From the haproxy mailing list: I now encourage everyone to migrate to maintenance release 1.3.19. The 3 last bugs I fixed can cause random pauses in traffic and sometimes return 502 server error for extremely small objects if the server closes too fast. This should be enough to motivate you to upgrade. The less nasty one (server error) also affected old versions such as 1.3.15 and 1.3.14, so I have released one of each as well. But now let's talk about the goodies. Having a development tree in parallel is nice, because it allows better testing with less risks, and the ability to backport riskless features. So there are a few minor features that I have backported since there is some demand : - support for multiple configuration files. You can now have up to 10 files in which you split your various sections (eg: global, http defaults, http frontends, http backends, tcp frontends, tcp backends). - a lot more verbose error reporting when checking the conf, even if you specify "-D". I have encountered huge errors in some configurations which were not detected, so some specific checks have been added for such errors, and the config parser now returns after reporting all the errors, not only the first one. This helps for instance when checking a config which has hostnames which do not resolve where you check it. - support for per-server source port ranges, because there are some people who needed to overcome the 64k concurrent connections limit (!) They promised to send me a screenshot of their stats page when they reach the million if ever :-) - support for setting a server's weight to zero, which helps for maintenance periods. It's better than setting it backup as you're certain it will never be used for new requests, even if it's alone.
From the mailing list: Hi all, With the very kind help from Cristian Ditoiu, I've just isolated a very nasty bug introduced in 1.3.19 by the fix for the timer drift : http://haproxy.1wt.eu/git?p=haproxy-1.3.git;a=commitdiff;h=4de2dea8076782d0685cd1f25edfb961d3a9d40c If you are running a configuration with some missing timeouts, your haproxy might crash a few minutes after it starts. In fact, it will crash as soon as it find an unexpirable task in the wait queue, that it will dequeue and reuse just as if it was queued. I'll release 1.3.20 and 1.4-dev2 shortly so that we don't stay too long with this bug. The simple workaround consists in ensuring that you never have an instance running without timeouts (there are some warnings emitted upon startup for this). And anyway you should never do that as you will certainly observe some session leakage over the time. In the mean time, and specially for distro maintainers (as I have noticed that at least Mandriva and Fedora maintainers have already updated their packages, which is really kind of them), I'm attaching the patch I'm going to merge (it's running fine right now). I suggest you merge it into your own 1.3.19 packages for now. I have another very minor thing to fix which can wait. More on all of that with the release, which I hope will be out on Sunday. Thanks for your patience, and of course, to all the guys who help chase bugs ! Willy Let's not stable 1.3.19, let's wait for 1.3.20.
Fixed in cvs by adding version s 1.3.14.14, 1.3.15.10 and 1.3.19.