[MM-62410] Automatically cleanup stale calls in RTCD enabled deployme…

…nts (#940)

* Update rtcd

* Cleanup stale calls in RTCD enabled deployments

* Add comment

* Bump RTCD to v1.0.0

go.mod

@@ -22,7 +22,7 @@ require (
 	github.com/mattermost/mattermost-plugin-calls/server/public v0.0.3
 	github.com/mattermost/mattermost/server/public v0.1.9
 	github.com/mattermost/morph v1.1.0
-	github.com/mattermost/rtcd v0.18.1-0.20250107081358-290c5ce0a692
+	github.com/mattermost/rtcd v1.0.0
 	github.com/mattermost/squirrel v0.2.0
 	github.com/pkg/errors v0.9.1
 	github.com/rudderlabs/analytics-go v3.3.3+incompatible

go.sum

@@ -375,8 +375,8 @@ github.com/mattermost/mattermost/server/public v0.1.9 h1:l/OKPRVuFeqL0yqRVC/Jpve
 github.com/mattermost/mattermost/server/public v0.1.9/go.mod h1:SkTKbMul91Rq0v2dIxe8mqzUOY+3KwlwwLmAlxDfGCk=
 github.com/mattermost/morph v1.1.0 h1:Q9vrJbeM3s2jfweGheq12EFIzdNp9a/6IovcbvOQ6Cw=
 github.com/mattermost/morph v1.1.0/go.mod h1:gD+EaqX2UMyyuzmF4PFh4r33XneQ8Nzi+0E8nXjMa3A=
-github.com/mattermost/rtcd v0.18.1-0.20250107081358-290c5ce0a692 h1:51Rxv6A5Esd11VW0PNHe0Vw3Zbye7J6iv3tXODTdknU=
-github.com/mattermost/rtcd v0.18.1-0.20250107081358-290c5ce0a692/go.mod h1:hsRBk1e6dQZrZSK2FoqZAhO+oXg+1Je5zhF1ClLbbdA=
+github.com/mattermost/rtcd v1.0.0 h1:7FjRcr7sinOOoSr+in/N/eeg/+BlowkY6m0FFBlM/6M=
+github.com/mattermost/rtcd v1.0.0/go.mod h1:hsRBk1e6dQZrZSK2FoqZAhO+oXg+1Je5zhF1ClLbbdA=
 github.com/mattermost/squirrel v0.2.0 h1:8ZWeyf+MWQ2cL7hu9REZgLtz2IJi51qqZEovI3T3TT8=
 github.com/mattermost/squirrel v0.2.0/go.mod h1:NPPtk+CdpWre4GxMGoOpzEVFVc0ZoEFyJBZGCtn9nSU=
 github.com/mattn/go-colorable v0.0.9/go.mod h1:9vuHe8Xs5qXnSaW/c/ABM9alt+Vo+STaOChaDxuIBZU=

lt/go.mod

@@ -6,7 +6,7 @@ require (
 	github.com/aws/aws-sdk-go v1.50.3
 	github.com/hajimehoshi/go-mp3 v0.3.4
 	github.com/mattermost/mattermost/server/public v0.0.12
-	github.com/mattermost/rtcd v0.18.1-0.20250107081358-290c5ce0a692
+	github.com/mattermost/rtcd v1.0.0
 	github.com/pion/rtp v1.8.10
 	github.com/pion/webrtc/v4 v4.0.7
 	gopkg.in/hraban/opus.v2 v2.0.0-20230925203106-0188a62cb302

lt/go.sum

@@ -85,8 +85,8 @@ github.com/mattermost/logr/v2 v2.0.21 h1:CMHsP+nrbRlEC4g7BwOk1GAnMtHkniFhlSQPXy5
 github.com/mattermost/logr/v2 v2.0.21/go.mod h1:kZkB/zqKL9e+RY5gB3vGpsyenC+TpuiOenjMkvJJbzc=
 github.com/mattermost/mattermost/server/public v0.0.12 h1:iunc9q4/XkArOrndEUn73uFw6v9TOEXEtp6Nm6Iv218=
 github.com/mattermost/mattermost/server/public v0.0.12/go.mod h1:Bk+atJcELCIk9Yeq5FoqTr+gra9704+X4amrlwlTgSc=
-github.com/mattermost/rtcd v0.18.1-0.20250107081358-290c5ce0a692 h1:51Rxv6A5Esd11VW0PNHe0Vw3Zbye7J6iv3tXODTdknU=
-github.com/mattermost/rtcd v0.18.1-0.20250107081358-290c5ce0a692/go.mod h1:hsRBk1e6dQZrZSK2FoqZAhO+oXg+1Je5zhF1ClLbbdA=
+github.com/mattermost/rtcd v1.0.0 h1:7FjRcr7sinOOoSr+in/N/eeg/+BlowkY6m0FFBlM/6M=
+github.com/mattermost/rtcd v1.0.0/go.mod h1:hsRBk1e6dQZrZSK2FoqZAhO+oXg+1Je5zhF1ClLbbdA=
 github.com/matttproud/golang_protobuf_extensions v1.0.1/go.mod h1:D8He9yQNgCq6Z5Ld7szi9bcBfOoFv/3dc6xSMkL2PC0=
 github.com/microcosm-cc/bluemonday v1.0.1/go.mod h1:hsXNsILzKxV+sX77C5b8FSuKF00vh2OMYv+xgHpAMF4=
 github.com/modern-go/concurrent v0.0.0-20180306012644-bacd9c7ef1dd/go.mod h1:6dJC0mAP4ikYIbvyc7fijjWJddQyLn8Ig3JB5CqoB9Q=

