Bug 617

Summary: ungraceful failure when client limit is reached on NT
Product: TAO Reporter: Marina Spivak <marina>
Component: ORBAssignee: DOC Center Support List (internal) <tao-support>
Status: ASSIGNED ---    
Severity: normal    
Priority: P3    
Version: 1.1.5   
Hardware: x86   
OS: Windows NT   

Description Marina Spivak 2000-07-22 20:53:53 CDT 
When the maximum number (FD_SETSIZE imposed limit) of clients connected to a 
server is reached, the next client attempting to connect to the server is 
forever blocked in its own reactor.  Seems like an 'ungraceful' failure.  
Ideally, the client call should either result in failure or it should block
until a handle in the server frees up, and succeed ...

The behaviour descirbed above was observed with TAO server running on NT, and 
all clients on Solaris 5.7.  I don't know whether same behaviour is present 
with other platform combinations, e.g., NT/NT, Solaris/Solaris, etc.
Comment 1 Irfan Pyarali 2000-07-23 15:01:59 CDT 
Sounds like connection recycling is involved. Kirthika, can you please take a 
look at this?
Comment 2 kirthika 2000-07-23 15:43:12 CDT 
Put it on my Todo list. Marina could you provide a stack trace if possible?
Comment 3 Marina Spivak 2000-12-03 23:26:52 CST 
Bala has revamped connection management so this may or may not still be a 
problem.  We need to test it out.  A simple test is to set FD_SETSIZE to some 
small value, start a simple server, and have a perl script start a bunch of 
clients (# clients > FD_SETSIZE). Each client establishes a connection to the 
server (i.e., invokes some method) and then sleeps for some amount of time 
before exiting (to achieve the first client still being connected to the server 
when the last one is trying to connect).  If everything is fine, all the 
clients will eventually terminate.  If the problem is still there, some of the 
clients will get forever blocked.

To tao-support pool.
Comment 4 Marina Spivak 2000-12-03 23:29:12 CST 
Accepting for tao-support.