A high-availability, active-active dual-datacenter architecture featuring APISIX gateway, CMS, OPA, RAG services, etcd, and managed databases for robust co
graph TD
%% Users
PLATFORM_ADMIN[Platform Admin]
PROJECT_USER[Project User]
%% Network Layer
LB[Load Balancer HA Pair]
%% DC1
subgraph DC1[Datacenter 1 Active]
APISIX1[APISIX Gateway]
CMS1[CMS Service]
OPA1[OPA Service]
RAG1[RAG Service]
ETCD1[etcd]
VDB1[Vector DB]
end
%% DC2
subgraph DC2[Datacenter 2 Active]
APISIX2[APISIX Gateway]
CMS2[CMS Service]
OPA2[OPA Service]
RAG2[RAG Service]
ETCD2[etcd]
VDB2[Vector DB]
end
%% Managed/External
subgraph MANAGED[External and Managed Services]
ATLAS[MongoDB Atlas - Platform Data]
COSMOS[Azure Cosmos DB - RAG App Data]
VAULT[Managed Vault HCP]
LLM[External LLMs]
end
%% Ingress
PLATFORM_ADMIN --> LB
PROJECT_USER --> LB
LB --> APISIX1
LB --> APISIX2
%% Routing to services
APISIX1 --> CMS1
APISIX1 --> OPA1
APISIX1 --> RAG1
APISIX2 --> CMS2
APISIX2 --> OPA2
APISIX2 --> RAG2
%% Services to self-hosted data
CMS1 --> ETCD1
CMS1 --> ETCD2
CMS2 --> ETCD1
CMS2 --> ETCD2
OPA1 --> ETCD1
OPA1 --> ETCD2
OPA2 --> ETCD1
OPA2 --> ETCD2
RAG1 --> VDB1
RAG1 --> VDB2
RAG2 --> VDB1
RAG2 --> VDB2
%% Managed services connectivity
CMS1 --> ATLAS
CMS2 --> ATLAS
RAG1 --> COSMOS
RAG2 --> COSMOS
APISIX1 --> VAULT
APISIX2 --> VAULT
%% Hairpin to LLM via APISIX
RAG1 --> APISIX1
RAG2 --> APISIX2
APISIX1 --> LLM
APISIX2 --> LLM
%% Replication links
ETCD1 --> ETCD2
ETCD2 --> ETCD1
VDB1 --> VDB2
VDB2 --> VDB1
The diagram illustrates an active-active, dual-datacenter architecture for a configuration management system, incorporating OPA for policy enforcement and RAG services. It details user access via a Load Balancer to redundant APISIX gateways in two datacenters. Each DC hosts CMS, OPA, and RAG services, interacting with replicated etcd and Vector DBs. External managed services like MongoDB Atlas, Azure Cosmos DB, Managed Vault, and External LLMs are also integrated.
This architecture is ideal for applications requiring high availability, disaster recovery, and low latency across geographical regions. It's suitable for critical services needing centralized configuration management, policy enforcement, and AI-driven capabilities (RAG) with robust data persistence and security.
This design can be adapted by integrating different API gateways, replacing etcd with other distributed key-value stores, or using alternative vector databases. Security can be enhanced by adding WAFs or more granular access controls. The RAG component can be extended with various LLM providers or fine-tuned models. Scaling can be achieved by adding more instances within each DC or additional datacenters.