server/grpc: Hardcoded 1s graceful shutdown timeout causes active gRPC streams to be forcefully terminated

[Grantmartin2002]

Problem

The gRPC server plugin has a hardcoded 1-second timeout for graceful shutdown in both v4 and v5:

select {
case <-exit:
case <-time.After(time.Second):
    g.srv.Stop()
}

After calling GracefulStop(), the server waits only 1 second before calling Stop(), which forcefully terminates all active connections. This is problematic for applications with long-lived gRPC streams (e.g., server-side streaming RPCs) that need more than 1 second to drain.

In Kubernetes environments, this causes ECONNRESET errors for clients with active streams during pod rolling updates, even when terminationGracePeriodSeconds is configured to allow for graceful draining.

Proposed Solution

Make the graceful shutdown timeout configurable via a new GracefulStopTimeout server option, defaulting to time.Second for backwards compatibility.

Usage:

import (
    grpcsrv "github.com/go-micro/plugins/v4/server/grpc"
)

srv := micro.NewService(
    micro.Server(grpcsrv.NewServer(
        grpcsrv.GracefulStopTimeout(3 * time.Minute),
    )),
)

I have a PR ready that implements this for both v4 and v5.

Environment

• go-micro plugins v4 (v1.2.0) and v5
• Kubernetes with gRPC streaming services
• Observed with terminationGracePeriodSeconds: 120 — streams still killed after 1s