I will try to explain the problem in开发者_如何学Python shortest possible words. I am using c++ builder 2010.
I am using TIdTCPServer and sending voice packets to a list of connected clients. Everything works ok untill any client is disconnected abnormally, For example power failure etc. I can reproduce similar disconnect by cutting the ethernet connection of a connected client.
So now we have a disconnected socket but as you know it is not yet detected at server side so server will continue to try to send data to that client too.
But when server try to write data to that disconnected client ...... Write() or WriteLn() HANGS there in trying to write, It is like it is wating for somekind of Write timeout. This hangs the hole packet distribution process as a result creating a lag in data transmission to all other clients. After few seconds "Socket Connection Closed" Exception is raised and data flow continues.
Here is the code
try
{
EnterCriticalSection(&SlotListenersCriticalSection);
for(int i=0;i<SlotListeners->Count;i++)
{
try
{
//Here the process will HANG for several seconds on a disconnected socket
((TIdContext*) SlotListeners->Objects[i])->Connection->IOHandler->WriteLn("Some DATA");
}catch(Exception &e)
{
SlotListeners->Delete(i);
}
}
}__finally
{
LeaveCriticalSection(&SlotListenersCriticalSection);
}
Ok i already have a keep alive mechanism which disconnect the socket after n seconds of inactivity. But as you can imagine, still this mechnism cant sync exactly with this braodcasting loop because this braodcasting loop is running almost all the time.
So is there any Write timeouts i can specify may be through iohandler or something ? I have seen many many threads about "Detecting disconnected tcp socket" but my problem is little different, i need to avoid that hangup for few seconds during the write attempt.
So is there any solution ?
Or should i consider using some different mechanism for such data broadcasting for example the broadcasting loop put the data packet in some kind of FIFO buffer and client threads continuously check for available data and pick and deliver it to themselves ? This way if one thread hangs it will not stop/delay the over all distribution thread.
Any ideas please ? Thanks for your time and help.
Regards
Jams
There are no write timeouts implemented in Indy. For that, you will have to use the TIdSocketHandle.SetSockOpt() method to set the socket-level timeouts directly.
The FIFO buffer is a better option (and a better design in general). For example:
void __fastcall TForm1::IdTCPServer1Connect(TIdContext *AContext)
{
...
AContext->Data = new TIdThreadSafeStringList;
...
}
void __fastcall TForm1::IdTCPServer1Disconnect(TIdContext *AContext)
{
...
delete AContext->Data;
AContext->Data = NULL;
...
}
void __fastcall TForm1::IdTCPServer1Execute(TIdContext *AContext)
{
TIdThreadSafeStringList *Queue = (TIdThreadSafeStringList*) AContext->Data;
TStringList *Outbound = NULL;
TStringList *List = Queue->Lock();
try
{
if( List->Count > 0 )
{
Outbound = new TStringList;
Outbound->Assign(List);
List->Clear();
}
}
__finally
{
Queue->Unlock();
}
if( Outbound )
{
try
{
AContext->Connection->IOHandler->Write(Outbound);
}
__finally
{
delete Outbound;
}
}
...
}
...
try
{
EnterCriticalSection(&SlotListenersCriticalSection);
int i = 0;
while( i < SlotListeners->Count )
{
try
{
TIdContext *Ctx = (TIdContext*) SlotListeners->Objects[i];
TIdThreadSafeStringList *Queue = (TIdThreadSafeStringList*) Ctx->Data;
Queue->Add("Some DATA");
++i;
}
catch(const Exception &e)
{
SlotListeners->Delete(i);
}
}
}
__finally
{
LeaveCriticalSection(&SlotListenersCriticalSection);
}
精彩评论