Am 07.02.2018 um 14:25 schrieb jocelyn fournier:
each layer brings inevitable it's own problems to the mix and if it's only added latency for non-cacheable queries and complexer dependencies for long-term maintainance of oyur whole application stack
I agree before adopting this kind of software, you have to check the cacheability of your application (which is a metric displayed by the heimdalldata dashboard BTW) because there's always a small overhead. But I like the concept of heimdalldata where the cache is actually hosted on each front server, instead of a potentially remote MySQL server, hence *reducing* the latency of your application by removing the network round-trip delay for cached queries well, our mysqld are conencted via unix-sockets, so no TCP and no failovers needed on that layer - HA is done by the virtualization cluster on a deeper layer and when the whole cluster fails game over anyways :-)