Re: Fedora distribution build times
Tomasz Torcz wrote: On Sat, Jan 15, 2011 at 07:38:21AM -0800, John Reiser wrote: This is a full rebuild of all 10k packages. Due to the RemoveSUID feature in Fedora 15, mock can no longer use tmpfs buildroots, meaning this run takes 4 days instead of 1. Progress? A slowdown by a factor of 4 is a high price to pay for the impact of RemoveSUID. I'd rather pay at most 30%, and not a factor of 4. That's the extreme corner case, caused by bug in tmpfs (lack of filecaps?). Upstream kernel bug, I would say. Which filesystems support filecaps? Will the RemoveSUID feature reduce the number of filesystem usable in F15? -- Roberto Ragusamail at robertoragusa.it -- devel mailing list devel@lists.fedoraproject.org https://admin.fedoraproject.org/mailman/listinfo/devel
Re: Fedora distribution build times
On Sun, 16 Jan 2011 10:14:33 +0100 Roberto Ragusa m...@robertoragusa.it wrote: Tomasz Torcz wrote: On Sat, Jan 15, 2011 at 07:38:21AM -0800, John Reiser wrote: This is a full rebuild of all 10k packages. Due to the RemoveSUID feature in Fedora 15, mock can no longer use tmpfs buildroots, meaning this run takes 4 days instead of 1. Progress? A slowdown by a factor of 4 is a high price to pay for the impact of RemoveSUID. I'd rather pay at most 30%, and not a factor of 4. That's the extreme corner case, caused by bug in tmpfs (lack of filecaps?). Upstream kernel bug, I would say. Which filesystems support filecaps? I don't know if there is a list anywhere... all the main ones do: ext*, btrfs, xfs, etc. Will the RemoveSUID feature reduce the number of filesystem usable in F15? Depends on what you mean by usable. Usable as filesystems that contain packaged files? See also: http://lkml.indiana.edu/hypermail/linux/kernel/1101.1/01210.html Patch submitted for upstream feedback that adds caps to tmpfs. This should hopefully get into the mainstream/fedora kernels soon. If it doesn't, we will probibly revert the feature. kevin signature.asc Description: PGP signature -- devel mailing list devel@lists.fedoraproject.org https://admin.fedoraproject.org/mailman/listinfo/devel
Re: Fedora distribution build times
On 01/16/2011 07:57 PM, Kevin Fenzi wrote: On Sun, 16 Jan 2011 10:14:33 +0100 Roberto Ragusa m...@robertoragusa.it wrote: Which filesystems support filecaps? I don't know if there is a list anywhere... all the main ones do: ext*, btrfs, xfs, etc. What about reiserfs? I've been using it everywhere for years. Will the RemoveSUID feature reduce the number of filesystem usable in F15? Depends on what you mean by usable. Usable as filesystems that contain packaged files? Usable as root partition, for example. -- Roberto Ragusamail at robertoragusa.it -- devel mailing list devel@lists.fedoraproject.org https://admin.fedoraproject.org/mailman/listinfo/devel
Re: Fedora distribution build times
Matt Domsch wrote: On Wed, Jan 12, 2011 at 12:24:41PM +0530, Shakthi Kannan wrote: Greetings! I would like to know as to how long it takes to build a complete Fedora distribution, say i686, from .src.rpms. Are there any statistics available or is it possible to provide any info on the hardware specifications (RAM, number of processors, kernel used etc.), time it takes to build the complete Fedora distribution, and other build related statistics? It took my build system 96 hours to build all of rawhide (10k packages) for both x86_64 and i386. Builders are 10 Dell PowerEdge 1955 servers, each with 2 sockets 3GHz Xeon 5160 CPUs (4 cores each), 8GB RAM. Builders running Fedora 14. Interesting. So it took 2x4x10x96=7680 corehours to build 2 packages, at an average of 7680*60/2=23 coreminutes/package. And we get that on average one package takes 6 minutes on a 4 core machine. That makes sense. -- Roberto Ragusamail at robertoragusa.it -- devel mailing list devel@lists.fedoraproject.org https://admin.fedoraproject.org/mailman/listinfo/devel
Re: Fedora distribution build times
