Linux App Service (B1): system process memory rises to ~95% every night since Nov 10, not caused by the app

Paul van den Tillaart 0 Reputation points
2026-11-12T21:55:00.3533333+00:00

Since 10 November 2026 we see a daily block of extra CPU and memory usage on our Linux App Service plans that does not come from our application. We would like to know whether this is known platform activity, and how to avoid it.

Setup

  • Region: West Europe
  • Two Linux App Service plans, both Basic B1 with 1 instance, each running one Node.js web app (production and staging)
  • Both plans have the identical set of outbound IP addresses, so we assume they share a scale unit. They run on different worker instances.

What we see

  • Every night since 9/10 November there is a block of about 8 to 9 hours:
    • production plan: 02:00 to 11:00 UTC
    • staging plan: 23:00 to 07:00 UTC (3 hours earlier, starting the same night)
  • During that block the plan CPU is about 38 to 42% (baseline 20 to 23%), and the plan memory peaks at 88 to 91% (hourly average). Our memory alert (above 85%) fires daily.
  • In the 27 days before, neither plan showed this pattern.

Why we think it is not the application

  • The app has almost no traffic in these hours, and its working set does not rise (it actually drops).
  • In "Diagnose and solve problems" > "Linux Memory Drill Down" the "System Processes Memory Usage (ps)" graph shows the total system memory rising to 90 to 95% in exactly this window, while the "App Svc Platform processes" line stays flat or even drops (see attached graph). The user processes of the site stay flat.
  • No deployments or configuration changes before or during the first occurrence.
  • Resource Health shows "Available" for both plans and both sites. There are no Service Health events and no activity log entries for this period.

Questions

  1. Is there known platform activity (maintenance, agents, updates) on Linux workers in West Europe since 10 November that could explain this?
  2. Is there a way to see which system processes use this memory, or to avoid it?
  3. Would scaling up (B1 to B2) or moving the plan to another scale unit be the recommended mitigation?

Thank you.

Azure App Service
Azure App Service

Azure App Service is a service used to create and deploy scalable, mission-critical web apps.

0 comments No comments

1 answer

Sort by: Most helpful
  1. Pavan Addepalli 0 Reputation points
    2026-11-19T11:58:28.2+00:00

    Based on your description, the repeated increase outside the application’s working set warrants a worker/platform-level investigation. However, the graph alone does not identify a particular agent, maintenance task, or platform defect.

    Is this known platform activity?

    I couldn’t find a public Microsoft advisory confirming this specific West Europe pattern beginning November 10, 2026. App Service performs platform and operating-system maintenance, but that does not establish that your recurring eight-to-nine-hour resource increase is expected maintenance.

    The identical outbound IP sets are consistent with shared deployment infrastructure, but do not indicate that the apps share a worker or identify the cause. Also, Basic plans already have compute dedicated to the plan; B1 is not the shared-compute hosting tier.

    Resource Health showing Available does not establish that resource consumption is normal: its documented App Service checks focus on host/service availability, reachability, and content storage.

    Can you identify the processes?

    There is an important visibility limitation: SSH into a Linux App Service gives access to the application container, not the underlying host operating system. Consequently, top or ps there cannot provide a complete inventory of worker-host processes.

    Since you have already used Linux Memory Drill Down, the next useful step is to preserve that evidence and open an Azure support request for host-level CPU and memory attribution during the recurring window. Include both plan/app resource IDs, worker instance IDs, exact UTC windows, diagnostic screenshots or exports, and a normal-period comparison.

    For that comparison, use the plan’s CPU and memory metrics at one-minute granularity, split by instance, rather than relying only on hourly averages. Both metrics support this granularity and dimension. Ask support to identify the growing processes and distinguish process memory from other host memory consumption, then correlate the onset with platform changes.

    Should you scale up or move?

    B1 → B2 is a reasonable temporary capacity test if headroom is a concern, not a confirmed fix. B2 provides two cores and 3.5 GB RAM versus B1’s one core and 1.75 GB. I would test staging first and compare absolute consumption and application performance across the next nightly window—not just the lower percentage on a larger machine.

    A move to another deployment unit is not a normal “Change plan” operation. Microsoft documents that existing plans cannot change webspaces, and supported app-to-plan moves are restricted to the same webspace. Do not assume that creating another plan or changing its resource group will move the existing workload to a different scale unit. Discuss relocation with support before undertaking a redeployment.

    The public documentation establishes these diagnostic and mitigation limits, but it does not explain your November 10 onset; identifying the responsible host activity requires the resource-specific support investigation.

    Was this answer helpful?

    0 comments No comments

Your answer

Answers can be marked as 'Accepted' by the question author and 'Recommended' by moderators, which helps users know the answer solved the author's problem.