feat(core): a client can work jobs without standing for leader

client.go

@@ -174,7 +174,9 @@ func (c *Client[TTx]) Start(ctx context.Context) error {
 	if c.caps.Listen {
 		c.listenWG.Go(func() { c.listenLoop(c.bgCtx) })
 	}
-	c.leaderWG.Go(func() { c.leaderLoop(c.claimCtx) })
+	if !c.cfg.NeverLead {
+		c.leaderWG.Go(func() { c.leaderLoop(c.claimCtx) })
+	}
 	c.leaseWG.Go(func() { c.leaseLoop(c.bgCtx) })
 
 	c.state = clientStarted

config.go

@@ -60,6 +60,13 @@ type Config struct {
 	// consumers that start from the earliest event or are moved back.
 	// Defaults to 7 days; negative keeps them forever.
 	StreamRetention time.Duration
+	// NeverLead keeps this client out of leader elections. It still claims
+	// and works jobs, but never runs the leader's duties: rescuing jobs from
+	// lost clients, retention, maintenance, periodic jobs and stream
+	// delivery. Use it for clients that share an install but must not run it,
+	// such as a developer's process against a shared environment. At least
+	// one client of every install must be able to lead.
+	NeverLead bool
 	// LeaseTTL is how long this client's lease lasts without a renewal. It
 	// is renewed three times per TTL; when it lapses, the client's running
 	// jobs are rescued and the client cancels them. A longer lease rides

docs/PLAN.md

@@ -767,6 +767,10 @@ Leader duties: periodic jobs, the rescuer, expiring TTL'd jobs, creating history
 partitions ahead of time and dropping expired ones, removing stale `hopper_clients`
 rows, and resolving batch and workflow completion.
 
+A client with `Config.NeverLead` never attempts the lease. It claims and works jobs
+like any other, so an install can include processes that should do work but not run
+the install, such as a developer's process against a shared environment.
+
 ### 7.9 Periodic jobs
 `hopper.Every(d, …)` and `hopper.Cron(spec, …)` are both available from v0.1. Cron
 supports standard five-field syntax, an optional seconds field, `@hourly`-style

docs/operations.md

@@ -10,9 +10,15 @@
   traffic (claims, finalizes, leases, leader duties) plus one dedicated
   connection for `LISTEN`, outside the pool. Give the pool what the
   application needs on top of that.
-- **Replicas.** Adding replicas adds throughput. Every replica competes for
+- **Replicas.** Adding replicas adds throughput. Every replica works jobs;
+  every replica except one with `Config.NeverLead` also competes for
   leadership; only the leader runs periodic jobs, rescue and retention, and a
   replacement takes over within 15 seconds of the leader stopping.
+- **Clients that must not lead.** Set `Config.NeverLead` on a client that
+  shares an install it should not run, such as a developer's process pointed
+  at a shared environment. It claims and works jobs like any other, but never
+  rescues, prunes, runs periodic jobs or delivers streams. An install whose
+  clients all set it has no leader, so nothing is ever rescued or pruned.
 
 ## Postgres
 

reliability_test.go

@@ -305,6 +305,55 @@ func TestFencedClientLetsALastAttemptFinish(t *testing.T) {
 	}
 }
 
+func TestNeverLeadClientWorksJobsWithoutLeading(t *testing.T) {
+	t.Parallel()
+	h := newHarness(t)
+	ctx := context.Background()
+
+	worked := make(chan hopper.JobID, 1)
+	workers := hopper.NewWorkers()
+	hopper.AddWorkFunc(workers, func(_ context.Context, job *hopper.Job[noop]) error {
+		worked <- job.ID
+		return nil
+	})
+	follower := h.client(&hopper.Config{
+		Queues:    map[string]hopper.QueueConfig{hopper.QueueDefault: {MaxWorkers: 1}},
+		Workers:   workers,
+		NeverLead: true,
+	})
+	if err := follower.Start(ctx); err != nil {
+		t.Fatal(err)
+	}
+
+	res, err := follower.Insert(ctx, noop{}, nil)
+	if err != nil {
+		t.Fatal(err)
+	}
+	select {
+	case id := <-worked:
+		if id != res.Job.ID {
+			t.Fatalf("worked %s, want %s", id, res.Job.ID)
+		}
+	case <-time.After(10 * time.Second):
+		t.Fatal("the job was never worked")
+	}
+
+	// Several leader intervals pass without it standing.
+	time.Sleep(500 * time.Millisecond)
+	if follower.IsLeader() {
+		t.Fatal("a NeverLead client became leader")
+	}
+	if n := h.count("SELECT count(*) FROM hopper_leader"); n != 0 {
+		t.Errorf("leader rows = %d, want 0", n)
+	}
+
+	leader := h.started(hopper.NewWorkers(), 1)
+	waitFor(t, leader.IsLeader)
+	if follower.IsLeader() {
+		t.Error("the NeverLead client became leader alongside another")
+	}
+}
+
 func TestLeaderFailover(t *testing.T) {
 	t.Parallel()
 	h := newHarness(t)