feat: add node rpc & subscription feature#911
Conversation
weigen393
added
the
feat
| } | ||
|
|
||
| type ServiceEvent struct { | ||
| ServiceID string `json:"service_id"` |
| t.Fatalf("First subscribe request returned error: %v", resp1.Error) | ||
| } | ||
| subID1 = uint64(resp1.Result.(float64)) | ||
| t.Logf("First subscription ID: %s", subID1) |
| t.Fatalf("Second subscribe request returned error: %v", resp2.Error) | ||
| } | ||
| subID2 = uint64(resp2.Result.(float64)) | ||
| t.Logf("Second subscription ID: %s", subID2) |
| Method: sub.MethodName, | ||
| Params: NotificationParams{ | ||
| Subscription: sub.ID, | ||
| Result: result, |
| SetupJAMProtocol(chainPath) | ||
|
|
||
| ctx, cancel := context.WithCancel(ctx) | ||
| defer cancel() |
There was a problem hiding this comment.
Blocker — RPC server lifecycle vs process exit
The RPC server is started in a goroutine with go func() { ... rpcServer.Start(ctx) }, while the main goroutine returns immediately after cancel() (triggered by signal) without waiting for the server goroutine to finish shutting down.
context.Cancel + http.Server.Shutdown (inside Start) is asynchronous with respect to the node() return. The process can exit while connections are still open or ListenAndServe / Shutdown is still in progress, leading to unclean termination and lost in-flight work.
Suggestion: use a sync.WaitGroup (or a dedicated done channel) so that after cancel() the main path waits until Start has returned (or until a bounded timeout), matching the 5s shutdown in Start.
|
|
||
| go func() { | ||
| chainState := blockchain.GetInstance() | ||
| eventBus := eventbus.NewEventBus() |
There was a problem hiding this comment.
Architecture — EventBus must be a shared process dependency
eventbus.NewEventBus() is created inside the RPC goroutine and only passed to NewRPCServer. The rest of the node (e.g. blockchain, sync, block import) has no way to Publish on the same bus, so real chain events cannot drive RPC subscribers without duplicating the bus or resorting to polling (e.g. ChainWatcher only).
Suggestion: construct a single EventBus (or EventPublisher / EventSubscriber interfaces) at the node composition root (node() or immediately after SetupJAMProtocol), and inject it into both the RPC server and any component that should emit JIP-2–relevant events, so Publish and Subscribe use the same instance.
2 participants
No description provided.