godoxy-yusing/internal/idlewatcher

Idlewatcher

Idlewatcher manages container lifecycle based on idle timeout. When a container is idle for a configured duration, it can be automatically stopped, paused, or killed. When a request comes in, the container is woken up automatically.

Architecture Overview

graph TB
    subgraph Request Flow
        HTTP[HTTP Request] -->|Intercept| W[Watcher]
        Stream[Stream Request] -->|Intercept| W
    end

    subgraph Wake Process
        W -->|Wake| Wake[Wake Container]
        Wake -->|Check Status| State[Container State]
        Wake -->|Wait Ready| Health[Health Check]
        Wake -->|Events| SSE[SSE Events]
    end

    subgraph Idle Management
        Timer[Idle Timer] -->|Timeout| Stop[Stop Container]
        State -->|Running| Timer
        State -->|Stopped| Timer
    end

    subgraph Providers
        Docker[DockerProvider] --> DockerAPI[Docker API]
        Proxmox[ProxmoxProvider] --> ProxmoxAPI[Proxmox API]
    end

    W -->|Uses| Providers

Directory Structure

idlewatcher/
├── cmd                    # Command execution utilities
├── debug.go               # Debug utilities for watcher inspection
├── errors.go              # Error types and conversion
├── events.go              # Wake event types and broadcasting
├── handle_http.go         # HTTP request handling and loading page
├── handle_http_debug.go   # Debug HTTP handler (dev only)
├── handle_stream.go       # Stream connection handling
├── health.go              # Health monitoring interface
├── loading_page.go        # Loading page HTML/CSS/JS templates
├── state.go               # Container state management
├── watcher.go             # Core Watcher implementation
├── provider/              # Container provider implementations
│   ├── docker.go          # Docker container management
│   └── proxmox.go         # Proxmox LXC management
├── types/
│   └── provider.go        # Provider interface definition
└── html/
    ├── loading_page.html  # Loading page template
    ├── style.css          # Loading page styles
    └── loading.js         # Loading page JavaScript

Core Components

Watcher

The main component that manages a single container's lifecycle:

classDiagram
    class Watcher {
        +string Key() string
        +Wake(ctx context.Context) error
        +Start(parent task.Parent) gperr.Error
        +ServeHTTP(rw ResponseWriter, r *Request)
        +ListenAndServe(ctx context.Context, predial, onRead HookFunc)
        -idleTicker: *time.Ticker
        -healthTicker: *time.Ticker
        -state: synk.Value~*containerState~
        -provider: synk.Value~Provider~
        -dependsOn: []*dependency
    }

    class containerState {
        +status: ContainerStatus
        +ready: bool
        +err: error
        +startedAt: time.Time
        +healthTries: int
    }

    class dependency {
        +*Watcher
        +waitHealthy: bool
    }

    Watcher --> containerState : manages
    Watcher --> dependency : depends on

Provider Interface

Abstraction for different container backends:

classDiagram
    class Provider {
        <<interface>>
        +ContainerPause(ctx) error
        +ContainerUnpause(ctx) error
        +ContainerStart(ctx) error
        +ContainerStop(ctx, signal, timeout) error
        +ContainerKill(ctx, signal) error
        +ContainerStatus(ctx) (ContainerStatus, error)
        +Watch(ctx) (eventCh, errCh)
        +Close()
    }

    class DockerProvider {
        +client: *docker.SharedClient
        +watcher: watcher.DockerWatcher
        +containerID: string
    }

    class ProxmoxProvider {
        +*proxmox.Node
        +vmid: int
        +lxcName: string
        +running: bool
    }

    Provider <|-- DockerProvider
    Provider <|-- ProxmoxProvider

Container Status

stateDiagram-v2
    [*] --> Napping: Container stopped/paused
    Napping --> Waking: Wake request
    Waking --> Running: Container started
    Running --> Starting: Container is running but not healthy
    Starting --> Ready: Health check passes
    Ready --> Napping: Idle timeout
    Ready --> Error check fails: Health
    Error --> Waking: Retry wake

Lifecycle Flow

Wake Flow (HTTP)

sequenceDiagram
    participant C as Client
    participant W as Watcher
    participant P as Provider
    participant H as HealthChecker
    participant SSE as SSE Events