Roberto Ragusa wrote: Matt Domsch wrote: It took my build system 96 hours to build all of rawhide (10k packages) for both x86_64 and i386. Builders are 10 Dell PowerEdge 1955 servers, each with 2 sockets 3GHz Xeon 5160 CPUs (4 cores each), 8GB RAM. Builders running Fedora 14. Perhaps it was Fedora 15 (and not Fedora 14) that took 4 days? See below. So it took 2x4x10x96=7680 corehours to build 2 packages, at an average of 7680*60/2=23 coreminutes/package. And we get that on average one package takes 6 minutes on a 4 core machine. That makes sense. In this message: Fedora rawhide FTBFS status 2010-12-01 x86_64 Mon Dec 6 15:25:22 UTC 2010 http://lists.fedoraproject.org/pipermail/devel/2010-December/146682.html Matt says: This is a full rebuild of all 10k packages. Due to the RemoveSUID feature in Fedora 15, mock can no longer use tmpfs buildroots, meaning this run takes 4 days instead of 1. Progress? A slowdown by a factor of 4 is a high price to pay for the impact of RemoveSUID. I'd rather pay at most 30%, and not a factor of 4. -- -- devel mailing list devel@lists.fedoraproject.org https://admin.fedoraproject.org/mailman/listinfo/devel
Re: Fedora distribution build times
On Sat, Jan 15, 2011 at 07:38:21AM -0800, John Reiser wrote: This is a full rebuild of all 10k packages. Due to the RemoveSUID feature in Fedora 15, mock can no longer use tmpfs buildroots, meaning this run takes 4 days instead of 1. Progress? A slowdown by a factor of 4 is a high price to pay for the impact of RemoveSUID. I'd rather pay at most 30%, and not a factor of 4. That's the extreme corner case, caused by bug in tmpfs (lack of filecaps?). Upstream kernel bug, I would say. -- Tomasz Torcz God, root, what's the difference? xmpp: zdzich...@chrome.pl God is more forgiving. -- devel mailing list devel@lists.fedoraproject.org https://admin.fedoraproject.org/mailman/listinfo/devel
Re: Fedora distribution build times
On 01/15/2011 10:38 AM, John Reiser wrote: Perhaps it was Fedora 15 (and not Fedora 14) that took 4 days? See below. So it took 2x4x10x96=7680 corehours to build 2 packages, at an average of 7680*60/2=23 coreminutes/package. And we get that on average one package takes 6 minutes on a 4 core machine. That makes sense. Is it possible to break down how much time is in compiling versus packaging versus whatever else is involved ? -- devel mailing list devel@lists.fedoraproject.org https://admin.fedoraproject.org/mailman/listinfo/devel
Re: Fedora distribution build times
Genes MailLists wrote: Is it possible to break down how much time is in compiling versus packaging versus whatever else is involved ? What filesystem was used instead of tmpfs (which does not support the Capabilities that are required because of RemoveSUID in Fedora 15)? Perhaps ext2 (no journal) would save 30% elapsed time in contrast to ext3, for example. -- -- devel mailing list devel@lists.fedoraproject.org https://admin.fedoraproject.org/mailman/listinfo/devel
Re: Fedora distribution build times
On Sat, Jan 15, 2011 at 11:14:42 -0500, Genes MailLists li...@sapience.com wrote: On 01/15/2011 10:38 AM, John Reiser wrote: Is it possible to break down how much time is in compiling versus packaging versus whatever else is involved ? The time to run mock is significant (even without the setuid issue) and if you don't need to check that packages have the correct build requires declarations you could save significant time by cutting that step out. -- devel mailing list devel@lists.fedoraproject.org https://admin.fedoraproject.org/mailman/listinfo/devel
Re: Fedora distribution build times
On Fri, 2011-01-14 at 16:51 -0600, Matt Domsch wrote: It took my build system 96 hours to build all of rawhide (10k packages) for both x86_64 and i386. Builders are 10 Dell PowerEdge 1955 servers, each with 2 sockets 3GHz Xeon 5160 CPUs (4 cores each), 8GB RAM. Builders running Fedora 14. OOI, do you sequentially build one package at a time on a builder, or do you do parallel builds? Jon. -- devel mailing list devel@lists.fedoraproject.org https://admin.fedoraproject.org/mailman/listinfo/devel
Re: Fedora distribution build times
On Sat, Jan 15, 2011 at 03:52:34PM -0500, Jon Masters wrote:
On Fri, 2011-01-14 at 16:51 -0600, Matt Domsch wrote:
It took my build system 96 hours to build all of rawhide (10k
packages) for both x86_64 and i386. Builders are 10 Dell PowerEdge
1955 servers, each with 2 sockets 3GHz Xeon 5160 CPUs (4 cores each),
8GB RAM. Builders running Fedora 14.
