MM-61904: Make reliable websockets work in HA (#29489)
We do a cluster request to get the active and dead queues from other nodes in the cluster to sync any missing information. We check the dead queue in the other nodes to see if there's been any message loss or not. Accordingly, we send just the active queue or both active and dead queues. There's still an edge case that is left out where a client could have potentially connected and reconnected to multiple nodes leaving multiple active queues in multiple nodes. We don't handle this scenario because then potentially we need to create a slice of sendQueueSize * number_of_nodes. And then this can happen again, leading to an infinite increase in sendQueueSize. We leave this edge-case to Redis, acknowledging a limitation in our architecture. In this PR, when there's no message loss, we just take the active queue from the last node it connected to. And if there's message loss where the client's seqNum is within the last node's dead queue, we also handle that. But if there's severe message loss where the client's seqNum falls within the dead queue of another node, then we just send the data from that node to reconstruct the data as much as possible. It could be possible to set a new connection ID in this case, but this involves more data transfer always from all nodes and recomputing the state in the requestor node. https://mattermost.atlassian.net/browse/MM-61904 ```release-note NONE ``` Co-authored-by: Mattermost Build <build@mattermost.com>
Этот коммит содержится в:
коммит произвёл
GitHub
родитель
921604cf39
Коммит
cb75a20c54
@@ -74,7 +74,7 @@ type WebConnConfig struct {
|
||||
XForwardedFor string
|
||||
|
||||
// These aren't necessary to be exported to api layer.
|
||||
sequence int
|
||||
sequence int64
|
||||
activeQueue chan model.WebSocketMessage
|
||||
deadQueue []*model.WebSocketEvent
|
||||
deadQueuePointer int
|
||||
@@ -161,9 +161,20 @@ func (ps *PlatformService) PopulateWebConnConfig(s *model.Session, cfg *WebConnC
|
||||
return nil, fmt.Errorf("invalid connection id: %s", cfg.ConnectionID)
|
||||
}
|
||||
|
||||
// Sequence_number must be sent with connection id.
|
||||
// A client must be either non-compliant or fully compliant.
|
||||
if seqVal == "" {
|
||||
return nil, errors.New("sequence number not present in websocket request")
|
||||
}
|
||||
|
||||
seqNum, err := strconv.ParseInt(seqVal, 10, 0)
|
||||
if err != nil {
|
||||
return nil, fmt.Errorf("invalid sequence number %s in query param: %w", seqVal, err)
|
||||
}
|
||||
|
||||
// This does not handle reconnect requests across nodes in a cluster.
|
||||
// It falls back to the non-reliable case in that scenario.
|
||||
res := ps.CheckWebConn(s.UserId, cfg.ConnectionID)
|
||||
res := ps.CheckWebConn(s.UserId, cfg.ConnectionID, seqNum)
|
||||
if res == nil {
|
||||
// If the connection is not present, then we assume either timeout,
|
||||
// or server restart. In that case, we set a new one.
|
||||
@@ -175,17 +186,7 @@ func (ps *PlatformService) PopulateWebConnConfig(s *model.Session, cfg *WebConnC
|
||||
cfg.deadQueuePointer = res.DeadQueuePointer
|
||||
cfg.Active = false
|
||||
cfg.ReuseCount = res.ReuseCount
|
||||
// Now we get the sequence number
|
||||
if seqVal == "" {
|
||||
// Sequence_number must be sent with connection id.
|
||||
// A client must be either non-compliant or fully compliant.
