Potential recurrence of #13309

[josalem]

I'm experiencing this issue with .NET Core 3.1

This causes the Linux container to hit OOM. Running the app in VS in debugger has the same issue.

Unfortunately, I cannot share my memory dump. But the summary is that I have the Azure Service Bus lib in my app. The service bus lib tries to connect to the topic you specify. If the topic does not exists it keeps throwing exceptions indefinitely. This causes the AI to keep gobbling memory.

This also causes the container to fail to swap at release time bc it runs OOM.

Originally posted by @nurevaSaeed in #13309 (comment)

[josalem]

@nurevaSaeed, could you describe the scenario a little more? Specifically

• What applications are running (including your own) inside your container or when you are debugging with VS?
• Are you tracing your application using dotnet-trace or using another .NET Diagnostics Tool when this happens?
• Are you using DiagnosticListener in your application?
• Are you using prometheus-net.DotNetRuntime in your app?
• What specifically makes you think you're seeing a recurrence of [2.2] Potential native memory leak from EventPipe #13309, e.g., are you seeing a leak of buffers of the same size as described there?

Lastly, could you explain this a bit more:

This causes the AI to keep gobbling memory.

Are you talking about Application Insights?

If possible, a minimal repro of the behavior, some output from the !heap -stat command showing the leak, or some other concrete evidence of the issue would be very helpful in triaging this.

[nurevaSaeed]

Sorry @josalem, I'd deleted my post before seeing your message. I will send an update with what you're asking soon. For now, I don't think my issue applies to this bug anymore.

[tommcdon]

@semyon2105 are you still experiencing the CPU issues mentioned in djluck/prometheus-net.DotNetRuntime#6?

[semyon2105]

@tommcdon yes, still seeing the same CPU increase in repro using the latest .NET SDK Docker image

[tommcdon]

@sywhang can you take a look?

[sywhang]

@semyon2105 What version of .NET Core 3.1 are you using?

[semyon2105]

@sywhang I'm using version 3.1.6

[sywhang]

Thanks for confirming that. Just to make sure, is https://github.com/semyon2105/leak-repro still a valid repro for your issue?

[semyon2105]

Yes, and I've just updated it to use the latest .NET image

[sywhang]

@semyon2105 Thanks. I tried running your repro and have some findings I'd like to share, but since the issue is unrelated to any runtime issues (including the one posted here), I will post my findings in the issue you initially reported it in the prometheus-net repo.

@josalem I believe the recurrence that you saw here is related to the mmap fix I made back in 3.1.5 - I'm going to close this issue since we've addressed that issue.