Hi Conrad, On Fri, Jul 25, 2014 at 03:34:39AM +0200, Conrad Hoffmann wrote: > Hey Willy, > > I know you're not that much into the systemd stuff, but although we > don't use systemd we actually us the wrapper a lot, and I thought it > could use a little make-over.
OK. > There was only one small bug really, but once I got going I thought i > might as well simplify it a litle. I pulled everything out of the signal > handlers, switched from signal() to sigaction(), got rid of global state > and got rid of the wrapper exec()'ing itself (not sure if there was some > reason for that, but it seems to be unneccessary). >From what I remember, the purpose was to be able to upgrade the wrapper itself without having to kill it. Typically in order to apply changes like you just performed... So I think you should bring that feature back. > It is late at night and I might have to revise a few things after some > sleep, but I wanted to get this out for you and others to look at. It is > against tonights master of the haproxy-1.5 repo, I could also create it > for 1.6 instead (or maybe it even applies out of the box). It should apply to both, there's no difference between the two. In the future you'd rather make patches against the development version, it's the only way to ensure the patch is correctly applied in forward versions people test less. The backport from dev to stable is our job and if we break something during this task, everyone will notice it and report it. You cleanup patch is interesting but it does too many things at once for a single patch, I'd strongly prefer if you would cut it into pieces each addressing a specific issue. It would make the code more easily reviewable and would also help troubleshooting in the event it would cause any minor regression. Thanks, Willy
