Lattency-based routing not working

Question

Lattency-based routing not working

Ray Chamidullin 20

Hello,

I am configuring Azure Front Door to route traffic to my Azure ContainerApps (one in US, one in Germany). The latency-based routing does not seem to work in my setup. Do I have anything wrong with my setup?

Here is my setup:

One endpoint with one route to an origin group that has 2 origins: Azure ContainerApp in US and Azure ContainerApp in Germany.
Both origins have priority 1 and weight 1000.
Session Affinity is enabled.
Health probes are enabled.
Load balancing has settings: Sample size = 4, Successful samples required = 3, Latency sensitivity (in milliseconds) = 0

Applications in ContainerApps return x-region header with values "eastus" or "germanywestcentral" depending on the region where they are deployed.

I did the following experiment to test the routing:

Call ContainerApp in US - average response time 400 ms
Call ContainerApp in Germany - average response time 100 ms
Call Front Door endpoint 100 times - 80% of times US instance responded with ~400 ms response time and only 20% of times Germany instance responded with ~100 ms response time

I would expect that majority of calls would route to Germany instance.

Accepted answer

0 additional answers

Your answer

Answer 1

KapilAnanth-MSFT 49,021 Microsoft Employee

@Ray Chamidullin ,

Welcome to the Microsoft Q&A Platform. Thank you for reaching out & I hope you are doing well.

From your verbatim,

From your region, you find that,
- ContainerApp in US has a response time of 400 ms
- ContainerApp in Germany has a response time of 100 ms
So, from the same region, you try to access the AFD which has both of these or Origins
And you find that 80% of the requests is being forwarded to ContainerApp in US (one with a higher response time)

From Traffic routing methods to origin

The 'closest' origin isn't necessarily closest as measured by geographic distance. Instead, Azure Front Door determines the closest origin by measuring network latency

Response time is not same as latency, as health probes may probe a different page than the origin actually serves or health probes may use HEAD Request instead of GET request.
You say you made 100 calls, can you increase the sample size and give it a try - say 1000 to 2000 requests?
Also, you mentioned accessing the container apps from both AFD and directly have the same latency (400 ms for US and 100 ms for Germany)
- Is this correct?
- Introducing the AFD did not improve the response time?

Cheers,

Kapil

Ray Chamidullin 20 Reputation points

2024-08-09T07:18:47.06+00:00
Thank you for your response, Kapil.

In my experiment I calling health probe endpoints (I don't use any other endpoints). I also tried using HEAD requests instead of GET, I increased sample size to 2000 and included random sleeps between calls. But the results are similar.

Here are the updated measurements when calling AFD.

Gemany - 136 ms on average, 887 / 2000

US - 361 ms on average, 1101 / 2000

Calling ContainerApps directly has a bit lower latency..

Germany - 92 ms on average

US - 355 ms on average

I am mostly trying to understand how the routing works and want to confirm that when I will go to production my users won't have unnecessary delays.
KapilAnanth-MSFT 49,021 Reputation points Microsoft Employee

2024-08-09T09:53:06.8633333+00:00
@Ray Chamidullin ,

Ideally, with latency sensitivity set as 0, traffic should route to the fastest available origins and weights on the origin don't take effect unless two origins have the same network latency.

From your test, it looks like the traffic is actually being Load Balanced.

It is possible that the health probes are failing

I believe you are experiencing Complete health probe failure

If health probes fail for every origin in an origin group, then Front Door considers all origins unhealthy and routes traffic in a round robin distribution across all of them.

See : How Front Door determines origin health

As next steps,

Please check the Origin Health Percentage Metrics and make sure the health percentage is non zero

If the above is zero,

consider checking the Health probe log

Cheers,

Kapil
Ray Chamidullin 20 Reputation points

2024-08-12T11:50:38.7966667+00:00

Thank you, Kapil, for your help. Indeed, I had issues with health probes. It works fine now after fixing the health probe configuration.
KapilAnanth-MSFT 49,021 Reputation points Microsoft Employee

2024-08-12T12:16:36.15+00:00

@Ray Chamidullin , you are most welcome :)

I would greatly appreciate if you could Accept the answer and close this thread

Original posters help the community find answers faster by identifying the correct answer.

Thanks,

Kapil

Share via

Lattency-based routing not working

0 additional answers

Your answer