OOI, do you sequentially build one package at a time on a builder, or do
you do parallel builds?
Each builder runs 2 build processes for each of i386 and x86_64 in
parallel, so 4 parallel builds per server. Each package may be using
make %{_smp_flags} which would then overcommit the CPUs. But in all,
it's disk-bound, not CPU-bound, especially now that I can't use tmpfs.
The builders have ext4 (Fedora 14 default) where the mock chroots
live. When the runs start, each builder creates the basic chroot, and
mock caches that and unpacks it at the start of each package build, to
save some time. The non-base buildrequires packages get fetched from
a local server over 1GbE, so that could become a bottleneck, but isn't
terribly limiting.
--
Matt Domsch
Technology Strategist
Dell | Office of the CTO
--
devel mailing list
devel@lists.fedoraproject.org
https://admin.fedoraproject.org/mailman/listinfo/devel
Re: Fedora distribution build times
On Sat, Jan 15, 2011 at 11:14:42AM -0500, Genes MailLists wrote: On 01/15/2011 10:38 AM, John Reiser wrote: Perhaps it was Fedora 15 (and not Fedora 14) that took 4 days? See below. So it took 2x4x10x96=7680 corehours to build 2 packages, at an average of 7680*60/2=23 coreminutes/package. And we get that on average one package takes 6 minutes on a 4 core machine. That makes sense. Is it possible to break down how much time is in compiling versus packaging versus whatever else is involved ? patches to mock to produce such would be welcome I'm sure. For my part, I run '/usr/bin/time mock ...' on each package built, and those times are recorded alongside the output RPMs, for the curious: [vim-7.3.069-1.fc15.src.rpm]$ cat time.log real 253.96 user 211.30 sys 32.11 -- Matt Domsch Technology Strategist Dell | Office of the CTO -- devel mailing list devel@lists.fedoraproject.org https://admin.fedoraproject.org/mailman/listinfo/devel
Re: Fedora distribution build times
On Sat, Jan 15, 2011 at 04:49:37PM +0100, Tomasz Torcz wrote: A slowdown by a factor of 4 is a high price to pay for the impact of RemoveSUID. I'd rather pay at most 30%, and not a factor of 4. That's the extreme corner case, caused by bug in tmpfs (lack of filecaps?). Upstream kernel bug, I would say. Lack of a feature isn't a bug. -- Matthew Miller mat...@mattdm.org Senior Systems Architect -- Instructional Research Computing Services Harvard School of Engineering Applied Sciences -- devel mailing list devel@lists.fedoraproject.org https://admin.fedoraproject.org/mailman/listinfo/devel
Re: Fedora distribution build times
On Sat, 2011-01-15 at 19:54 -0500, Matthew Miller wrote: On Sat, Jan 15, 2011 at 04:49:37PM +0100, Tomasz Torcz wrote: A slowdown by a factor of 4 is a high price to pay for the impact of RemoveSUID. I'd rather pay at most 30%, and not a factor of 4. That's the extreme corner case, caused by bug in tmpfs (lack of filecaps?). Upstream kernel bug, I would say. Lack of a feature isn't a bug. Not to be an antagonist, but I still see no rational reason to remove setuid and replace it with this alternative implementation wholesale at this time. UNIX got away with setuid for a *long* time, and I'm sure we could have managed for another year (or 30) without a big switchover. Jon. -- devel mailing list devel@lists.fedoraproject.org https://admin.fedoraproject.org/mailman/listinfo/devel
