On Mon, 27 Jul 2026 15:56:34 GMT, Alan Bateman <[email protected]> wrote:

> > We don't know what configurations are built & tested at Oracle or other 
> > vendors but in our case where we do build and test debug builds, we see 
> > largely increased disk consumption which will be reduced/mitigated by that 
> > change.
> 
> My concern is more about changing all the jlink tests. The test for, e.g. 
> including locales, shouldn't be using .--strip-debug. Same thing for many 
> other tests.
> 
> The update the plugin is fine but if disk space during test runs is an issue 
> it would be better to have the tests do cleanup and delete the generated 
> run-time images when the test passes. If they are generated into the jtreg 
> scratch directory then they should be cleaned up already.

The problem is that we execute the jtreg tests with agents and a concurrency of 
16 - this causes up to 16 concurrent running jlink tests. When each test 
creates an image of 1.3 GB that would lead to 21 GB of peak disk usage - but 
several tests need to create more than one image at runtime.

At the moment I am in the process to check the different versions of this 
patch, with and with-out `--strip-debug `. I must admit that I also got a bit 
lost on the way by trying to find out why the debuginfo files in our nightly 
builds are a lot smaller compared to local builds I do while testing. The 
builds run with nearly the same configure calls, but the libjvm.debuginfo 
varies from 0.7 GB to 1.9GB. Currently I am trying to figure out the reason for 
this - my first finding is that adding  `--disable-precompiled-headers `caused 
the libjvm.debuginfo file to shrink from 1.9 GB to 1.3 GB

-------------

PR Comment: https://git.openjdk.org/jdk/pull/32009#issuecomment-5093763102

Reply via email to