SPIKE: Investigate why xz packaging on RPM platforms has poor performance
Summary
In !7128 (merged), we discovered that XZ package compression on RPM platforms takes approximately 2-3 times longer than on Debian platforms. This effect is consistent across all RPM platforms irrespective of architecture. In response, we disabled XZ packaging for RPM. We would like the ~300MB savings from using XZ if we can get it, so we should find out why the performance difference exists and fix it if we can.
Steps to reproduce
- Set
compression_type :xz
for RPM packages inconfig/projects/gitlab.rb
- Run a pipeline with the change and another without it
- Observe the
Packaging time:
line in a task for each pipeline and compare them
What is the current unexpected behavior?
For the same RPM distribution, the packaging time with XZ compression will be 2-3 times higher.
What is the expected correct behavior?
The packaging times with and without xz
should be within a few percent of each other.