don't see what it buys our users if they get one big update over 2 small ones. In most cases the biggest part (consuming time and cpu cycles) of the updates is not installing them but everything else like checking for new packages, downloading the metadata, This portion of the list is saved. calculating dependencies, downloading the packages and running the transaction test. Especially for small updates this takes much more time than the actual rpm -U part. But this portion of your list is dependent on the size of the transaction so it isn't going to halve the time to go from two small updates to a single large update here. -Toshio signature.asc Description: OpenPGP digital signature -- fedora-devel-list mailing list fedora-devel-list@redhat.com https://www.redhat.com/mailman/listinfo/fedora-devel-list
Re: Signing server? (Re: Updates testing for F-11)
Am Donnerstag, den 11.06.2009, 20:54 -0400 schrieb Josh Boyer: On Thu, Jun 11, 2009 at 8:39 PM, Christoph Wickertchristoph.wick...@googlemail.com wrote: Some updates are processed after a day, others not for two weeks. I'm a bit confused where your date is coming from. 2 weeks seems wrong lately. In fact, since I took over the push stuff, it's normally done daily or as often as the composes allow. Right now, the compose for f11-updates alone is 7-8 hours, so doing it daily often just doesn't work out. But 2 weeks seems wrong. OK, 12 days to be correct. You said it will not happen before the weekend, and if it happens on Sunday, two of my requests have reached 12 days. Also, signing server won't really help any of the above. Especially at release time this is annoying as some of the packages I submitted were bugfixes I wanted to be in the Xfce Spin. Another package was renamed, if it got pushed in time this would have happend smoothly between releases. So any chance we get a more reliable push mechanism? If by reliable you mean auto-pushed, maybe. Two options: More manpower or automatic pushes. For me it is important that I get a time frame that I can count on. But that would need auto-sign and an agreed upon schedule for updates and code. IMO this is something we should discuss on this list. We need to find a fine balance between pushing updates in time to make maintainers happy and not too many updates for the users. Maybe something like security/urgent updates daily, everything else once or twice a week. But this needs further discussion. And right now, we have a number of things that cause backend failures which necessitates manually fixing them. If you'd like to volunteer to code on bodhi and fix some of these, it would be most welcome. With the number of people currently working on it (2-3), it will be a while before we get to that point. I will see what I can do, but currently I'm swamped with other things, both Fedora (LinuxTag Berlin) and $dayjob. Also it will take some time for me to get involved in infrastructure, but if possible, I'll do it. josh Regards, Christoph -- fedora-devel-list mailing list fedora-devel-list@redhat.com https://www.redhat.com/mailman/listinfo/fedora-devel-list