    C->>W: HTTP Request
    W->>W: resetIdleTimer()
    alt Container already ready
        W->>W: return true (proceed)
    else
        alt No loading page configured
            W->>P: ContainerStart()
            W->>H: Wait for healthy
            H-->>W: Healthy
            W->>C: Continue request
        else Loading page enabled
            W->>P: ContainerStart()
            W->>SSE: Send WakeEventStarting
            W->>C: Serve loading page
            loop Health checks
                H->>H: Check health
                H-->>W: Not healthy yet
                W->>SSE: Send progress
            end
            H-->>W: Healthy
            W->>SSE: Send WakeEventReady
            C->>W: SSE connection
            W->>SSE: Events streamed
            C->>W: Poll/retry request
            W->>W: return true (proceed)
        end
    end

Stream Wake Flow

sequenceDiagram
    participant C as Client
    participant W as Watcher
    participant P as Provider
    participant H as HealthChecker

    C->>W: Connect to stream
    W->>W: preDial hook
    W->>W: wakeFromStream()
    alt Container ready
        W->>W: Pass through
    else
        W->>P: ContainerStart()
        W->>W: waitStarted()
        W->>H: Wait for healthy
        H-->>W: Healthy
        W->>C: Stream connected
    end

Idle Timeout Flow

sequenceDiagram
    participant Client as Client
    participant T as Idle Timer
    participant W as Watcher
    participant P as Provider
    participant D as Dependencies

    loop Every request
        Client->>W: HTTP/Stream
        W->>W: resetIdleTimer()
    end

    T->>W: Timeout
    W->>W: stopByMethod()
    alt stop method = pause
        W->>P: ContainerPause()
    else stop method = stop
        W->>P: ContainerStop(signal, timeout)
    else kill method = kill
        W->>P: ContainerKill(signal)
    end
    P-->>W: Result
    W->>D: Stop dependencies
    D-->>W: Done

Dependency Management

Watchers can depend on other containers being started first:

graph LR
    A[App] -->|depends on| B[Database]
    A -->|depends on| C[Redis]
    B -->|depends on| D[Cache]

sequenceDiagram
    participant A as App Watcher
    participant B as DB Watcher
    participant P as Provider

    A->>B: Wake()
    Note over B: SingleFlight prevents<br/>duplicate wake
    B->>P: ContainerStart()
    P-->>B: Started
    B->>B: Wait healthy
    B-->>A: Ready
    A->>P: ContainerStart()
    P-->>A: Started

Event System

Wake events are broadcast via Server-Sent Events (SSE):

classDiagram
    class WakeEvent {
        +Type: WakeEventType
        +Message: string
        +Timestamp: time.Time
        +Error: string
        +WriteSSE(w io.Writer) error
    }

    class WakeEventType {
        <<enumeration>>
        WakeEventStarting
        WakeEventWakingDep
        WakeEventDepReady
        WakeEventContainerWoke
        WakeEventWaitingReady
        WakeEventReady
        WakeEventError
    }

    WakeEvent --> WakeEventType

State Machine

stateDiagram-v2
    note right of Napping
        Container is stopped or paused
        Idle timer stopped
    end note

    note right of Waking
        Container is starting
        Health checking active
        Events broadcasted
    end note

    note right of Ready
        Container healthy
        Idle timer running
    end note

    Napping --> Waking: Wake()
    Waking --> Ready: Health check passes
    Waking --> Error: Health check fails
    Error --> Waking: Retry
    Ready --> Napping: Idle timeout
    Ready --> Napping: Manual stop

Key Files

File	Purpose
watcher.go	Core Watcher implementation with lifecycle management
handle_http.go	HTTP interception and loading page serving
handle_stream.go	Stream connection wake handling
provider/docker.go	Docker container operations
provider/proxmox.go	Proxmox LXC container operations
state.go	Container state transitions
events.go	Event broadcasting via SSE
health.go	Health monitor interface implementation

Configuration

See types.IdlewatcherConfig for configuration options:

• IdleTimeout: Duration before container is put to sleep
• StopMethod: pause, stop, or kill
• StopSignal: Signal to send when stopping
• StopTimeout: Timeout for stop operation
• WakeTimeout: Timeout for wake operation
• DependsOn: List of dependent containers
• StartEndpoint: Optional endpoint restriction for wake requests
• NoLoadingPage: Skip loading page, wait directly

Thread Safety

• Uses synk.Value for atomic state updates
• Uses xsync.Map for SSE subscriber management
• Uses sync.RWMutex for watcher map access
• Uses singleflight.Group to prevent duplicate wake calls