System Deployments

Tech stack

1. AWS + Terraform

   flowchart LR;
       subgraph A[VPC];
           subgraph AA[Public Subnet]
           end
           subgraph AB[Private Subnet]
               direction LR;
               subgraph ABA[Fargate]
                   direction TB;
                   ABA1[ECS]
                   ABA2[ECS]
                   ABA3[ECS]
               end
               ABB[NLB] --> ABA;
               ABC[APIGW] --> ABB;
           end
           AC[Internet Gateway] --> AA;
           AA --> AD[NAT];
           AD --> AB;
       end;
       B[ECR] --> AC;
       C[Cognito] --> AC;
   

2. Azure bicep

   flowchart LR;
       subgraph B[Azure Cloud]
           direction TB;
           B1[ACR];
           B2[Key Vault];
           subgraph BA[Container Apps]
               BA1[frontend] --> BA2[backend];
           end
           subgraph BB[Serverless]
               BB1[EventGrid] --> BB2[Function Apps];
               BB2 --> BB3[OCR];
           end
           subgraph BC[Storage]
               BC1[Blob]
               BC2[CosmosDB]
           end
           B1 --> BA;
           BA --> BC;
           BC --> BA;
           BC --> BB;
           BB --> BC;
           BB --> BA2;
       end
       subgraph A[Azure Devops]
           subgraph AA[Repository]
               subgraph AAA[CICD Pipeline]
                   AAA1[CI Testings]
               end
               AAA -->|package| AAB[Artefacts]
   
           end
       end
       A -->|docker image| B1;
       A -->|iac| B;
       BA2 -->|install package| A;
       C((User)) --> BA1;
   

3. Dockers + Helm + Kubernetes

   flowchart LR;
       subgraph AA[Node]
           direction LR;
           subgraph A[Deployment]
               A1[Pod];
               A2[Pod];
               A3[Pod];
           end
       end
       B[NodeIP] --> AA;
   

4. Google Cloud Platform

   flowchart LR;
       A[Artifact Registry] --> B[Cloud Run];
   

5. FastAPI + Prometheus + Grafana

   flowchart LR;
       A[FastAPI] -->|/metrics| B[Prometheus];
       B --> C[Grafana]
   

   

Archive

• API calls

  sequenceDiagram
      participant app;
      participant api;
      participant location_api;
      participant sensor_api;
      participant prediction_api;
      Note over app: requirements <br>- delay: 2s;
      app->>+api: GET;
      Note over location_api: <br>- endpoint: /location <br>- delay: 0.5s;
      api->>+location_api: GET /location;
      location_api-->>-api: return (lat,lng);
      Note over sensor_api: <br>- endpoint: /sensor <br>- delay: 2s;
      api->>+sensor_api: GET /sensor?lat,lng;
      sensor_api-->>-api: return (timestamp, sensor_data);
      Note over prediction_api: <br>- endpoint: /predict <br>- delay: 4s;
      api->>+prediction_api: GET /sensor?lat,lng,sensor_data;
      prediction_api-->>-api: return (prediction_data);
      api-->>-app: ;
  

• MLOps

  flowchart TB;
      subgraph A[Continuous Integration]
          direction TB
          subgraph AA[Data]
              AAB[Database] --- AAA[Artefacts];
          end
          subgraph AB[Preprocessing]
              ABA[Exploration] --> ABB[Processing];
              ABB-->ABC[Feature Extraction];
          end
          subgraph AC[Model]
              ACA[Model Selection]-->ACB[Train];
              ACB-->ACC[Validate];
              ACC-->ACD[Model Registration];
              ACD-->ACA;
          end
          AA-->|batch/stream processing|AB;
          AB-->|features|AA;
          AC-->|model artefacts|AA;
          AA-->AC;
      end
      a((client));
      a-->|raw data|AA;
      subgraph B[Continuous Deployment]
          direction LR
          subgraph BA[Deploy]
              BAA[Containers]-->BAB[Servers];
              BAB-->BAC[Load Balancer] & BAD[Logging];
          end
          subgraph BB[Monitoring]
              BBA[Visualisation Tool];
          end
          BAD-->BBA;
      end
      AC-->B;
      B-->|endpoints, kpis|a;
  

  Role	Responsibilities
  Data engineer	Data, Preprocessing
  Data scientist	Preprocessing, Model
  ML engineer	Preprocessing, Model, Deploy
  Devops	Deploy, Monitoring

  Responsibility	Skills
  Data	SQL, Postgresql, Pgadmin, SQLAlchemy, SQLModel, Alembic
  Preprocessing	Spark, Pandas, Databricks
  Model	Tensorflow, Pytorch, MLflow
  Deploy	FastAPI, Tensorflow, Pytorch, Prometheus
  Monitoring	Prometheus, Grafana

• Kafka Broker and Database Crash

  flowchart LR;
      A[Collectors]-->B[Kafka Broker];
      subgraph C[Workers]
          C1[Worker1];
          C2[Worker2];
      end
      B-->|work1 topic|C1;
      B-->|work2 topic|C2;
      subgraph D[Database]
          D1[Table1];
          D2[Table2];
      end
      C1-->|1. write|D1;
      D-->|2. crash|D;
      C2-->|3. disconnect, can't write|D2;
  

  ---
  displayMode: compact
  ---
  gantt
      title Workers Timetable
      dateFormat HH:mm
      axisFormat %H:%M
      tickInterval 1hour
  
      section Worker 1
      w1: a1, 00:20, 15m
      w1: a1, 02:20, 15m
      w1: a1, 04:20, 15m
      w1: a1, 06:20, 15m
      w1: a1, 08:20, 15m
      w1: a1, 10:20, 15m
  
      section Worker 2
      w2: a2, 00:00, 1h
      w2: a2, 04:00, 1h
      w2: a2, 08:00, 1h
  

Others

1. Compnay Hierarchy

   flowchart TB;
       A[root]-->B;
       subgraph B[scope];
           direction TB;
               B1[service-publisher];
               B2[infra-builder];
               B3[developer];
       end
   
       subgraph C[service];
           direction TB
               C1[ecr];
               C2[ecs];
               C3[s3];
       end
   
       E[(credentials)];
       F[(AWS Management Console)];
   
       D[Leo]-->B3;
       B1 -->|push| C1;
       B2 -->|build| C1 & C2 & C3;
       B3 -->|read| C1 & C3;
       D --> E;
       B1 --> E;
       B2 --> E;
       D & A -->|access| F;