RFR: JDK-8319307: DCmds should not assert on truncation and should report truncation [v2]

[Thomas Stuefe]

On Mon, 27 Nov 2023 11:49:24 GMT, Kevin Walls <kevinw at openjdk.org> wrote:

> Thanks, JDK-8319055 does indeed need more investigation.
> 
> While nobody would want to _continually_ raise the buffer size, raising it is the easy one-liner with no side-effects on multiple diagnostic commands... Would it be 200MB? 500MB? 8-)
> 
> (sorry, should have written something like "multiple other files", not diagnostic commands)

Raising the limit is unrelated to this PR. We can certainly raise the limit, but it is not possible to prevent the limit from being hit during normal operations. Or more precise, you can prevent it but you would have to set it so high that it's basically useless for preventing most OOM kills, especially in the cloud. Which was the original intent of this limit. As an example, we have seen process maps > 2GB for large processes with very fragmented address spaces.

What this PR tries to change is how the VM *reacts* to the limit, regardless of where the threshold is. IMHO the VM should indicate that it hit a limit. It should not silently truncate the output. Nor should it assert.

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

PR Comment: https://git.openjdk.org/jdk/pull/16474#issuecomment-1827704454