Our nightly builds currently create more than is strictly required. We may be able to reduce the number of build artifacts and our image sizes. Storage on the the autobuilder is always a concern and reducing disk consumed per build will help a lot.
We talked recently about removing RPMs, .debs, and some machine images from nightly builds after a short delay. This would not reduce build sizes but would reduce storage needs. Most machine images are never downloaded. I can pull stats on that. RPMs and .debs aren't needed after the package feeds are tested.
Michael, we had discussed a list of what is actually downloaded. Can we get that for daisy. That way we know where and what to trim.
I'm going to stop publishing RPMs and debs today. I doubt anyone will scream about it but if someone does, we now know about it. They're the least likely thing people will scream about. Halstead, any idea on download stats.
I quick look through the logs shows: 158 rpms and 38 debs downloaded in May from the releases And from the nightly builds: RPMs 2013-10: 554 2013-11: 618 2013-12: 238 2013-6: 208 2013-7: 611 2013-8: 311 2013-9: 231 2014-1: 209 2014-2: 172 2014-3: 249 2014-4: 347 2014-5: 204 2014-6: 131 Debs 2013-10: 1 2013-11: 65 2013-12: 20 2014-1: 135 2014-2: 189 2014-3: 6 2014-5: 5 2014-6: 1 It seems safe to stop producing RPMs and certainly safe to stop making Debs.
So, there isn't, at the time, much we can do for what we build out via the autobuilder. Some of that needs to be kicked over to QA and they need to go through and figure out what they absolutely need. Where we have been able to reduce things is by: Removing the rpm/deb directory in the release. Not needed. note: preserve links when copying the machines dir for x in `find . -name *.gz|grep -v p1022|grep core`; do rm -rf $x; done; for x in `find . -name *.bz2|grep -v p1022|grep core`; do rm -rf $x; done; for x in `find . -name *.ext3|grep -v qemu`; do rm -rf $x; done; This gets us down to 191GB from 255GB I'm going to close this for now. We're at a 20% reduction in size and while we can keep going, at some point we're going to impact endusers by yanking something they need. I'm also going to remove the dependency on 3060. What was implemented here just needs to be reflected in that work.
Hi, This feature cannot be tested by QA as it impact the public AB machines and the backend storage for which we don't have access. As Beth changed status to Resolved we will assume that this issue was fixed. Cheers, Ionut C