|
||||
return nil, errors.New("sequence number not present in websocket request")
|
||||
}
|
||||
var err error
|
||||
cfg.sequence, err = strconv.Atoi(seqVal)
|
||||
if err != nil || cfg.sequence < 0 {
|
||||
return nil, fmt.Errorf("invalid sequence number %s in query param: %v", seqVal, err)
|
||||
}
|
||||
cfg.sequence = seqNum
|
||||
}
|
||||
return cfg, nil
|
||||
}
|
||||
@@ -235,7 +236,7 @@ func (ps *PlatformService) NewWebConn(cfg *WebConnConfig, suite SuiteIFace, runn
|
||||
send: cfg.activeQueue,
|
||||
deadQueue: cfg.deadQueue,
|
||||
deadQueuePointer: cfg.deadQueuePointer,
|
||||
Sequence: int64(cfg.sequence),
|
||||
Sequence: cfg.sequence,
|
||||
WebSocket: cfg.WebSocket,
|
||||
lastUserActivityAt: model.GetMillis(),
|
||||
UserId: cfg.Session.UserId,
|
||||
@@ -655,34 +656,48 @@ func (wc *WebConn) addToDeadQueue(msg *model.WebSocketEvent) {
|
||||
// hasMsgLoss indicates whether the next wanted sequence is right after
|
||||
// the latest element in the dead queue, which would mean there is no message loss.
|
||||
func (wc *WebConn) hasMsgLoss() bool {
|
||||
return _hasMsgLoss(wc.deadQueue, wc.deadQueuePointer, wc.Sequence)
|
||||
}
|
||||
|
||||
// isInDeadQueue checks whether a given sequence number is in the dead queue or not.
|
||||
// And if it is, it returns that index.
|
||||
func (wc *WebConn) isInDeadQueue(seq int64) (bool, int) {
|
||||
return _isInDeadQueue(wc.deadQueue, seq)
|
||||
}
|
||||
|
||||
// _hasMsgLoss is called from 2 places: wc.hasMsgLoss and ps.GetWSQueues.
|
||||
// It is done this way because it is difficult to call wc.hasMsgLoss from inside
|
||||
// ps.GetWSQueues
|
||||
func _hasMsgLoss(deadQueue []*model.WebSocketEvent, deadQueuePtr int, seq int64) bool {
|
||||
var index int
|
||||
// deadQueuePointer = 0 means either no msg written or the pointer
|
||||
// has rolled over to its starting position.
|
||||
if wc.deadQueuePointer == 0 {
|
||||
// If last entry is nil, it means no msg is written.
|
||||
if wc.deadQueue[deadQueueSize-1] == nil {
|
||||
if deadQueuePtr == 0 {
|
||||
// If first entry is nil, it means no msg is written.
|
||||
if deadQueue[0] == nil {
|
||||
return false
|
||||
}
|
||||
// If it's not nil, that means it has rolled over to start, and we
|
||||
// check the last position.
|
||||
index = deadQueueSize - 1
|
||||
} else { // deadQueuePointer != 0 means it's somewhere in the middle.
|
||||
index = wc.deadQueuePointer - 1
|
||||
index = deadQueuePtr - 1
|
||||
}
|
||||
|
||||
if wc.deadQueue[index].GetSequence() == wc.Sequence-1 {
|
||||
if deadQueue[index].GetSequence() == seq-1 {
|
||||
return false
|
||||
}
|
||||
return true
|
||||
}
|
||||
|
||||
// isInDeadQueue checks whether a given sequence number is in the dead queue or not.
|
||||
// And if it is, it returns that index.
|
||||
func (wc *WebConn) isInDeadQueue(seq int64) (bool, int) {
|
||||
// _isInDeadQueue is called from 2 places: wc.isInDeadQueue and ps.GetWSQueues.
|
||||
// It is done this way because it is difficult to call wc.isInDeadQueue from inside
|
||||
// ps.GetWSQueues
|
||||
func _isInDeadQueue(deadQueue []*model.WebSocketEvent, seq int64) (bool, int) {
|
||||
// Can be optimized to traverse backwards from deadQueuePointer
|
||||
// Hopefully, traversing 128 elements is not too much overhead.
|
||||
for i := 0; i < deadQueueSize; i++ {
|
||||
elem := wc.deadQueue[i]
|
||||
elem := deadQueue[i]
|
||||
if elem == nil {
|
||||
return false, 0
|
||||
}
|
||||
|
||||
Ссылка в новой задаче
Block a user