High CPU utilization in Azure Durable Function on AKS due to excessive file watcher threads (600–800 threads per pod)

[VikramaVutakanti]

Description

We are running an Azure Durable Function App on AKS with KEDA using a custom Docker image based on the official Azure Functions runtime. We are consistently observing high CPU utilization caused by a very large number of file watcher threads (600–800) being spawned inside the function pod.

This occurs intermittently:

• Sometimes in one pod
• Sometimes across two pods
• Even when traffic is low or idle

Despite explicitly disabling file watching via environment variables and host.json, the issue persists.

---

Environment

• Platform: Azure Kubernetes Service (AKS)
• Scaling: KEDA
• Function type: Durable Functions
• Runtime: Azure Functions v4
• .NET version: .NET 8
• OS: Linux container

---

Dockerfile

FROM mcr.microsoft.com/dotnet/sdk:8.0 AS installer-env

FROM mcr.microsoft.com/azure-functions/dotnet:4-dotnet8.0

ENV AzureWebJobsScriptRoot=/home/site/wwwroot \
    AzureFunctionsJobHost__Logging__Console__IsEnabled=true \
    DOTNET_HOSTBUILDER__RELOADCONFIGONCHANGE=false \
    DOTNET_USE_POLLING_FILE_WATCHER=1 \
    AzureFunctionsJobHost__fileWatchingEnabled=false \
    WEBSITE_RUN_FROM_PACKAGE=1 \
    ASPNETCORE_URLS=http://+:8080 \
    WEBSITE_HOSTNAME=localhost:8080

---

host.json

{
  "version": "2.0",
  "logging": {
    "fileLoggingMode": "never",
    "console": {
      "isEnabled": "false"
    },
    "logLevel": {
      "default": "Information",
      "Host": "Error"
    },
    "applicationInsights": {
      "samplingSettings": {
        "isEnabled": false,
        "maxTelemetryItemsPerSecond": 500
      }
    }
  },
  "extensions": {
    "durableTask": {
      "storageProvider": {
        "controlQueueBatchSize": 32,
        "controlQueueBufferThreshold": 256,
        "controlQueueVisibilityTimeout": "00:05:00",
        "maxQueuePollingInterval": "00:00:05",
        "partitionCount": 4,
        "workItemQueueVisibilityTimeout": "00:05:00",
        "useAppLease": true
      },
      "maxConcurrentActivityFunctions": 40,
      "maxConcurrentOrchestratorFunctions": 40
    }
  },
  "watchFiles": false,
  "watchDirectories": false,
  "fileWatchingEnabled": false
}

---

Observed Behavior

• CPU spikes to high levels even during low or no load

• Thread dump shows 600–800 file watcher–related threads

• High CPU correlates directly with these file watcher threads

• Scaling events (scale down / scale up via KEDA) seem to increase the likelihood

• Issue reproduces even with:

  • watchFiles=false
  • fileWatchingEnabled=false
  • DOTNET_HOSTBUILDER__RELOADCONFIGONCHANGE=false
  • WEBSITE_RUN_FROM_PACKAGE=1

---

Expected Behavior

• File watching should be fully disabled when explicitly configured
• Durable Function pods should not spawn hundreds of file watcher threads
• CPU utilization should remain low during idle or low-traffic periods

---

Additional Notes

• This setup is fully containerized (not App Service)
• Running Azure Functions on AKS with KEDA
• Looks like file watching might still be enabled internally or re-enabled during scaling events
• This causes unnecessary CPU consumption and impacts cluster cost and stability

---

Request

• Clarification on why file watcher threads are still being created

• Guidance on fully disabling file watching in Azure Functions running on AKS

• Confirmation if this is a known issue with:

  • Azure Functions v4
  • Durable Functions
  • .NET 8
  • KEDA-based scaling

---