tcp is fine.
Also to go down you need to miss 3 retries in a row by default. So it makes even more suspect.
We should investigate why it was reported down even if the proxy responded.
What type of CSS ?
What software version ?
What keepalive frequency ?
What keepalive retries ?
How often do you see the problem ?
Does the service come back up immediately after ?
Sniffer trace available ?
Thanks,
Gilles.