I have create Azure VMSS behind Public Azure Std LB with HTTP based Health Probe. Azure Loadbalancer is working as per expectation. But If VM is unhealthy then it must be deleted or re-provisioned. So that machine can attain healthy state again.153 votes
I’d like to ask you for more feedback on this request please. Load Balancer doesn’t control the VMSS. I think what you’re looking for is a way for VMSS to replace any instances with a LB health probe status of 0. I’ve reached out to VMSS team to get their input. LB is likely not the right place to do this.
Azure load balancer should report the status of backend pool VMs based on the health probes we have created and not just report if the VM is running or not. Recently one of my backend pool VM went into high load and it took me minutes to identify the problematic one. LB was not sending any traffic to this VM however the portal was still showing it 'Running' instead of "Unhealthy" or "Failed". This would save a lot of time and will be easy to see health of backend pool vms.107 votes
We recently released a pre-configured dashboard that will help with this. Let us know if you find this useful. Thanks for your feedback. https://azure.microsoft.com/en-us/blog/introducing-azure-load-balancer-insights-using-azure-monitor-for-networks/– Anavi N [MSFT]
The current Azure Load Balancer checks are very basic. They should be extended to allow
* Custom headers added to HTTP checks
* Auth headers on HTTP checks
* Check returned content for a pattern
* Allow POST rather than GET with customisable content
* Accept other HTTP return statuses (for example, 3xx) as valid
* TCP checks should allow configuration of Send/Expect strings
* Maybe additional checks for such services as LDAP, DNS, etc
As comparison, the popular HAProxy load balancer supports most of these capabilities.15 votes
Thank you for the feedback.
Is it possible to add the swicth "--internal-dns-name <Name>" for an internal loadbalancer ?
This is possible for an NIC today, please add this to loadbalancer12 votes
Hi there – thanks so much for the feedback.
Curious to know about your use case for this? This helps us plan and prioritize better.
Looking forward to hearing from you.
– Anavi N [MSFT]
We would like to implement an application-specific https load balancer probe for rolling upgrades. This always fails initial deployment because none of the VMs in the scale set have the application installed yet. It would be nice if the rolling upgrade threshold could be ignored until it passes for the first time.8 votes
thank you for the feedback. how would you want this to work? do you want to be able to configure how we detect it the first time, automatically probe up until first time probed up, something else?
Currently you need to provide a custom probe service or use applications (e g SQL Always-On Availability Group) which has built-in probe services or use other services' ports (e g RPC 135) for the probe service. Would it be possible to provide a probe service application that you can install as a service on the nodes instead of having to write your own probe service? I have seen some C# samples of such TCP port probe services, which you could install as a service, but it would be better to have official in case there are improvements, updates or changes required.
Currently you need to provide a custom probe service or use applications (e g SQL Always-On Availability Group) which has built-in probe services or use other services' ports (e g RPC 135) for the probe service. Would it be possible to provide a probe service application that you can install as a service on the nodes instead of having to write your own probe service? I have seen some C# samples of such TCP port probe services, which you could install as a service, but it would be better to have official in case there are improvements, updates or changes…6 votes
Could you please reply with details on the scenario and how you envision this to work?— Anavi N [MSFT]
At the moment the Azure load balancer supports only 1 IPv6 IP on the internet frontend.
The IPv4 adresses where sold, the future is the usage of IPv6. But a loadbalancer can only handly one IPv6 Address???
It minimum we should be able to terminate one IPv6 Präfix. Better that the LB can handle many dedicated IPv6 addresses.3 votes
Hi there – does multiple frontend IP addresses help in your scenario?
– Anavi N [MSFT]
The probe use RST to disconnect a TCP connection that established 3 way handshake. If I use an software to monitor some paket, the software will detect some errors by RST paket. I hope we can use FIN sequence to close the TCP connection.3 votes
Thank you for the feedback. Can you clarify please if you are asking for the health probe to close the connection
a) WITH a 4 way close or
b) WITHOUT a 4 way close and send RST instead?
In addition to health probes, provide a probe to determine latency when pushing packets to the backend.3 votes
Hi there – could you share more details about your scenario and need for this so we can look into it further?
– Anavi N [MSFT]
need longer time for 'keep alive' of socket. just like you can configure them with NGINX for example. We need min. 10 hours per socket.2 votes
Hi there – I assume you are already leveraging keep alives? Why is 10 hours needed?
Would like to learn more of your scenario.
— Anavi N [MSFT]
- Don't see your idea?