While it's running createrepo (and createrepo_c) create various temp files and directories. They are supposed to clean up after themselves, even if an error occurs, but certain kinds of failures may prevent them from doing so (for example being killed with an unhandled signal). If a createrepo child process fails, Beaker should make an effort to clean up any temporary files/directories so that the system is not left in a state where new createrepo processes refuse to run (and thus Beaker cannot accept new task uploads).
Most important one is the /var/www/beaker/rpms/.repodata directory which is created by createrepo_c. We should check for other temp files/directories, and also check what naming scheme createrepo uses as well.
I still haven't figured out what is causing createrepo_c to be killed with SIGHUP but this should be an acceptable bandaid solution.
On Gerrit: http://gerrit.beaker-project.org/3121
This bug fix has been applied to the release-0.16 branch, however we have elected not to do another maintenance release of the 0.16.x series. This fix will be included in 0.17.0 instead.
Suggested steps for verification: 1. Upload a task to Beaker 2. On the Beaker server, wait for the createrepo_c process to be started and then: killall -9 createrepo_c 3. Task upload should fail with an XML-RPC fault. 4. Try task upload again, it should succeed.
Beaker 0.17.0 has been released.