On 13/11/18 9:37 π.μ., Bruno Henc wrote: > Hello, > > > Not sure if there is a direct way to do this, but you can always create a > monitoring process that > will use the haproxy runtime API to MAINT or DRAIN a server until the CPU / > network load drops. So > you have a simple watchdog process which reads the output from your > monitoring tools to decide if a > server needs to be disabled or re-enabled. >
Another approach is to use agent-check from haproxy to query a sidecar process on the backend servers. That sidecar process determines, based on various criteria, the load of the server and instructs haproxy to either drain or reduce the percentage of traffic that it gets. Cheers, Pavlos
signature.asc
Description: OpenPGP digital signature