On Wed, Feb 08, 2023 at 11:37:11AM +0100, Thomas Pedoussaut wrote: > > On 08/02/2023 10:09, Aurelien DARRAGON wrote: > > > In fact at some point I had a backend with 5 srv from config + 3 > > > dynamically inserted. Those new ones got about 50 requests pushed to > > > them, until they reaches the slowstart delay(I think, must investigate > > > more), when they stopped being selected for new connection. The status > > > was L7OK for the 8 srv regardless of the way they arrived. > > You're right, I made the same observation during the tests but I forgot > > to mention it: > > A server that is in STARTING phase is immune to this issue: it only > > occurs when server state is RUNNING (either when no slowstart delay is > > specified, or at the end of the slowstart period if specified) > > > > Also, last observation: if you remove the maxconn on a "buggy" server > > (that previously had the maxconn set using the cli): > > > > echo "set maxconn server farm/t1 0" | nc -U /tmp/ha.sock > > > > The issue is gone: you can set the maxconn back to a positive value > > without problems. > > > > I'll keep investigating by looking at the code :) > > > > I made another test, inserted a new server without maxconn, it rightfully > took traffic, then adjusted maxconn to a positive integer and it continued > to work. > > I have a good workaround now for my setup, where would be the best place to > document it ?
Have you seen my previous message about minconn very likely being the culprit ? Once confirmed we'll need to fix it, so it doesn't even need to be documented (since we prefer to fix bugs rather than document how to work around them :-)). thanks, Willy