Since the change on fedorapeople was a success, we need to make this change on the builders as well.
The problem is that when we create rawhide each night, we have random builders create a chroot in order to run pungi in. The chroot is generated from the freshly made rawhide repos, which have the newer sha256 sum in them. We need a newer yum and the python-hashlib build in order for the host yum to be able to process that metadata and generate the chroot.
On 2009-01-28 01:51:46 PM, Jesse Keating wrote:
Since the change on fedorapeople was a success, we need to make this change on the builders as well.
The problem is that when we create rawhide each night, we have random builders create a chroot in order to run pungi in. The chroot is generated from the freshly made rawhide repos, which have the newer sha256 sum in them. We need a newer yum and the python-hashlib build in order for the host yum to be able to process that metadata and generate the chroot.
+1
Thanks, Ricky
On Wed, 28 Jan 2009, Jesse Keating wrote:
Since the change on fedorapeople was a success, we need to make this change on the builders as well.
The problem is that when we create rawhide each night, we have random builders create a chroot in order to run pungi in. The chroot is generated from the freshly made rawhide repos, which have the newer sha256 sum in them. We need a newer yum and the python-hashlib build in order for the host yum to be able to process that metadata and generate the chroot.
+1 low risk, easy to revert.
Side question: Do we know what we're looking for if this breaks something?
-Mike
On Wed, 2009-01-28 at 15:59 -0600, Mike McGrath wrote:
On Wed, 28 Jan 2009, Jesse Keating wrote:
Since the change on fedorapeople was a success, we need to make this change on the builders as well.
The problem is that when we create rawhide each night, we have random builders create a chroot in order to run pungi in. The chroot is generated from the freshly made rawhide repos, which have the newer sha256 sum in them. We need a newer yum and the python-hashlib build in order for the host yum to be able to process that metadata and generate the chroot.
+1 low risk, easy to revert.
Side question: Do we know what we're looking for if this breaks something?
the crowd of people screaming REGRESSION in rhel 5.3 bug reports?
More or less anything completely hurky happening in mock is what to look out for.
-sv
Jesse Keating wrote:
Since the change on fedorapeople was a success, we need to make this change on the builders as well.
The problem is that when we create rawhide each night, we have random builders create a chroot in order to run pungi in. The chroot is generated from the freshly made rawhide repos, which have the newer sha256 sum in them. We need a newer yum and the python-hashlib build in order for the host yum to be able to process that metadata and generate the chroot.
So you need both yum and python-hashlib upgrades?
-Toshio
On Wed, 2009-01-28 at 14:07 -0800, Toshio Kuratomi wrote:
Jesse Keating wrote:
Since the change on fedorapeople was a success, we need to make this change on the builders as well.
The problem is that when we create rawhide each night, we have random builders create a chroot in order to run pungi in. The chroot is generated from the freshly made rawhide repos, which have the newer sha256 sum in them. We need a newer yum and the python-hashlib build in order for the host yum to be able to process that metadata and generate the chroot.
So you need both yum and python-hashlib upgrades?
Yes. sha256 support lives in python-hashlib, and the updated yum handles such checksums, using the updated hashlib.
On Wed, 2009-01-28 at 14:07 -0800, Toshio Kuratomi wrote:
Jesse Keating wrote:
Since the change on fedorapeople was a success, we need to make this change on the builders as well.
The problem is that when we create rawhide each night, we have random builders create a chroot in order to run pungi in. The chroot is generated from the freshly made rawhide repos, which have the newer sha256 sum in them. We need a newer yum and the python-hashlib build in order for the host yum to be able to process that metadata and generate the chroot.
So you need both yum and python-hashlib upgrades?
yum needs to update to the version from rhel 5.3 - 3.2.19<mumble>
python-hashlib doesn't need an update - it just needs to be installed.
-sv
On Wed, 2009-01-28 at 13:51 -0800, Jesse Keating wrote:
Since the change on fedorapeople was a success, we need to make this change on the builders as well.
This change did cause some fallout.
The previous yum would not consider scriptlet (like %post) errors to be fatal. The new yum does. This matters because bash apparently has had a %post failure since December, both F11 and F10 builds it seems. Once the new yum was in place, no buildroot for F10 or F11 could init. This is quite bad.
Instead of rolling back yum, I'm attempting to replace bash's %post that was in bash, and instead writing it in native lua. This will allow bash to complete its %post without any external deps. If this works, I'll create a F10 bash for dist-f10-override so that buildroots there will init as well.
To get the build done, I've reverted yum on ppc10 and x86-2, and disabled every other builder. I'll re-enable things once we're able to reliably create buildroots.
On Wed, 2009-01-28 at 22:44 -0800, Jesse Keating wrote:
Instead of rolling back yum, I'm attempting to replace bash's %post that was in bash, and instead writing it in native lua. This will allow bash to complete its %post without any external deps. If this works, I'll create a F10 bash for dist-f10-override so that buildroots there will init as well.
To get the build done, I've reverted yum on ppc10 and x86-2, and disabled every other builder. I'll re-enable things once we're able to reliably create buildroots.
This was successful. I've got new f10 and f11 repos with a fixed bash, and all builders have been re-enabled with the new yum.
On Thu, 2009-01-29 at 00:05 -0800, Jesse Keating wrote:
On Wed, 2009-01-28 at 22:44 -0800, Jesse Keating wrote:
Instead of rolling back yum, I'm attempting to replace bash's %post that was in bash, and instead writing it in native lua. This will allow bash to complete its %post without any external deps. If this works, I'll create a F10 bash for dist-f10-override so that buildroots there will init as well.
To get the build done, I've reverted yum on ppc10 and x86-2, and disabled every other builder. I'll re-enable things once we're able to reliably create buildroots.
This was successful. I've got new f10 and f11 repos with a fixed bash, and all builders have been re-enabled with the new yum.
Thanks, Jesse. -sv
infrastructure@lists.fedoraproject.org
-
Jesse Keating -
Mike McGrath -
Ricky Zhou -
Seth Vidal -
Toshio Kuratomi