Merge pull request #8829 from zl03jsj/fix/TestPaychGetRestartAfterAdd…

…FundsMsg_may_stuck_in_forever_waiting

Fix: PaychGetRestartAfterAddFundsMsg may stuck in forever waiting

paychmgr/manager.go

@@ -87,6 +87,7 @@ func newManager(pchstore *Store, pchapi managerAPI) (*Manager, error) {
 		channels: make(map[string]*channelAccessor),
 		pchapi:   pchapi,
 	}
+	pm.ctx, pm.shutdown = context.WithCancel(context.Background())
 	return pm, pm.Start()
 }
 

paychmgr/mock_test.go

@@ -136,7 +136,7 @@ func newMockPaychAPI() *mockPaychAPI {
 func (pchapi *mockPaychAPI) StateWaitMsg(ctx context.Context, mcid cid.Cid, confidence uint64, limit abi.ChainEpoch, allowReplaced bool) (*api.MsgLookup, error) {
 	pchapi.lk.Lock()
 
-	response := make(chan types.MessageReceipt)
+	response := make(chan types.MessageReceipt, 1)
 
 	if response, ok := pchapi.waitingResponses[mcid]; ok {
 		defer pchapi.lk.Unlock()
@@ -151,8 +151,12 @@ func (pchapi *mockPaychAPI) StateWaitMsg(ctx context.Context, mcid cid.Cid, conf
 	pchapi.waitingCalls[mcid] = &waitingCall{response: response}
 	pchapi.lk.Unlock()
 
-	receipt := <-response
-	return &api.MsgLookup{Receipt: receipt}, nil
+	select {
+	case receipt := <-response:
+		return &api.MsgLookup{Receipt: receipt}, nil
+	case <-ctx.Done():
+		return nil, ctx.Err()
+	}
 }
 
 func (pchapi *mockPaychAPI) receiveMsgResponse(mcid cid.Cid, receipt types.MessageReceipt) {

paychmgr/paychget_test.go

@@ -631,7 +631,7 @@ func TestPaychGetRestartAfterAddFundsMsg(t *testing.T) {
 	require.NoError(t, err)
 
 	// Simulate shutting down system
-	mock.close()
+	require.NoError(t, mgr.Stop())
 
 	// Create a new manager with the same datastore
 	mock2 := newMockManagerAPI()

paychmgr/simple.go

@@ -3,6 +3,7 @@ package paychmgr
 import (
 	"bytes"
 	"context"
+	"errors"
 	"fmt"
 	"sort"
 	"sync"
@@ -351,6 +352,11 @@ func (ca *channelAccessor) queueSize() int {
 // msgWaitComplete is called when the message for a previous task is confirmed
 // or there is an error.
 func (ca *channelAccessor) msgWaitComplete(ctx context.Context, mcid cid.Cid, err error) {
+	// if context is canceled, should Not mark message to 'bad', just return.
+	if errors.Is(err, context.Canceled) || errors.Is(err, context.DeadlineExceeded) {
+		return
+	}
+
 	ca.lk.Lock()
 	defer ca.lk.Unlock()