One Application Insights instance per tenant

Question

One Application Insights instance per tenant

Jonathan Cook 0

I am looking at the approach to using Application Insights with a mutli-tenant solution described here:

https://learn.microsoft.com/en-us/azure/architecture/guide/multitenant/service/application-insights#one-application-insights-instance-per-tenant

The goal is having a single App Service write to multiple Application Insights instances based on the tenant that is making the request to the App Service. Is there an example or further documentation into how I would achieve what is described in the link?

EDIT: I see that you can set the InstrumentationKey for a telemetry item using ITelemetry.Context.InstrumentationKey. I am getting conflicting information as to whether or not setting it in this way effects only the individual item, or if the change might effect other items using the same context.

AnuragSingh-MSFT 21,566 Reputation points Moderator

2023-09-07T05:37:53.03+00:00

Jonathan Cook, thank you for the question. The second half of this doc has provided features of AppInsights which can be used to achieve this - Application Insights features that support multitenancy. I have reached out to the team for additional details. I will update this thread with further update.
Jonathan Cook 0 Reputation points

2023-09-11T15:23:35.2033333+00:00

Hi @AnuragSingh-MSFT , I saw that section on those features but it did not really tie it all together for me. From what I have seen elsewhere, it appears that I might need to configure multiple TelemetrySinks, and then use TelemetryProcessors to determine which sink should handle them. But if there is an established pattern or some examples to work from that would be a huge help. Thank you for looking into it and I look forward to your update.
AnuragSingh-MSFT 21,566 Reputation points Moderator

2023-09-15T03:54:06.3933333+00:00

Jonathan Cook, I wanted to update that the AppInsights team is looking into it and I will get back to this thread as soon as I have some update.
Jonathan Cook 0 Reputation points

2023-10-02T18:15:01.9+00:00

Hi @AnuragSingh-MSFT , just following up to see if the AppInsights team was able to provide any additional info. I also wanted to point out my edit to the question, since maybe it will be an easier question to answer. I took a brief look at the source code and it seems like the TelemetryContext gets deep cloned for every message, and therefore setting the Instrumentation Key on each message would not have any unintended side effects. The doc comments for the InstrumentationKey property seem a bit ambiguous about this.
AnuragSingh-MSFT 21,566 Reputation points Moderator

2023-10-09T05:32:51.96+00:00

Jonathan Cook, I apologize that it is taking a while. Based on the primary information that I have received; this would not be under supported scenario for ApplicationInsights implementation as AI SDK was not designed for this kind of scenario.

I have requested more details and also to get the referenced doc updated with this information to avoid any confusion.

Apologies for the inconvenience.

1 answer

Your answer

AnuragSingh-MSFT 21,566 Reputation points Moderator

2023-09-07T05:37:53.03+00:00

Jonathan Cook, thank you for the question. The second half of this doc has provided features of AppInsights which can be used to achieve this - Application Insights features that support multitenancy. I have reached out to the team for additional details. I will update this thread with further update.
Jonathan Cook 0 Reputation points

2023-09-11T15:23:35.2033333+00:00

Hi @AnuragSingh-MSFT , I saw that section on those features but it did not really tie it all together for me. From what I have seen elsewhere, it appears that I might need to configure multiple TelemetrySinks, and then use TelemetryProcessors to determine which sink should handle them. But if there is an established pattern or some examples to work from that would be a huge help. Thank you for looking into it and I look forward to your update.
AnuragSingh-MSFT 21,566 Reputation points Moderator

2023-09-15T03:54:06.3933333+00:00

Jonathan Cook, I wanted to update that the AppInsights team is looking into it and I will get back to this thread as soon as I have some update.
Jonathan Cook 0 Reputation points

2023-10-02T18:15:01.9+00:00

Hi @AnuragSingh-MSFT , just following up to see if the AppInsights team was able to provide any additional info. I also wanted to point out my edit to the question, since maybe it will be an easier question to answer. I took a brief look at the source code and it seems like the TelemetryContext gets deep cloned for every message, and therefore setting the Instrumentation Key on each message would not have any unintended side effects. The doc comments for the InstrumentationKey property seem a bit ambiguous about this.
AnuragSingh-MSFT 21,566 Reputation points Moderator

2023-10-09T05:32:51.96+00:00

Jonathan Cook, I apologize that it is taking a while. Based on the primary information that I have received; this would not be under supported scenario for ApplicationInsights implementation as AI SDK was not designed for this kind of scenario.

I have requested more details and also to get the referenced doc updated with this information to avoid any confusion.

Apologies for the inconvenience.

Answer 1

Jonathan Cook, apologies for the delayed response as I had been discussing this scenario with the AppInsights team. Here are the key outcomes:

By default, this is not possible with AppInsights SDK as available currently. There could be work-around which might make it work but those are untested and unsupported at the moment.
Another option could be to use **opentelemetry-dotnet **which provides additional benefits including sampling, post-processing, and more. The following links contain features of OTel which could help achieve this ()

Customizing OpenTelemetry .NET SDK for Tracing

Tech Preview: OpenTelemetry Transformation Language statements as routing conditions

These features of OTel will depend on the ability to recognize the tenant-based telemetry for which additional tweaks might be required (for example inclusion of tenant header, if not readily available). While these are some of the options available, they have not been tested extensively for their efficacy and supportability in the current scenario.

I have also shared feedback and review request for the referenced document in the question for it to include caveats to the architecture discussed from Azure Monitor perspective.

Share via

One Application Insights instance per tenant

1 answer

Your answer