pointyhat is overloaded to do too many things. The plan is to instantiate multiple machines to divide up these tasks.
pointyhat's tasks
-
Creation of packages for -stable branches for uploading. In general we do not often update src. These packages need to be made in as secure an environment as possible.
-
Creation of packages for -current for uploading. In general we do not often update src. These packages also need to be made in as secure an environment as possible.
-
Creation of packages for release for uploading. These need a particular src build. These packages need to be made in as secure an environment as possible.
-
Regression tests of infrastructure changes. These packages do not need to be uploaded.
-
Regression tests for sweeping ports changes. These packages do not need to be uploaded.
-
Regression tests for major src changes. These packages do not need to be uploaded.
-
Stress-testing of -current/zfs on pointyhat itself. This is often incompatible with the need to create packages for upload.
-
Web-server for current build and error logs.
-
Web-server for historical build and error logs.
-
dinoex' check-errlogs script (cronjob).
-
fenner's (broken) portsurvey script (cronjob).
-
Various periodic QA scripts. (?)
meta-issues
-
How do we divide up the package builds between multiple machines?
-
How do we make sure that the same build directory names are not generated by two different machines?
-
Where do the http archives get served from? How do we synchronize them?
-
How do we refactor the links so that historical data and current work can live on different disks?
-
In all cases,
-full
should be renamed-previous
. -
The current code that assumes the rotation of
-latest
into-full
(i.e.previous
) needs to be parameterized. -
We currently overload
-exp
to also mean-release
. This is a bogus limitation and should go away.
current www layout
See this link.
current disk layout
Not yet completed. Note: it's not the same as the above.
Last modified: Tue Apr 13 14:41:22 UTC 2010