Lightweight Deployment Trimming Plan Verification¶

Considering that in some scenarios, especially in the military industry, cloud software deployment faces stringent resource constraints, this document introduces a lightweight trimming and verification plan for deploying DCE Community Edition, aiming to meet customers' requirements for lightweight deployment.

This trimming plan is verified step by step through four phases, detailed as follows:

Verification Environment¶

• Operating System: Kylin Linux Advanced Server V10 (Halberd) for ARM
• CPU: 10C
• Memory: 17Gi

Installed Components and Resource Statistics¶

The installation components, trimming plan, and phased trimming approach for DCE are as follows:

Full view of Phase 1 lightweight trimming:

Optimization Measures¶

1. Insight can stop the following Pods while ensuring normal monitoring capabilities:

   Pod name	Mem Size
   insight-agent-fluent-bit-5x2rn	99.62 MiB
   insight-agent-otel-kubernetes-collector-69f67cc745-xt5hj	74.94 MiB
   insight-agent-tailing-sidecar-operator-6f85f7bb75-67xc8	46.81 MiB
   insight-elastic-alert-64bbb468dc-l4mk5	30.38 MiB
   insight-jaeger-collector-5cd5b94dcc-mwgcl	32.50 MiB
   insight-jaeger-query-5495c59bbd-fk287	28.88 MiB
   insight-opentelemetry-collector-5d47dd6c6b-nk54t	62.12 MiB
   Optimizable memory	375.25 MiB

2. Remove the Istio sidecar using the script clean_istio_proxy.sh

3. Shut down the Bootstrap kind-cluster and the elasticsearch component

   1. Disable the elasticSearch component in manifest.yaml before the installer deployment

      [Not Feasible] insight-server strongly depends on es

   2. Shut down the kind-cluster container after the installer deployment is complete

      [Feasible] The image pull policy has hidden risks and needs to be adjusted to IfNotPresent

4. Deploy an image registry in the global management cluster, and manage it through kangaroo [Feasible]

5. Deploy a single-instance MySQL, and use it as the external MySQL instance for Container Management [Feasible]

Optimization by Phase¶

Phase 1 Optimization

Phase 1 trims non-essential infrastructure and component modules.

1. Resource usage inside Bootstrap

   

2. Resource usage of Global + kind Bootstrap

   

   

According to script-based statistics, the total memory consumption is: 13.6 GiB

Estimated memory consumption of removable components:

If these four categories of components are not installed, the estimated memory consumption would be 7.36 GiB. However, previous community edition trimming tests (without these four components) showed that 8 GB memory could only run the system barely, and the system was unstable.

Phase 2 Optimization

In Phase 2, MySQL was consolidated from dual instances into a single instance, while some Insight components were further trimmed and optimized.

Phase 3 Optimization

Optimization item: remove ES (replicas = 0), based on issue 2268

Phase 4 Optimization

Conclusion¶

This lightweight plan is derived by progressively trimming and scaling down a full K8s + DCE5 Community Edition environment under sufficient memory, reaching an idle state. Based on the memory statistics, the total memory just meets the requirements.

In actual real-world scenarios, with an 8Gi memory environment, installation proceeds sequentially (adaptive trimming). However, since some steps at the end require manually executing scripts, it triggers deploy rolling updates, causing a temporary surge in memory demand. Additionally, the operating system itself consumes part of the dynamic memory resources.

Therefore, in summary, running the DCE lightweight trimmed environment in an 8G memory setup still lacks sufficient resources.

That is, installation memory requirement ≠ idle state memory usage

• Using Phase 4 as the final trimming goal, at least 10G+ memory is required.
• Using Phase 3 as the final trimming goal (including observability components), at least 12G+ memory is required.

For how to achieve lightweight deployment through the installer, refer to Installer Lightweight Deployment Plan.

Comments