Skip to content

daemon.md

Latest commit

 

History

History
238 lines (173 loc) · 6 KB

daemon.md

File metadata and controls

238 lines (173 loc) · 6 KB
Raw

Daemon

The daemon (grite-daemon) is optional and exists only to improve performance and coordination. Correctness never depends on it.

Quick Start

# Daemon auto-spawns on first command (no manual start needed)
grite issue list

# Manual control
grite daemon start --idle-timeout 300
grite daemon status
grite daemon stop

# Force local execution (skip daemon)
grite --no-daemon issue list

Auto-Spawn

The daemon automatically spawns when you run CLI commands:

  1. CLI checks for running daemon
  2. If no daemon, spawns grite-daemon in background
  3. Waits for daemon to become ready (up to 5 seconds)
  4. Routes command through IPC
  5. Daemon runs until idle timeout

Default idle timeout is 5 minutes (300 seconds).

Disabling Auto-Spawn

Use --no-daemon to force local execution:

grite --no-daemon issue list

Idle Timeout

The daemon automatically shuts down after a period of inactivity:

# Start with 10-minute idle timeout
grite daemon start --idle-timeout 600

# Start with no timeout (runs until stopped)
grite daemon start --idle-timeout 0

The idle timer resets on each command. When timeout is reached:

  1. Daemon logs "Idle timeout reached"
  2. Workers shut down gracefully
  3. All locks released
  4. Process exits

Responsibilities

  • Maintain a warm materialized view for fast reads
  • Handle concurrent CLI requests efficiently
  • Refresh daemon lock heartbeat
  • Release locks on shutdown

Non-Responsibilities

  • Never rewrites refs or force-pushes
  • Never writes to the working tree
  • Never becomes required for correctness
  • No background sync (sync is explicit via grite sync)

Architecture

+----------------+     +----------------+     +----------------+
|    CLI         | --> |   Supervisor   | --> |    Worker      |
|  (grite)        |     |  (manages IPC) |     | (per repo/actor)|
+----------------+     +----------------+     +----------------+
        |                      |                      |
        v                      v                      v
   IPC Request          Route to Worker         Execute Command
                                                      |
                                                      v
                                               +-------------+
                                               | LockedStore |
                                               |   (sled)    |
                                               +-------------+

Supervisor

  • Listens on IPC socket (/tmp/grite-daemon.sock)
  • Routes requests to appropriate worker
  • Manages worker lifecycle
  • Tracks idle time for auto-shutdown

Worker

  • One worker per (repo, actor) pair
  • Holds exclusive flock on sled database
  • Spawns concurrent tokio tasks for commands
  • Refreshes daemon lock heartbeat

Concurrency

The daemon handles concurrent requests efficiently:

  1. Supervisor receives IPC request
  2. Routes to worker for (repo, actor)
  3. Worker spawns tokio task
  4. Sled MVCC handles concurrent access
  5. Response sent back via IPC

Multiple CLI processes can issue commands simultaneously. The daemon serializes database access internally while allowing concurrent execution.

Database Locking

Filesystem Lock (flock)

The daemon acquires an exclusive flock on sled.lock:

.git/grite/actors/<actor_id>/sled.lock

This prevents other processes from opening the sled database while the daemon is running.

Daemon Lock (ownership marker)

The daemon creates a JSON lock file for coordination:

.git/grite/actors/<actor_id>/daemon.lock

Example:

{
  "pid": 12345,
  "started_ts": 1700000000000,
  "repo_root": "/path/to/repo",
  "actor_id": "64d15a2c383e2161772f9cea23e87222",
  "host_id": "hostname",
  "ipc_endpoint": "/tmp/grite-daemon.sock",
  "lease_ms": 30000,
  "last_heartbeat_ts": 1700000000000,
  "expires_ts": 1700000030000
}

Lock Rules

Scenario CLI Behavior
No daemon lock Execute locally or auto-spawn
Lock valid, IPC reachable Route through daemon
Lock valid, IPC unreachable Error (daemon may have crashed)
Lock expired Take over, execute locally

CLI Integration

Status

$ grite daemon status
Daemon is running
  PID:            12345
  Host ID:        my-laptop
  IPC Endpoint:   /tmp/grite-daemon.sock
  Started:        2024-01-15 10:30:00 UTC
  Expires in:     25s

JSON Output

$ grite daemon status --json
{
  "running": true,
  "pid": 12345,
  "host_id": "my-laptop",
  "ipc_endpoint": "/tmp/grite-daemon.sock",
  "started_ts": 1705315800000,
  "expires_ts": 1705315830000,
  "time_remaining_ms": 25000
}

Failure Behavior

Failure Recovery
Daemon crashes Lock expires, CLI can take over
IPC timeout CLI retries 3 times, then errors
Worker panics Supervisor continues, worker restarted on next request
Command error Error returned via IPC, daemon continues

Logging

The daemon logs to stderr. Control verbosity with --log-level:

grite-daemon --log-level debug

Log levels: trace, debug, info, warn, error

When auto-spawned, daemon runs with --log-level info and stdout/stderr redirected to /dev/null.

IPC Protocol

  • Socket: /tmp/grite-daemon.sock (Unix domain socket)
  • Framing: length-prefixed (u32 BE + payload)
  • Serialization: rkyv (zero-copy)
  • Concurrency: one task per connection

See IPC Protocol for message format details.

Configuration

The daemon reads configuration from:

  1. Command-line arguments
  2. Environment variables (for log level)

No configuration file is used. The daemon is stateless except for the workers it manages.

Comparison: With vs Without Daemon

Aspect Without Daemon With Daemon
First command latency Higher (open sled) Lower (sled warm)
Concurrent commands Serialize at flock Concurrent in daemon
Memory usage Per-process Shared in daemon
Complexity Simple More moving parts
Correctness Same Same