server/activate.go

@@ -162,6 +162,10 @@ func (p *Plugin) OnActivate() (retErr error) {
 		p.LogDebug("rtcd client manager initialized successfully")
 
 		p.rtcdManager = rtcdManager
+
+		if err := p.cleanUpState(); err != nil {
+			p.LogError("failed to cleanup state", "err", err.Error())
+		}
 	} else {
 		rtcServerConfig := rtc.ServerConfig{
 			ICEAddressUDP:   cfg.UDPServerAddress,

server/interfaces/interfaces.go

@@ -57,4 +57,6 @@ type RTCDClient interface {
 	Close() error
 	GetVersionInfo() (rtcd.VersionInfo, error)
 	GetSystemInfo() (rtcd.SystemInfo, error)
+	GetSession(callID, sessionID string) (rtc.SessionConfig, int, error)
+	GetSessions(callID string) ([]rtc.SessionConfig, int, error)
 }

server/rtcd.go

@@ -10,6 +10,7 @@ import (
 	"fmt"
 	"math/rand"
 	"net"
+	"net/http"
 	"net/url"
 	"os"
 	"strings"
@@ -21,6 +22,7 @@ import (
 	"github.com/mattermost/mattermost-plugin-calls/server/db"
 	"github.com/mattermost/mattermost-plugin-calls/server/interfaces"
 	"github.com/mattermost/mattermost-plugin-calls/server/license"
+	"github.com/mattermost/mattermost-plugin-calls/server/public"
 
 	rtcd "github.com/mattermost/rtcd/service"
 	"github.com/mattermost/rtcd/service/random"
@@ -706,3 +708,54 @@ func (h *rtcdHost) isFlagged() bool {
 	defer h.mut.RUnlock()
 	return h.flagged
 }
+
+// hasCallEnded checks if the call has ended by querying the RTCD host assigned to the call.
+// Since this method is used to clean up the call state, it's important to be as conservative as possible
+// and only return true if we are absolutely sure the call has ended.
+func (m *rtcdClientManager) hasCallEnded(call *public.Call) bool {
+	m.ctx.LogDebug("RTCD host is set in call, checking...", "callID", call.ID, "rtcdHost", call.Props.RTCDHost)
+	host := m.getHost(call.Props.RTCDHost)
+	if host != nil {
+		m.ctx.LogDebug("RTCD host found", "callID", call.ID, "rtcdHost", call.Props.RTCDHost)
+
+		// Version compatibility check. We need to be talking to RTCD v1.0.0 or higher to be able to call GetSessions.
+		info, err := host.client.GetVersionInfo()
+		if err != nil {
+			m.ctx.LogDebug("failed to get version info", "err", err.Error(), "callID", call.ID, "rtcdHost", call.Props.RTCDHost)
+			return false
+		}
+
+		// Always support dev builds.
+		if info.BuildVersion == "" || info.BuildVersion == "master" || strings.HasPrefix(info.BuildVersion, "dev") {
+			m.ctx.LogDebug("skipping version compatibility check", "buildVersion", info.BuildVersion, "callID", call.ID, "rtcdHost", call.Props.RTCDHost)
+		} else if err := checkMinVersion("v1.0.0", info.BuildVersion); err != nil {
+			m.ctx.LogDebug("RTCD host version is not compatible", "err", err.Error(), "callID", call.ID, "rtcdHost", call.Props.RTCDHost)
+			return false
+		}
+
+		cfgs, code, err := host.client.GetSessions(call.ID)
+		if err != nil {
+			m.ctx.LogDebug("failed to get sessions for call", "err", err.Error(), "callID", call.ID, "rtcdHost", call.Props.RTCDHost)
+		}
+
+		if code != http.StatusOK && code != http.StatusNotFound {
+			m.ctx.LogDebug("unexpected status code from RTCD", "code", code, "callID", call.ID, "rtcdHost", call.Props.RTCDHost)
+			// The request above could fail for various reasons, in which case we can't assume the call has ended.
+			return false
+		}
+
+		if len(cfgs) > 0 {
+			// The call is ongoing so we skip state cleanup.
+			m.ctx.LogDebug("call is still ongoing", "callID", call.ID, "rtcdHost", call.Props.RTCDHost)
+			return false
+		}
+
+		// no call ongoing
+		m.ctx.LogDebug("call was not found", "callID", call.ID, "rtcdHost", call.Props.RTCDHost)
+	} else {
+		// If the host is not found we can assume the RTCD node is gone for good so there's no way a call would be ongoing.
+		m.ctx.LogDebug("RTCD host not found", "callID", call.ID, "rtcdHost", call.Props.RTCDHost)
+	}
+
+	return true
+}

server/state.go

@@ -368,10 +368,19 @@ func (p *Plugin) cleanUpState() error {
 			p.LogError("failed to lock call", "err", err.Error())
 			continue
 		}
+
+		// If a call has a RTCD host assigned, we want to check with the RTCD side whether the call is still ongoing or not before
+		// cleaning up the state.
+		if p.rtcdManager != nil && call.Props.RTCDHost != "" && !p.rtcdManager.hasCallEnded(call) {
+			p.unlockCall(call.ChannelID)
+			continue
+		}
+
 		if err := p.cleanCallState(call); err != nil {
 			p.unlockCall(call.ChannelID)
 			return fmt.Errorf("failed to clean up state: %w", err)
 		}
+
 		p.unlockCall(call.ChannelID)
 	}