I can share the setup that reproduces the issue (Note: I have excluded a lot of config to isolate the issue):
(Note: Just a note though, there was no traffic and I waited for 30 minutes. It only happens with two reload happens at same time) *1*. *Ubuntu 14.04 LTS* with *HAProxy* installed using backports sudo apt-get install haproxy/*trusty-backports* *2*. With out of the box config provided (see attached) for the ubuntu platform, simply reload the config twice (make sure that reload logic happens in background) : (See below) #!/bin/bash -c #This is a test script to reproduce the issue. This is not our production logic. #Simply reload twice in back ground and check for haproxy process service haproxy reload & service haproxy reload & *3*. Run the script : sudo bash reload.sh *4.* Grep for haproxy process : *ps -efl | grep haproxy.cfg | grep -v grep* 1 S haproxy 16253 2191 0 80 0 - 7245 ep_pol 00:33 ? 00:00:00 /usr/sbin/haproxy -f /etc/haproxy/haproxy.cfg -p /var/run/haproxy.pid -D -sf 16173 1 S haproxy 16254 2191 0 80 0 - 7245 ep_pol 00:33 ? 00:00:00 /usr/sbin/haproxy -f /etc/haproxy/haproxy.cfg -p /var/run/haproxy.pid -D -sf 16173 *5.* Repeat step 2-4 , you will notice there process count has increased by 1 more : 1 S haproxy 16253 2191 0 80 0 - 7245 ep_pol 00:33 ? 00:00:00 /usr/sbin/haproxy -f /etc/haproxy/haproxy.cfg -p /var/run/haproxy.pid -D -sf 16173 1 S haproxy 16303 2191 0 80 0 - 7245 ep_pol 00:34 ? 00:00:00 /usr/sbin/haproxy -f /etc/haproxy/haproxy.cfg -p /var/run/haproxy.pid -D -sf 16254 1 S haproxy 16304 2191 0 80 0 - 7245 ep_pol 00:34 ? 00:00:00 /usr/sbin/haproxy -f /etc/haproxy/haproxy.cfg -p /var/run/haproxy.pid -D -sf 16254 *6. *Modify the reload script and do not reload it in background #!/bin/bash -c service haproxy reload service haproxy reload *7*. Kill all processes and repeat steps 3-4 (several times) You will notice that old process do not stackup as before. ps -efl | grep haproxy.cfg | grep -v grep 1 S haproxy 16513 2191 0 80 0 - 7245 ep_pol 00:38 ? 00:00:00 /usr/sbin/haproxy -f /etc/haproxy/haproxy.cfg -p /var/run/haproxy.pid -D -sf 16494 Sukrit On Mon, Aug 4, 2014 at 12:00 AM, Baptiste <[email protected]> wrote: > On Mon, Aug 4, 2014 at 6:15 AM, Sukrit Khera <[email protected]> wrote: > > We are planning to setup proxy service using HaProxy. We realized that > when > > we send multiple reload signals at same time, haproxy process count > > increases and the old process do not die. > > > > If we serialize the reload requests( i.e. execute one after other), we do > > not see the leak. > > > > Note : This happens even when there is not traffic to the site and the > old > > process never die. > > > > sukrit@sukrit-dev:~$ haproxy -v > > HA-Proxy version 1.5.3 2014/07/25 > > > > Linux (Ubuntu Trusty 14.04) : 3.13.0-32-generic > > > > Test Script : > > service haproxy reload & > > service haproxy reload & > > > > Use Case: We are providing proxy as a service to our internal apps to do > a > > blue green deploy. Once new app comes up, we do a reload. SInce we have > > multiple workers running which processes the deployment request, at > times we > > might have 2 proxy requests being served at same time. > > > > I was wondering if that is intended behavior and our script should handle > > the serialization of request or if it is a bug. > > > > Thanks and Regards > > Sukrit Khera > > > Hi Sukrit, > > This behavior occurs by design. > Can you share with us your configuration? > Old processes remains active as long as it has an active connection. > A connection could remain active because there is data exchanged on it > or because you set up very long time outs. > > Could you share with us your configuration (don't forget to anonymize it) > > Baptiste > --
haproxy.cfg
Description: Binary data
