Maintenance page irule was not working properly

I don't have a test setup where I can easily reproduce this right now, but my guess would be that it may have to do with the event you are using (LB_FAILED). After your LB::reselect this may not trigger anymore, because now you are landing on a working pool again.

You already have a check for active pool members < 1, so you can use the HTTP_REQUEST event instead. It's a bit less efficient, because it triggers all the time, not just in case of an ongoing maintenance. But I've never noticed any performance impact.

I always do some variation of the following:

when HTTP_REQUEST {
	# Check if pool is unavailable
	if { [ active_members prod_pool ] < 1 } {
		# Don't do rewriting for resources of the maintenance page
		if { not ( [HTTP::uri] starts_with "/maintenance/" ) }{
			# Rewrite everything else to maintenance page
			HTTP::uri "/maintenance/"
		}
 
		# Rewrite Host Header, set different Pool
		HTTP::header replace Host "maintenance.server"
		pool maintenance_pool
	}
}

But as I said, I don't know if the event is actually what's causing your issue or if it's related to something else in your iRule or your maintenance page or your network environment.

Try adding some logging (log local0. "your text here") to your iRule in different places, to see what happens when you refresh the page. Does the event still trigger? Do your conditions still match? What's the value of [HTTP::uri] during the various steps of the process? This may be a bit of data to parse, but you should be able to figure out what happens.