dgraph-io/dgraph · error

Unable to reach leader of group: %d

Error message

Unable to reach leader of group: %d

What it means

Zero cannot obtain a gRPC connection to the RAFT leader of the given group while executing predicate deletion. The group exists in the membership state but s.Leader(gid) returned nil, meaning no healthy leader address is known to Zero (leader down, not yet elected, or membership maps out of sync). deletePredicates aborts and the caller only logs a warning, so cleanup is skipped.

Source

Thrown at dgraph/cmd/zero/zero.go:815

	for _, tablet := range group.Tablets {
		gid = tablet.GroupId
		break
	}
	if gid == 0 {
		return errors.Errorf("Unable to find group")
	}
	state, err := s.latestMembershipState(ctx)
	if err != nil {
		return err
	}
	sg, ok := state.Groups[gid]
	if !ok {
		return errors.Errorf("Unable to find group: %d", gid)
	}

	pl := s.Leader(gid)
	if pl == nil {
		return errors.Errorf("Unable to reach leader of group: %d", gid)
	}
	wc := pb.NewWorkerClient(pl.Get())

	for pred := range group.Tablets {
		if _, found := sg.Tablets[pred]; found {
			continue
		}
		glog.Infof("Tablet: %v does not belong to group: %d. Sending delete instruction.",
			pred, gid)
		in := &pb.MovePredicatePayload{
			Predicate: pred,
			SourceGid: gid,
			DestGid:   0,
		}
		if _, err := wc.MovePredicate(ctx, in); err != nil {
			return err
		}
	}

View on GitHub (pinned to 759e242be6)

Solutions

1. Verify all Alphas of the group are running and connected to Zero (check /health and Zero's /state endpoint).
2. Retry the predicate drop after the group elects a leader; the error is transient during elections.
3. Check network/firewall connectivity between Zero and the Alpha servers on their gRPC ports.
4. If the group should no longer exist (removed Alphas), wait for Zero's membership state to converge or use cleanupTool/dgraph zero removal for dead nodes.

Example fix

// before
curl -X POST localhost:8080/admin -d '{"query":"mutation { dropOp(pattern: "name") { response { message } } }"}'  # fails while group leader is down
// after
# restart the down Alpha first, confirm leader in Zero /state, then re-run the drop

// Before dropping predicates, check Zero's membership state
const res = await fetch('http://zero:6080/state');
const state = await res.json();
const group = state.groups && state.groups[gid];
if (!group || !group.members || !group.members.some(m => m.leader && m.healthy === true)) {
  throw new Error(`Group ${gid} has no healthy leader; defer predicate drop`);
}

function hasHealthyLeader(group) {
  return Boolean(group && Array.isArray(group.members) &&
    group.members.some(m => m.leader === true && m.healthy === true));
}

try {
  await dropPredicate(pattern);
} catch (err) {
  if (/Unable to reach leader of group/.test(String(err))) {
    await waitForGroupLeader(gid);   // poll /state until leader healthy
    await dropPredicate(pattern);    // retry once leader is up
  } else {
    throw err;
  }
}

When it happens

Trigger: A drop of a predicate triggers MovePredicate cleanup via UpdateMembership; Zero knows the group/tablets from a stale membership snapshot but the leader of that group has crashed, been restarted, or has not yet been elected when deletePredicates calls s.Leader(gid).

Common situations: Alpha process in the group crashed or is restarting during a schema drop; network partition between Zero and the group's Alphas; a single-node group whose member is down; dropping predicates right after resizing a cluster.

Related errors

• group: [%d] is not a known group
• namespace: %d. No tablet found for: %s
• namespace: %d. Tablet: [%s] is already being served by group
• Unable to move reserved predicate %s
• Tablet to be moved: [%v] is not being served

AI-assisted analysis of dgraph-io/dgraph@759e242be6 (2026-09-01). Data as JSON: /api/errors/f67b87bcc0f08087. Report an issue: GitHub.