Fix: Load Memory Sidecar Modules from Runtime Scripts in Hermes Memory Installer

Hermes Memory Installer is a utility for deploying and managing memory sidecar modules in containerized environments, commonly used alongside Kubernetes workloads. These modules handle tasks such as memory profiling, leak detection, and allocation tracking, running as sidecars to application containers. In the latest release, a critical fix addresses how these modules are loaded, enabling them to be sourced from runtime scripts rather than static configuration alone. This change improves flexibility for dynamic deployments.

The Problem

Previously, the installer loaded memory sidecar modules based on a static list defined in configuration files or environment variables during initialization. This approach required modules to be predetermined at build time or image creation, leading to several issues:

• Inflexibility: Changing modules for different environments (e.g., dev vs. prod) meant rebuilding images or maintaining multiple configuration overlays.
• Operational Overhead: Deployments with varying memory monitoring needs across nodes or clusters required complex scripting outside the installer.
• Error Prone: Hardcoded paths often broke when module names or versions changed across releases.

In dynamic Kubernetes environments, static loading limited the ability to adapt to runtime conditions, such as switching profilers based on workload profiles.

The Fix: Runtime Script Loading

This update introduces the ability to load memory sidecar modules through runtime scripts. The installer now supports executing a script during startup, whose output defines which modules to load. This allows deployment logic to be externalized and driven by environment variables, external metadata, or custom logic.

A typical configuration uses a ConfigMap mounted as a script:

apiVersion: v1
kind: ConfigMap
metadata:
  name: hermes-module-loader
data:
  load-sidecars.sh: |
    #!/bin/bash
    # Select modules based on cluster type
    if [ "$CLUSTER_ENV" == "production" ]; then
      echo "module=mem-profiler,config=prod-tunings"
    elif [ "$CLUSTER_ENV" == "staging" ]; then
      echo "module=mem-tracer,config=staging-tunings"
    else
      echo "module=mem-logger"
    fi

The installer processes this script, parses the output, and loads the specified modules with their configurations. This enables dynamic selection without image changes.

Implementation Notes

The core change is a new loading phase in the installer’s init sequence. It checks for a script path (defaulting to a known location like /etc/hermes/load-sidecars.sh), executes it with a shell, and interprets the standard output as module directives. This is backward compatible—if no script exists, it falls back to static configuration. The fix ensures that sidecar modules are consistently loaded even when orchestration triggers multi-step init processes, as previously these script-based loads were skipped.

Benefits for Developers

• Runtime Adaptability: Modules can be chosen based on cluster labels, node types, or external APIs.
• Simpler Deployments: CI/CD pipelines inject scripts rather than rebuild images.
• Reduced Image Bloat: Optional modules are never bundled but loaded on demand.

This fix solidifies Hermes Memory Installer as a practical tool for teams that need fine-grained control over memory sidecar management in dynamic stacks. Check the project’s release artifacts for the updated binaries and documentation on script syntax.