Since my mail I figured out it's the inter time yeah, which is
overriding the timeouts.
This is 2x apache _test_ servers with absolutely 0 traffic atm other than the healthchecks. So the apache maxconnection limits are not reached for sure.
Any idea why would those hiccups happen? Looks like it's happening ~1x/hour with 3secs inter param. Guess it's not that awfull, but 3secs is a lot when it comes to response time, testing through local net, on a 0byte static file, usual response time is <0.1sec.
Thank your for your response!
Idézet (Willy Tarreau <w#1wt.eu>):
> On Fri, Dec 05, 2008 at 06:51:14PM +0100, Joe7 wrote:
>> Just installed this great tool.
>> Unfortunately having strange issues, haproxy _randomly_ marks hosts as down.
>> They are up, can't see any error on them or anything.
>> Looks to be an issue if httpcheck is on, with tcp only it looked good.
>> Any idea why that woud happen?
> It sounds like the servers take a bit more time than expected to reply
> to health-checks. This happens on java-based servers when the GC is
> running for instance. You should increase the "inter" parameter in order
> to cover the occasional hiccups.
> Another possibility is that your servers cannot accept all connections
> for the traffic and the health-checks. In this case, it's wise to set
> the "maxconn" parameter slightly below the server's limit (eg: MaxClient
> on apache) so that there's always a free connection for the health-check.
> On Mongrel, it's a bit harder since you only have one connection at a
> time for everything. In this case, either you have fast response times
> and it will be OK with a large enough "inter", or you have slow response
> times and the only solution will be to stay with tcp checks unfortunately.
Received on 2008/12/08 08:46
This archive was generated by hypermail 2.2.0 : 2008/12/08 09:01 CET