= bug description = Scratch builds on staging Koji submitted by staging Koschei fail with traceback: "HTTPError: HTTP Error 404: Not Found", for example see http://koji.stg.fedoraproject.org/koji/taskinfo?taskID=90001370
= bug analysis = Koschei runs scratch builds from existing SRPMs, which in most cases are located in production Koji volume at /mnt/koji/vol/prod/. The problem is that this location is not exported through HTTP, so kojid on builders fails with HTTP 404 error.
= fix recommendation = Either of the following two solutions:
Add rewrite rule to kojiweb.conf {{{ RewriteRule ^/vol/prod/(.+) https://koji.fedoraproject.org/$1 [R=301,L] RewriteRule ^/vol/prod/packages/(.+) https://kojipkgs.fedoraproject.org/packages/$1 [R=301,L] }}} Pros: it should be faster - packages will be coming from squid cache at kojipkgs Cons: extra hosts entries may need to be added on some staging hosts so that they know how to connect to production koji and kojipkgs
Export contents of vol directory: {{{ Alias /work "/mnt/koji/vol/"
<Directory "/mnt/koji/vol/"> Options Indexes FollowSymLinks Require all granted </Directory> }}} Pros: simple configuration Cons: data is read directly from production NFS, which apparently is already under high load
I'd vote for option number 2, since it maintains more of the symmetry with the existing setup (i.e., it looks like staging koji has all of the real data). Although, I don't have a good sense of just how loaded our NFS setup is.
attachment 0001-Export-mnt-koji-vol-on-staging-Koji-through-HTTP.patch
Proposed patch for option nr 2 added.
Applied! https://infrastructure.fedoraproject.org/cgit/ansible.git/commit/?id=eb07c24d08023b8f2bfdc1d726583e4e0fedb2cd
Login to comment on this ticket.