Detecting peer died in multi thread environment

Newbie to this site..

I have a process with multiple threads inside. In one of the threads(A), there is a TCP socket
connection towards a server. Data is being sent from the thread A towards this server. At one point of
time, server is down abruptly, which is detected by the thread A only after 15 minutes.
During this 15 minutes, the data sent by thread A, using unix command "send" are successful.

My queries

1. why, the "send" command doesn't fail immediately? Why it takes 15 minutes for this detection?
2. I tried to catch the signal SIGPIPE in thread A itself using signal function, but it doesn't work, which means "send"
command still is successfull?
3. Should the signals be catched at thread level or process level?

Could you please clarify.

Thanks!

Your connection might have been buffering in which case you might want to set TCP_NODELAY on the socket, this is typically used in situations where each individual send is small but you want it sent right away instead of getting buffered before sending.

Thanks for the info.

Does it still work, even if the Peer host itself crashes or unreachable? Assume, its not the case of Peer process crashing.
Because, in case TCP is actively sending data, in case of Peer host itself crashes or unreachable, TCP will time out(may be after 15 minutes), and socket's pending error is set to ETIMEDOUT.
Will this case, also be solved by using TCP_NODELAY?
Could you please clarify. Thanks!

Originally Posted by raikme At one point of time, server is down abruptly, which is detected by the thread A only after 15 minutes. My queries 1. why, the "send" command doesn't fail immediately? Why it takes 15 minutes for this detection?

May be because server is down, but link still active?

raikme - yes if the peer has crashed you won't know it until bytes are actively sent so either when the keepalive timer kicks one out or your send fills its buffer and sends it out. The TCP_NODELAY tells send not to buffer but send out the data immediately, this should eliminate the delay of getting a SIGPIPE when the connection is down.