On Tue, Mar 10, 2015 at 01:32:03AM +0000, Nick Huanca wrote: > Willy Tarreau <w <at> 1wt.eu> writes: > > > That's intentional, and I'll reject any patch to document it. It's a > > debugging feature whose format changes between versions or for any > > reason we see fit when facing a new problem. This can only be used > > with the code at hand. > > > > Would it even be possible to give at least a guideline? After diagnosing > many idle timeouts and tunnel timeout issues it was very useful to watch > specific sessions and watch the counters. I wouldn't want them documented > to be consistent throughout versions but maybe at least a heads up on what > the value correlate to for connection diagnostics.
It's *really* meant to be read with the code in parallel, nothing else. We have changed it many times and it will change many times again. Sure we could imagine having a doc for it but it would make the process of updating it to track bugs down much more cumbersome, and some of the information we put there is hard to describe in a doc because we're basically dumping various fields from a few structs in relation with the session being dumped, just in case that may help sort out multiple hypothesis. And in fact, I think that's exactly what you did. The best thing that can be done in terms of doc would basically be to indicate that this format is subject to change anytime and is used for debugging with the code open in the editor, so that users know it exists and check by themselves if/when they believe that can help. > Is there a feature request process? Yes, by posting here and you did that pretty well. BTW, additions of extra fields there are possibly welcome if they can help. Thanks, Willy
