What causes "failed to send 'done' signal for group [%d]: %w" in dgraph-io/dgraph?

Check whether the done Send failed because the stream was already dead — inspect earlier gRPC errors and server logs.

failed to send 'done' signal for group [%d]: %w

Error message

failed to send 'done' signal for group [%d]: %w

What it means

Returned by streamBadger when sending the final Done packet fails (non-EOF). This packet tells the server the group's snapshot stream is complete; without a successful send the server treats the transfer as incomplete and will not finalize the snapshot.

Source

Thrown at dgraph/cmd/dgraphimport/import_client.go:242

		if _, err := out.Recv(); err != nil {
			return fmt.Errorf("failed to receive response for group ID [%v] from the server: %w", groupId, err)
		}
		glog.Infof("[import] Group [%v]: Received ACK for sending data chunk", groupId)

		return nil
	}

	// Execute the stream process
	if err := stream.Orchestrate(ctx); err != nil {
		return fmt.Errorf("stream orchestration failed for group [%v]: %w, badger path: %s", groupId, err, ps.Opts().Dir)
	}

	// Send the final 'done' signal to mark completion
	glog.Infof("[import] Sending completion signal for group [%d]", groupId)
	done := &api.StreamPacket{Done: true}

	if err := out.Send(&api.StreamExtSnapshotRequest{Pkt: done}); err != nil && !errors.Is(err, io.EOF) {
		return fmt.Errorf("failed to send 'done' signal for group [%d]: %w", groupId, err)
	}

	for {
		if ctx.Err() != nil {
			return ctx.Err()
		}
		resp, err := out.Recv()
		if errors.Is(err, io.EOF) {
			return fmt.Errorf("server closed stream before Finish=true for group [%d]", groupId)
		}
		if err != nil {
			return fmt.Errorf("failed to receive final response for group ID [%v] from the server: %w", groupId, err)
		}
		if resp.Finish {
			glog.Infof("[import] Group [%v]: Received final Finish=true", groupId)
			break
		}
		glog.Infof("[import] Group [%v]: Waiting for Finish=true, got interim ACK", groupId)

View on GitHub (pinned to 759e242be6)

Solutions

Check whether the done Send failed because the stream was already dead — inspect earlier gRPC errors and server logs.
Re-run the import for the affected group; a missing done signal means the server never finalized.
Fix stream stability (keepalive, LB timeouts) so the connection survives the whole transfer.
Verify all groups report completion server-side after the import.
Protect the transfer with a context sized for the dataset, not a default short deadline.

Example fix

// before
if err := out.Send(&api.StreamExtSnapshotRequest{Pkt: done}); err != nil && !errors.Is(err, io.EOF) {
	return fmt.Errorf("failed to send 'done' signal for group [%d]: %w", groupId, err)
}
// after
if err := out.Send(&api.StreamExtSnapshotRequest{Pkt: done}); err != nil && !errors.Is(err, io.EOF) {
	return fmt.Errorf("failed to send 'done' signal for group [%d] (transfer incomplete, will need re-run): %w", groupId, err)
}

Defensive patterns

Strategy: try-catch

Validate before calling

select {
case <-ctx.Done():
	return fmt.Errorf("context ended before done-signal; extend deadline: %w", ctx.Err())
default:
}

Type guard

func streamAlreadyDead(err error) bool {
	s, ok := status.FromError(err)
	return ok && (s.Code() == codes.Unavailable || s.Code() == codes.Canceled)
}

Try / catch

if err := out.Send(&api.StreamExtSnapshotRequest{Pkt: done}); err != nil && !errors.Is(err, io.EOF) {
	if streamAlreadyDead(err) {
		return fmt.Errorf("stream died before completion for group %d; full re-import required: %w", groupId, err)
	}
	return fmt.Errorf("failed to send 'done' signal for group [%d]: %w", groupId, err)
}

Prevention

Keep the connection alive for the entire transfer (keepalive, no LB drops).
Avoid Alpha restarts during imports (k8s PDBs, maintenance windows).
Verify server-side completion for every group after import.
Reserve headroom in the context deadline for the final handshake.

When it happens

Trigger: The gRPC stream already broke during data transfer (server crash, connection reset) so the done Send fails; server closed its side after an earlier error; context cancelled at the very end of a long transfer.

Common situations: Import that 'almost finished' but the Alpha restarted or was OOM-killed; LB/proxy dropping the long-lived stream just before completion; cluster churn (leader change) invalidating the stream.

Related errors

AI-assisted analysis of dgraph-io/dgraph@759e242be6 (2026-09-01). Data as JSON: /api/errors/6062435f886544e9. Report an issue: GitHub.