Aurélien D.

We've had a quick talk about that with Willy. We're thinking about relaxing the logformat compatibility checks in 3.0+ so that they only emit diag_warnings (haproxy will only report them...

>While you are doing that if you can squeeze it in to make the error message (diagnostic message) a tad more verbose and point out the specific object that would...

What about using `stick store-request` for `/start` requests and `stick on` for other requests? ```haproxy stick-table type ip size 200k expire 30m acl persist path -m str "/start" stick store-request...

I'm not sure to properly understand you're issue. But if multiple servers are expected to share the same IP address (from same FQDN), then you may want to set the...

Oh ok my bad, didn't see that you have 10k backends with 1 server each (and not 10k servers in 1 backend) I'm checking

Ok, I'm able to reproduce indeed, thanks for your inputs I'm investigating

Ok.. so again my bad, regression has been introduced with cd994407a9545a8d84e410dc0cc18c30966b70d8 (this commit is starting to become my nemesis!) Each time the loop of pending events is interrupted ([tune.events.max-events-at-once](https://docs.haproxy.org/dev/configuration.html#tune.events.max-events-at-once) which...

It should be fixed now (available in master, and will end up in 3.0 and 2.9 for upcoming releases)

Thanks for filing a bug report, as always :)

You're welcome We'll keep the issue open for tracking until the fix is backported to relevant versions