DataChannel.readLoop goroutine leak #2098
Description
Your environment.
- Version: Release or SHA
- Browser: include version
- Other Information - stacktraces, related issues, suggestions how to fix, links for us to have context
What did you do?
I got goroutine leak from our system which is based on ion-sfu.
(dlv) bt
0 0x000000000043cc65 in runtime.gopark
at /__w/_tool/go/1.16.4/x64/src/runtime/proc.go:337
1 0x000000000046fe38 in runtime.goparkunlock
at /__w/_tool/go/1.16.4/x64/src/runtime/proc.go:342
2 0x000000000046fe38 in sync.runtime_notifyListWait
at /__w/_tool/go/1.16.4/x64/src/runtime/sema.go:513
3 0x000000000047b879 in sync.(*Cond).Wait
at /__w/_tool/go/1.16.4/x64/src/sync/cond.go:56
4 0x000000000071fd56 in github.com/pion/sctp.(*Stream).ReadSCTP
at /github/home/go/pkg/mod/github.com/pion/sctp@v1.8.2/stream.go:108
5 0x00000000007271bd in github.com/pion/datachannel.(*DataChannel).ReadDataChannel
at /github/home/go/pkg/mod/github.com/pion/datachannel@v1.5.2/datachannel.go:186
6 0x00000000007ee3da in git.dev.hpcnt.com/hyperconnect/pion-webrtc/v3.(*DataChannel).readLoop
at /github/home/go/pkg/mod/git.dev.hpcnt.com/hyperconnect/pion-webrtc/v3@v3.1.2-hpcnt.release/datachannel.go:324
As you can see, Stream.ReadSCTP hangs while waiting for signal at Stream.readNotifier.
at that moment association and dataChannel have been already closed.
(dlv) p s.association.readLoopCloseCh
chan struct {} {
qcount: 0,
dataqsiz: 0,
buf: *[0]struct struct {} [],
elemsize: 0,
closed: 1,
elemtype: *runtime._type {size: 0, ptrdata: 0, hash: 670477339, tflag: tflagExtraStar|tflagRegularMemory (10), align: 1, fieldAlign: 1, kind: 25, equal: runtime.memequal0, gcdata: *0, str: 56587, ptrToThis: 543648},
sendx: 0,
recvx: 0,
recvq: waitq<struct {}> {
first: *sudog<struct {}> nil,
last: *sudog<struct {}> nil,},
sendq: waitq<struct {}> {
first: *sudog<struct {}> nil,
last: *sudog<struct {}> nil,},
lock: runtime.mutex {
lockRankStruct: runtime.lockRankStruct {},
key: 0,},}
(dlv) p s.association.closeWriteLoopCh
chan struct {} {
qcount: 0,
dataqsiz: 0,
buf: *[0]struct struct {} [],
elemsize: 0,
closed: 1,
elemtype: *runtime._type {size: 0, ptrdata: 0, hash: 670477339, tflag: tflagExtraStar|tflagRegularMemory (10), align: 1, fieldAlign: 1, kind: 25, equal: runtime.memequal0, gcdata: *0, str: 56587, ptrToThis: 543648},
sendx: 0,
recvx: 0,
recvq: waitq<struct {}> {
first: *sudog<struct {}> nil,
last: *sudog<struct {}> nil,},
sendq: waitq<struct {}> {
first: *sudog<struct {}> nil,
last: *sudog<struct {}> nil,},
lock: runtime.mutex {
lockRankStruct: runtime.lockRankStruct {},
key: 0,},}
(dlv) p s.association.streams
map[uint16]*github.com/pion/sctp.Stream [
0: *{
association: *(*"github.com/pion/sctp.Association")(0xc0431f9500),
lock: (*sync.RWMutex)(0xc01a279e08),
streamIdentifier: 0,
defaultPayloadType: PayloadTypeWebRTCBinary (53),
reassemblyQueue: *(*"github.com/pion/sctp.reassemblyQueue")(0xc072a4a840),
sequenceNumber: 1,
readNotifier: *(*sync.Cond)(0xc04c6bea80),
readErr: error nil,
writeErr: error nil,
unordered: false,
reliabilityType: 0,
reliabilityValue: 0,
bufferedAmount: 19,
bufferedAmountLow: 0,
onBufferedAmountLow: nil,
log: github.com/pion/logging.LeveledLogger(*github.com/pion/logging.DefaultLeveledLogger) ...,
name: "0:0xc0431f9500",},
]
I presume while SCTPTransport.Start is in progress, Association.readLoop is closed and
Association.unregisterStream which is defer action does not affect if Association.OpenStream not called yet.
I don't know exactly what causes this. I think may be It happens when peerConnection closed while data channel connection is being established.