Comment by vips7L
7-20 seconds depending on location.
7-20 seconds depending on location.
I’m going to preface this with I’m not an OPS person. But yes multi region. The main writer instance is us-east-1 it performs excellently when hitting this region. We have read replicas in us-west-2 and some in Europe/emea and Asia/pacific.
When hitting one of these with a write you end up with massive delays. The 7 seconds and below tends to be from us-west-2 and the higher numbers are from our Japanese users.
Our OPS team has struggled to figure out why the delays happen. There’s some code fixes we could probably do (i.e always write to the writer) but as team lead for the development side the deadline is too close and I don’t want to rewrite core parts of the app to split reads and writes. They engaged AWS support so I’m hoping something is just misconfigured or maybe this just isn’t the use case for Aurora.
Oh. Yeah the log file system and thus the consistent replica latency are local to a region.
It sounds like you might be using global aurora with write forwarding? That’s pretty new and not something I have experience sorry. AFAIU though it’s a whole different thing under the hood.
Location? Are you doing multi region?