ai-scheduling-server

Scheduling server for scheduling ai-profile generation jobs

Restores and Synchronizes job scheduling state via schedule_state.json file. (which is created at initial run)

Supports job status retrieval & engine status(ai-server) configuration by REST-API at runtime.

Architecture

flowchart TD
    A[Client] -->|Job| C{Scheduler}
    C -->|Job| D[Engine1]
    C -->|Job| E[Engine2]
    C -->|Job| F[Engine3]

Loading

Job sequence

sequenceDiagram
    Scheduler->>+Engine: Healthcheck
    Engine->>-Scheduler: Healthcheck reponse
    Scheduler->>+Engine: Job Request
    Engine->>+Model: Preprocess & Inference Request
    Model-->>-Engine: Result | Error
    Engine-->>+GCP: Result upload
    Engine-->>-Scheduler: Response

Loading

State Transfer

Job State

stateDiagram-v2
    Pending --> InProcess
    InProcess --> Processed
    InProcess --> Error

Loading

EngineStatus State

stateDiagram-v2
    Ready --> InProcess
    InProcess --> Ready
    InProcess --> Error
    Error --> Ready
    Ready --> Error

Loading

1. Build&Run

# git clone prequisites
git clone https://github.com/aiprofile-gdsc-koreauniv/ai-scheduling-server/

cd ai-scheduling-server/

# create error log
touch $PWD/error.txt

# build docker image
docker build -t MY_CONTAINER_NAME .

# docker run
docker run -d \
   -p MY_PORT:9000 \
   -v $PWD:/app \
   MY_CONTAINER_NAME

참고사항

• docs-url : http://localhost:MY_PORT/docs
• engine : ai-api-server 를 의미합니다.
• job : docs에 정의되어 있는 프로필 생성 요청 1건을 의미합니다.
• schedule_state.json : job의 state를 기록/복원하는 state 파일입니다.