Merging performance branch into master (#4268)

* improve performance on sendNotifications

* Fix SQL queries

* Remove get direct profiles, not needed anymore

* Add raw data to error details if AppError fails to decode

* men

* Fix decode (#4052)

* Fixing json decode

* Adding unit test

* Initial work for client scaling (#4051)

* Begin adding paging to profiles API

* Added more paging functionality

* Finish hooking up admin console user lists

* Add API for searching users and add searching to all user lists

* Add lazy loading of profiles

* Revert config.json

* Fix unit tests and some style issues

* Add GetProfilesFromList to Go driver and fix web unit test

* Update etag for GetProfiles

* Updating ui for filters and pagination (#4044)

* Updating UI for pagination

* Adjusting margins for filter row

* Adjusting margin for specific modals

* Adding relative padding to system console

* Adjusting responsive view

* Update client user tests

* Minor fixes for direct messages modal (#4056)

* Remove some unneeded initial load calls (#4057)

* UX updates to user lists, added smart counts and bug fixes (#4059)

* Improved getExplicitMentions and unit tests (#4064)

* Refactor getting posts to lazy load profiles correctly (#4062)

* Comment out SetActiveChannel test (#4066)

* Profiler cpu, block, and memory profiler. (#4081)

* Fix TestSetActiveChannel unit test (#4071)

* Fixing build failure caused by dependancies updating (#4076)

* Adding profiler

* Fix admin_team_member_dropdown eslint errors

* Bumping session cache size (#4077)

* Bumping session cache size

* Bumping status cache

* Refactor how the client handles channel members to be large team friendly (#4106)

* Refactor how the client handles channel members to be large team friendly

* Change Id to ChannelId in ChannelStats model

* Updated getChannelMember and getProfilesByIds routes to match proposal

* Performance improvements (#4100)

* Performance improvements

* Fixing re-connect issue

* Fixing error message

* Some other minor perf tweaks

* Some other minor perf tweaks

* Fixing config file

* Fixing buffer size

* Fixing web socket send message

* adding some error logging

* fix getMe to be user required

* Fix websocket event for new user

* Fixing shutting down

* Reverting web socket changes

* Fixing logging lvl

* Adding caching to GetMember

* Adding some logging

* Fixing caching

* Fixing caching invalidate

* Fixing direct message caching

* Fixing caching

* Fixing caching

* Remove GetDirectProfiles from initial load

* Adding logging and fixing websocket client

* Adding back caching from bad merge.

* Explicitly close go driver requests (#4162)

* Refactored how the client handles team members to be more large team friendly (#4159)

* Refactor getProfilesForDirectMessageList API into getAllProfiles API

* Refactored how the client handles team members to be more large team friendly

* Fix js error when receiving a notification

* Fix JS error caused by current user being overwritten with sanitized version (#4165)

* Adding error message to status failure (#4167)

* Fix a few bugs caused by client scaling refactoring (#4170)

* When there is no read replica, don't open a second set of connections to the master database (#4173)

* Adding connection tacking to stats (#4174)

* Reduce DB writes for statuses and other status related changes (#4175)

* Fix bug preventing opening of DM channels from more modal (#4181)

* 	Fixing socket timing error (#4183)

* Fixing ping/pong handler

* Fixing socket timing error

* Commenting out status broadcasting

* Removing user status changes

* Removing user status changes

* Removing user status changes

* Removing user status changes

* Adding DoPreComputeJson()

* Performance improvements (#4194)

* * Fix System Console Analytics queries
* Add db.SetConnMaxLifetime to 15 minutes
* Add "net/http/pprof" for profiling
* Add FreeOSMemory() to manually release memory on reload config

* Add flag to enable http profiler

* Fix memory leak (#4197)

* Fix memory leak

* removed unneeded nil assignment

* Fixing go routine leak (#4208)

* Merge fixes

* Merge fix

* Refactored statuses to be queried by the client rather than broadcast by the server (#4212)

* Refactored server code to reduce status broadcasts and to allow getting statuses by IDs

* Refactor client code to periodically fetch statuses

* Add store unit test for getting statuses by ids

* Fix status unit test

* Add getStatusesByIds REST API and move the client over to use that instead of the WebSocket

