leader

Documentation

Overview

Package leader implements single-node leader election via PostgreSQL session-level advisory locks. Only one Kronos node holds the lock at a time; the others block in the election loop. When the leader's connection drops or context is cancelled, PostgreSQL releases the lock automatically and one waiting node takes over.

Advisory locks are a lightweight alternative to etcd/ZooKeeper for deployments already using Postgres. The tradeoff is that split-brain is bounded by the TCP keepalive timeout rather than a lease TTL — acceptable for job schedulers where a few seconds of overlap is recoverable via idempotent handlers.

Index

• type Elector
  • func New(dsn string, log zerolog.Logger) *Elector
  • func (e *Elector) Run(ctx context.Context, onElected func(context.Context), onRevoked func())

Types

type Elector

type Elector struct {
	// contains filtered or unexported fields
}

Elector manages leader election for a single Kronos node.

func New

func New(dsn string, log zerolog.Logger) *Elector

New creates an Elector. dsn is a standard PostgreSQL connection string.

func (*Elector) Run

func (e *Elector) Run(ctx context.Context, onElected func(context.Context), onRevoked func())

Run blocks until ctx is cancelled. It repeatedly tries to acquire the advisory lock. Once acquired, it calls onElected with a derived context and blocks until leadership is lost (connection failure or ctx cancel). onRevoked is called after leadership ends, before retrying.

Typical usage in main.go:

go elector.Run(ctx,
    func(leaderCtx context.Context) { scheduler.Run(leaderCtx) },
    func() { log.Info().Msg("lost leadership, standing by") },
)