Stop the hanging Goroutines by dropping the old, unidentified TCP streams (#260)

* Close the hanging TCP message channels after a dynamically aligned timeout (base `10000` milliseconds)

* Bring back `source.Lazy`

* Add a one more `sync.Map.Delete` call

* Improve the formula by taking base Goroutine count into account

* Reduce duplication

* Include the dropped TCP streams count into the stats tracker and print a debug log whenever it happens

* Add `superIdentifier` field to `tcpStream` to check if it has identified

Also stop the other protocol dissectors if a TCP stream identified by a protocol.

* Take one step forward in fixing the channel closing issue (WIP)

Add `sync.Mutex` to `tcpReader` and make the loops reference based.

* Fix the channel closing issue

* Improve the accuracy of the formula, log better and multiply `baseStreamChannelTimeoutMs` by 100

* Remove `fmt.Printf`

* Replace `runtime.Gosched()` with `time.Sleep(1 * time.Millisecond)`

* Close the channels of other protocols in case of an identification

* Simplify the logic

* Replace the formula with hard timeout 5000 milliseconds and 4000 maximum number of Goroutines
This commit is contained in:
M. Mert Yıldıran
2021-09-12 08:26:48 +03:00
committed by GitHub
parent 819ccf54cd
commit 858a64687d
10 changed files with 225 additions and 44 deletions

View File

@@ -32,7 +32,7 @@ func init() {
type dissecting string
func (d dissecting) Register(extension *api.Extension) {
extension.Protocol = protocol
extension.Protocol = &protocol
}
func (d dissecting) Ping() {
@@ -41,7 +41,7 @@ func (d dissecting) Ping() {
const amqpRequest string = "amqp_request"
func (d dissecting) Dissect(b *bufio.Reader, isClient bool, tcpID *api.TcpID, counterPair *api.CounterPair, superTimer *api.SuperTimer, emitter api.Emitter) error {
func (d dissecting) Dissect(b *bufio.Reader, isClient bool, tcpID *api.TcpID, counterPair *api.CounterPair, superTimer *api.SuperTimer, superIdentifier *api.SuperIdentifier, emitter api.Emitter) error {
r := AmqpReader{b}
var remaining int
@@ -78,6 +78,10 @@ func (d dissecting) Dissect(b *bufio.Reader, isClient bool, tcpID *api.TcpID, co
var lastMethodFrameMessage Message
for {
if superIdentifier.Protocol != nil && superIdentifier.Protocol != &protocol {
return errors.New("Identified by another protocol")
}
frame, err := r.ReadFrame()
if err == io.EOF {
// We must read until we see an EOF... very important!
@@ -108,9 +112,11 @@ func (d dissecting) Dissect(b *bufio.Reader, isClient bool, tcpID *api.TcpID, co
switch lastMethodFrameMessage.(type) {
case *BasicPublish:
eventBasicPublish.Body = f.Body
superIdentifier.Protocol = &protocol
emitAMQP(*eventBasicPublish, amqpRequest, basicMethodMap[40], connectionInfo, superTimer.CaptureTime, emitter)
case *BasicDeliver:
eventBasicDeliver.Body = f.Body
superIdentifier.Protocol = &protocol
emitAMQP(*eventBasicDeliver, amqpRequest, basicMethodMap[60], connectionInfo, superTimer.CaptureTime, emitter)
default:
body = nil
@@ -134,6 +140,7 @@ func (d dissecting) Dissect(b *bufio.Reader, isClient bool, tcpID *api.TcpID, co
NoWait: m.NoWait,
Arguments: m.Arguments,
}
superIdentifier.Protocol = &protocol
emitAMQP(*eventQueueBind, amqpRequest, queueMethodMap[20], connectionInfo, superTimer.CaptureTime, emitter)
case *BasicConsume:
@@ -146,6 +153,7 @@ func (d dissecting) Dissect(b *bufio.Reader, isClient bool, tcpID *api.TcpID, co
NoWait: m.NoWait,
Arguments: m.Arguments,
}
superIdentifier.Protocol = &protocol
emitAMQP(*eventBasicConsume, amqpRequest, basicMethodMap[20], connectionInfo, superTimer.CaptureTime, emitter)
case *BasicDeliver:
@@ -165,6 +173,7 @@ func (d dissecting) Dissect(b *bufio.Reader, isClient bool, tcpID *api.TcpID, co
NoWait: m.NoWait,
Arguments: m.Arguments,
}
superIdentifier.Protocol = &protocol
emitAMQP(*eventQueueDeclare, amqpRequest, queueMethodMap[10], connectionInfo, superTimer.CaptureTime, emitter)
case *ExchangeDeclare:
@@ -178,6 +187,7 @@ func (d dissecting) Dissect(b *bufio.Reader, isClient bool, tcpID *api.TcpID, co
NoWait: m.NoWait,
Arguments: m.Arguments,
}
superIdentifier.Protocol = &protocol
emitAMQP(*eventExchangeDeclare, amqpRequest, exchangeMethodMap[10], connectionInfo, superTimer.CaptureTime, emitter)
case *ConnectionStart:
@@ -188,6 +198,7 @@ func (d dissecting) Dissect(b *bufio.Reader, isClient bool, tcpID *api.TcpID, co
Mechanisms: m.Mechanisms,
Locales: m.Locales,
}
superIdentifier.Protocol = &protocol
emitAMQP(*eventConnectionStart, amqpRequest, connectionMethodMap[10], connectionInfo, superTimer.CaptureTime, emitter)
case *ConnectionClose:
@@ -197,6 +208,7 @@ func (d dissecting) Dissect(b *bufio.Reader, isClient bool, tcpID *api.TcpID, co
ClassId: m.ClassId,
MethodId: m.MethodId,
}
superIdentifier.Protocol = &protocol
emitAMQP(*eventConnectionClose, amqpRequest, connectionMethodMap[50], connectionInfo, superTimer.CaptureTime, emitter)
default: