Hello Percona Community,
I am setting up a PMM Server (version 3.2.0) deployment in my local Kubernetes environment (running on macOS) using the official Percona Helm chart.
The Setup
-
PMM Version: 3.2.0 (using official Helm chart)
-
Environment: Local Kubernetes (on macOS, orbstack)
-
Client: Successfully pushing metrics from local MySQL instance.
The Problem (Replication Issue)
When I configure the StatefulSet with replicas: 1, everything works perfectly: the Pod starts, I can log into the PMM UI, and client metrics are displayed.
However, when I increase the replica count to replicas: 3, all three Pods come up successfully. But upon attempting to log into the PMM UI (accessed via the service), its logging in I am able to access the UI, but after few seconds OR sometimes instantaneously logged out with an “Unauthorized” error popup. It appears to be an immediate session invalidation or redirect loop triggered post-login.
The Core Question
This behavior strongly suggests an issue with session management or state synchronization between the multiple Pods.
-
What is the specific relationship between PMM Server’s login/session mechanism and the number of replicas that could cause this “Unauthorized” auto-logout error? Does PMM require a specific shared storage or configuration for session handling in a multi-replica setup that is not handled automatically by the Helm chart’s default StatefulSet configuration?
-
How is PMM High Availability (HA) fundamentally designed to work in a Kubernetes environment? Is the goal of increasing replicas in the StatefulSet to simply avoid an SPOF (Single Point of Failure), or is there a specific requirement for leader election/clustering that needs additional configuration?
-
Will a Pod failure be automatically handled if I increase the replica count, or does the PMM architecture require external components (like a specific database/store) to manage state across Pods for true HA?
Any insights into how PMM manages state and sessions across multiple Kubernetes replicas would be greatly appreciated. Thank you!
