On Tue, 6 Jan 2026 07:38:19 GMT, Aleksey Shipilev <[email protected]> wrote:

>> I have noticed that Linux x86_64 fastdebug build jobs started to fail in GHA 
>> with no free space left on device. Instrumenting GHA runs shows we have 
>> about 17G of free space on those runners. 
>> 
>> 
>> Filesystem  Size  Used  Avail Use% Mounted on
>> /dev/root    72G   56G   17G   78% /
>> 
>> 
>> [GHA 
>> docs](https://docs.github.com/en/actions/reference/runners/github-hosted-runners#standard-github-hosted-runners-for-public-repositories)
>>  say there is 14G of guaranteed disk space on runners. This is a bit beyond 
>> what our fastdebug builds need. My local build says disabling debug symbols 
>> bundling reduces the space requirement from 16G -> 11G, without regressing 
>> tests.
>> 
>> Testing:
>>  - [x] Ad-hoc JVM crash to verify that stack trace generation still works
>>  - [x] GHA
>
> Thank you! I am integrating now to unbreak GHA.

> @shipilev I'm a bit late to the game here but maybe we should try 
> --with-external-symbols-in-bundles=public on Windows in order to get file and 
> method names in case of crashes. Not sure whether it'll be again too much of 
> disk space needed, though...

Oh. Yes, please try that? Really we wanted to improve Linux side of things, 
that is where we are scratching the runner space limits. I have not looked at 
Windows side at all; so please experiment and see how bad is it? I also have 
this improvement for Linux/MacOS: https://github.com/openjdk/jdk/pull/29056 -- 
again, does not help Windows, since I cannot find a useful/relevant MSVC toggle 
there.

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

PR Comment: https://git.openjdk.org/jdk/pull/29030#issuecomment-3713766995

Reply via email to