* Adding multiple threads to websocket hub (#4230)

* Adding multiple threads to websocket hub

* Fixing unit tests

* Fixing so websocket connections from the same user end up in the same… (#4240)

* Fixing so websocket connections from the same user end up in the same list

* Removing old comment

* Refactor user autocomplete to query the server (#4239)

* Add API for autocompleting users

* Converted at mention autocomplete to query server

* Converted user search autocomplete to query server

* Switch autocomplete API naming to use term instead of username

* Split autocomplete API into two, one for channels and for teams

* Fix copy/paste error

* Some final client scaling fixes (#4246)

* Add lazy loading of profiles to integration pages

* Add lazy loading of profiles to emoji page

* Fix JS error when receiving post in select team menu and also clean up channel store

1 files changed, 96 insertions, 56 deletions

diff --git a/api/web_hub.go b/api/web_hub.go
index 4a9719d80..5f480880e 100644
--- a/api/web_hub.go
+++ b/api/web_hub.go
@@ -5,6 +5,8 @@ package api
 
 import (
 	"fmt"
+	"hash/fnv"
+	"runtime"
 
 	l4g "github.com/alecthomas/log4go"
 
@@ -14,27 +16,77 @@ import (
 )
 
 type Hub struct {
-	connections       map[*WebConn]bool
-	register          chan *WebConn
-	unregister        chan *WebConn
-	broadcast         chan *model.WebSocketEvent
-	stop              chan string
-	invalidateUser    chan string
-	invalidateChannel chan string
+	connections    map[*WebConn]bool
+	register       chan *WebConn
+	unregister     chan *WebConn
+	broadcast      chan *model.WebSocketEvent
+	stop           chan string
+	invalidateUser chan string
 }
 
-var hub = &Hub{
-	register:          make(chan *WebConn),
-	unregister:        make(chan *WebConn),
-	connections:       make(map[*WebConn]bool),
-	broadcast:         make(chan *model.WebSocketEvent),
-	stop:              make(chan string),
-	invalidateUser:    make(chan string),
-	invalidateChannel: make(chan string),
+var hubs []*Hub = make([]*Hub, 0)
+
+func NewWebHub() *Hub {
+	return &Hub{
+		register:       make(chan *WebConn),
+		unregister:     make(chan *WebConn),
+		connections:    make(map[*WebConn]bool, model.SESSION_CACHE_SIZE),
+		broadcast:      make(chan *model.WebSocketEvent, 4096),
+		stop:           make(chan string),
+		invalidateUser: make(chan string),
+	}
+}
+
+func TotalWebsocketConnections() int {
+	// XXX TODO FIXME, this is racy and needs to be fixed
+	count := 0
+	for _, hub := range hubs {
+		count = count + len(hub.connections)
+	}
+
+	return count
+}
+
+func HubStart() {
+	l4g.Info(utils.T("api.web_hub.start.starting.debug"), runtime.NumCPU()*2)
+
+	// Total number of hubs is twice the number of CPUs.
+	hubs = make([]*Hub, runtime.NumCPU()*2)
+
+	for i := 0; i < len(hubs); i++ {
+		hubs[i] = NewWebHub()
+		hubs[i].Start()
+	}
+}
+
+func HubStop() {
+	l4g.Info(utils.T("api.web_hub.start.stopping.debug"))
+
+	for _, hub := range hubs {
+		hub.Stop()
+	}
+
+	hubs = make([]*Hub, 0)
+}
+
+func HubRegister(webConn *WebConn) {
+	hash := fnv.New32a()
+	hash.Write([]byte(webConn.UserId))
+	index := hash.Sum32() % uint32(len(hubs))
+	hubs[index].Register(webConn)
+}
+
+func HubUnregister(webConn *WebConn) {
+	for _, hub := range hubs {
+		hub.Unregister(webConn)
+	}
 }
 
 func Publish(message *model.WebSocketEvent) {
-	hub.Broadcast(message)
+	message.DoPreComputeJson()
+	for _, hub := range hubs {
+		hub.Broadcast(message)
+	}
 
