RFR: 8320691: Timeout handler on Windows takes 2 hours to complete

[Jaikiran Pai]

On Fri, 24 Nov 2023 07:58:18 GMT, Daniel Jeliński <djelinski at openjdk.org> wrote:

> The recent cdb versions do not support `.dump /f`:
> 
> *****************************************************************************
> * .dump /f is not supported on a user mode process.                         *
> *                                                                           *
> * .dump /ma creates a complete memory dump of a user mode process.          *
> *****************************************************************************
> 
> and after printing that message, cdb ignores the rest of the command line and never quits.
> 
> This PR updates the dump command to use the recommended `/ma` parameter. This allows the command to produce a dump and complete in a timely manner.

test/failure_handler/src/share/conf/windows.properties line 61:

> 59: native.core.app=cdb
> 60: native.core.args=-c ".dump /ma core.%p;qd" -p %p
> 61: native.core.params.timeout=3600000

Hello Daniel, I found it surprising that this takes 2 hours to complete. The failure handler infrastructure has timeout handling built in, after which it kills the failure handler action (the process). Looking at the value specified here it translates to a timeout of 60 minutes (which is too high by the way). So I looked around in some other files and I think there might be a bug here. In other files (linux.properties and mac.properties), I notice the timeout is specified as:


native.core.timeout=600000

Notice the absence of "params" part in that key. I wonder if that is playing a role here and whether we should fix this key. While at it, perhaps we should also reduce this timeout to may be something lesser (1 hour seems to high). Linux and macosx use a value of `600000` which is 10 minutes. If Windows requires a few more minutes then that's understandable but perhaps we should set it to a maximum of 30 minutes maybe?

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

PR Review Comment: https://git.openjdk.org/jdk/pull/16806#discussion_r1404106460