Puppet

Mostly I think this is a duplicate issue – not that I can find the dup issue. :)

I just ran into this also. You have to explicitly set autoflush when using a file log destination, otherwise log output may be buffered until the next log message is written. There have been some issues (#4139) with the --autoflush command line argument, but it should be fixed in the version you are using

Josh Cooper wrote:

When logging to a file, puppet should always flush the message in the fflush sense, not the fsync sense. It shouldn’t require users to enable autoflush

…this is nice, but isn’t the root cause of the problem on Windows. Ruby will flush internal buffers safely when it shuts down in the normal case, so something on Windows is causing it to exit uncleanly, or without the normal cleanup from triggering that flush.

Daniel Pittman wrote:

…this is nice, but isn’t the root cause of the problem on Windows. Ruby will flush internal buffers safely when it shuts down in the normal case, so something on Windows is causing it to exit uncleanly, or without the normal cleanup from triggering that flush.

Yep. On Unix, the daemon.stop method gets called to cleanly shutdown, but the same is not true when running puppet agent via third-party service wrapper. I’ll file a separate ticket for that. Likely we will need to register a console ctrl handler, SetConsoleCtrlHandler, and close the logs, etc.

Recently ran into a support case where the customers were using the —autoflush option, so we definitely will need a deprecation cycle for this option.