Background
The RealTheory Collector is a lightweight agent that runs inside your Kubernetes cluster and continuously streams metrics, performance, cost, and health data to the RealTheory platform.
The agent is designed to restart periodically as part of its routine maintenance and upgrade workflow. These restarts are intentional, controlled, and help ensure the agent remains healthy, up to date, and aligned with the latest RealTheory improvements.
Kubernetes increments the pod restartCount whenever the container process exits, regardless of whether the exit was planned or due to an error. Because Kubernetes does not distinguish between the two, planned maintenance restarts cannot be differentiated from crash‑driven restarts in the pod status.
Expected Restart Behavior
By design, each Collector agent performs a self-restart weekly. This automatic restart ensures that:
-
The latest image version is pulled from the official RealTheory container registry because the agent's
imagePullPolicyisAlways -
The agent remains in sync with the RealTheory backend and can take advantage of new features and compatibility updates
In addition to this scheduled restart, the agent might also restart under other expected conditions, such as:
-
Security token rotation – Refreshing credentials to maintain secure communication with RealTheory services
-
TLS certificate renewal – Updating certificates used for encrypted data transmission
-
Configuration or secret updates – Reloading specific configuration changes without requiring a full redeployment
-
Memory optimization – Performing a controlled restart to clear memory fragmentation and maintain stable operation
-
Explicit management actions – Restarting when an upgrade or restart command is issued through the RealTheory Management Console
These restarts typically take less than a second and are transparent. In-flight data is buffered and is resent automatically after the agent is back online, so no data is lost, and no action is required.
When to Investigate
Under normal conditions, an agent should restart no more than a few times per week (including the scheduled restart). A higher restart count alone does not indicate a problem if the agent remains healthy and data continues to flow. However, if you observe restarts occurring multiple times per day or notice data gaps in the RealTheory console, this might indicate an underlying issue such as:
-
Network instability or failing health checks
-
A misconfigured deployment (for example, custom registry or pinned image version)
-
Resource constraints on the node hosting the Collector
If frequent restarts persist, please contact RealTheory Support so we can review your agent logs and metrics to diagnose further.
See Also