hyperledger-labs / fabric-smart-client / 41292654539

92%

Build Type

push

github

Committed by web-flow

Commit Message

feat(node)!: drain running views on shutdown (#2417)

On SIGTERM, SIGINT or `Node.Stop` a node now lets running views finish before it stops, and stops its components in an order where none outlives what it uses. Before, `Node.Stop` cancelled a context and returned, and the web, gRPC and operations servers, the KVS, the Fabric network provider and the comm service each tore down in a goroutine of their own, so the KVS could close under a component still using it.

The shutdown runs in four steps:
- Drain. `/readyz` answers 503 with reason `draining`, the gRPC health service reports `NOT_SERVING` and `fsc_shutdown_draining` is 1. A new view-API command gets gRPC `Unavailable` (HTTP 503, WebSocket close 1013). A new P2P protocol gets `draining: node is draining`, unless the sender already takes part in the responder context it names.
- Wait for running views, for at most `fsc.shutdown.drainTimeout` (default 30s). At the deadline, initiator views still running are cancelled with `view.ErrCancelledByShutdown`, which the view API returns as gRPC `Aborted` (HTTP 409), since such a view may have submitted a transaction that still commits.
- Cancel the node context.
- Call `Stop()` on every SDK in reverse install order, within `fsc.shutdown.teardownTimeout` (default 30s). The Fabric SDK stops its networks, then the view SDK stops its servers, the comm service and the KVS.

A second signal or `Stop` call skips the rest of the drain.

`pkg/node` finds the drain as a `node.Drainer` in the node's registry, where the view SDK registers the view manager. Wrapper SDKs embed the `commondig.SDK` interface, so a method on the SDK itself would not reach `pkg/node` through them. No method is added to an SDK interface.

The view manager counts initiator views, view-API commands and responders from the moment they are admitted, checking for a drain and counting in one step. A view's context is cancelled only at the drain deadline, not when the view returns. The comm servic... (continued)

Coverage Stats

329 of 362 new or added lines in 19 files covered. (90.88%)

4 existing lines in 2 files now uncovered.

30420 of 33250 relevant lines covered (91.49%)

1283.95 hits per line