5

Has anyone successfully done rolling updates with Azure Front Door? We have an application in 2 regions, and we want to disable the backend in region 1 while it gets updated and same for the backend in region 2. However, there seems to be a ridiculous amount of lag time between when you disable or remove a backend from a pool, making this basically impossible.

We've tried:

  • Disabling/totally removing backends
  • Setting high/low backend priorities/weights
  • Modifying health probe intervals
  • Changing sample size/successful samples/latency to 1/1/100

I have an endpoint that I watch during the deployment process which tells me which region it's in, and it never changes during the operation, and becomes unavailable when the region is being updated. There's gotta be a way to do this, right?

Thomas
  • 24,234
  • 6
  • 81
  • 125
mrosack
  • 161
  • 1
  • 5
  • Original post was 2 years ago but doesn't look like Front Door delays in config changes has improved. I was looking into these delays today and fud adding/removing backends took over 30mins to take affect. https://camerondwyer.com/2021/07/16/how-to-avoid-downtime-during-blue-green-deployment-of-service-behind-azure-front-door/ – CamD Jul 16 '21 at 13:10

1 Answers1

0

I have a suggestion,

  1. Reduce the Health Probe Interval.
  2. Reduce the sample size and successful sample required. (Make sure you are probing to a simple HTTP page so your backend resource can handle the loads. You will start receiving probes from all the POP servers with the interval you specified.) 3.For the sever which you need to do maintenance, stop the service or make the probe fail, so that all traffic will switch to the healthy server. Then do the maintenance and start the service again. This will make sure your service is not disrupted.
msrini-MSIT
  • 1,444
  • 1
  • 7
  • 11
  • I have a similar setup to mrosack and I see the same behaviour. I tried the steps suggested by @msrini-MSIT (for step 3 I stopped the service), but I get about 15 seconds of downtime before the traffic switches to the healthy backend. Probe config is 5/1/1 (interval/sample size/successful samples/required). Not bad but it's not 'zero downtime'. – Tim Webster Jul 09 '19 at 19:30