 	if einterfaces.GetClusterInterface() != nil {
 		einterfaces.GetClusterInterface().Publish(message)
@@ -42,11 +94,19 @@ func Publish(message *model.WebSocketEvent) {
 }
 
 func PublishSkipClusterSend(message *model.WebSocketEvent) {
-	hub.Broadcast(message)
+	message.DoPreComputeJson()
+	for _, hub := range hubs {
+		hub.Broadcast(message)
+	}
 }
 
 func InvalidateCacheForUser(userId string) {
-	hub.invalidateUser <- userId
+
+	Srv.Store.Channel().InvalidateAllChannelMembersForUser(userId)
+
+	for _, hub := range hubs {
+		hub.InvalidateUser(userId)
+	}
 
 	if einterfaces.GetClusterInterface() != nil {
 		einterfaces.GetClusterInterface().InvalidateCacheForUser(userId)
@@ -54,11 +114,17 @@ func InvalidateCacheForUser(userId string) {
 }
 
 func InvalidateCacheForChannel(channelId string) {
-	hub.invalidateChannel <- channelId
 
-	if einterfaces.GetClusterInterface() != nil {
-		einterfaces.GetClusterInterface().InvalidateCacheForChannel(channelId)
-	}
+	// XXX TODO FIX ME
+	// This can be removed, but the performance branch
+	// needs to be merged into master so it can be removed
+	// from the enterprise repo as well.
+
+	// hub.invalidateChannel <- channelId
+
+	// if einterfaces.GetClusterInterface() != nil {
+	// 	einterfaces.GetClusterInterface().InvalidateCacheForChannel(channelId)
+	// }
 }
 
 func (h *Hub) Register(webConn *WebConn) {
@@ -79,6 +145,10 @@ func (h *Hub) Broadcast(message *model.WebSocketEvent) {
 	}
 }
 
+func (h *Hub) InvalidateUser(userId string) {
+	h.invalidateUser <- userId
+}
+
 func (h *Hub) Stop() {
 	h.stop <- "all"
 }
@@ -108,6 +178,7 @@ func (h *Hub) Start() {
 				if !found {
 					go SetStatusOffline(userId, false)
 				}
+
 			case userId := <-h.invalidateUser:
 				for webCon := range h.connections {
 					if webCon.UserId == userId {
@@ -115,26 +186,20 @@ func (h *Hub) Start() {
 					}
 				}
 
-			case channelId := <-h.invalidateChannel:
-				for webCon := range h.connections {
-					webCon.InvalidateCacheForChannel(channelId)
-				}
-
 			case msg := <-h.broadcast:
 				for webCon := range h.connections {
-					if shouldSendEvent(webCon, msg) {
+					if webCon.ShouldSendEvent(msg) {
 						select {
 						case webCon.Send <- msg:
 						default:
+							l4g.Error(fmt.Sprintf("webhub.broadcast: cannot send, closing websocket for userId=%v", webCon.UserId))
 							close(webCon.Send)
 							delete(h.connections, webCon)
 						}
 					}
 				}
 
-			case s := <-h.stop:
-				l4g.Debug(utils.T("api.web_hub.start.stopping.debug"), s)
-
+			case <-h.stop:
 				for webCon := range h.connections {
 					webCon.WebSocket.Close()
 				}
@@ -144,28 +209,3 @@ func (h *Hub) Start() {
 		}
 	}()
 }
-
-func shouldSendEvent(webCon *WebConn, msg *model.WebSocketEvent) bool {
-	// If the event is destined to a specific user
-	if len(msg.Broadcast.UserId) > 0 && webCon.UserId != msg.Broadcast.UserId {
-		return false
-	}
-
-	// if the user is omitted don't send the message
-	if _, ok := msg.Broadcast.OmitUsers[webCon.UserId]; ok {
-		return false
-	}
-
-	// Only report events to users who are in the channel for the event
-	if len(msg.Broadcast.ChannelId) > 0 {
-		return webCon.IsMemberOfChannel(msg.Broadcast.ChannelId)
-	}
-
-	// Only report events to users who are in the team for the event
-	if len(msg.Broadcast.TeamId) > 0 {
-		return webCon.IsMemberOfTeam(msg.Broadcast.TeamId)
-
-	}
-
-	return true
-}