Sahan Gamage wrote:
Samisa Abeysinghe wrote:
Ryzhov, Evgeny wrote:
Hi colleagues,
-1-
I tried to reuse client but determined that multiple invocations
followed with memory leaks(about 300k per call) and unclosed sockets.
You can see this on the following test:
1) run axis_http_server.exe
2) run echooo.exe (source echooo.c is applyed)
3) watch out for sockets state before and after sending echo(there
will be sockets in CLOSE_WAIT and FIN_WAIT_2 states)
4) watch out for memory usage on client and server sides
Memory leaks is a known issue. Even 0.93 would have considerable
amount of leaks. Given the amount of cycles the current developer base
has, it is expected that it would only be possible to fix leaks
leading towards 1.0.
-2-
I explored process of socket shutdown in network_handler.c and
determind that
axis2_network_handler_close_socket function returns only after 2 minutes
You can see this on the following test:
1) modify axis2_network_handler_close_socket function in
network_handler.c as shown below:
---
AXIS2_EXTERN axis2_status_t AXIS2_CALL
axis2_network_handler_close_socket (const axis2_env_t *env,
axis2_socket_t socket)
{
int i = 0;
char buf[32];
AXIS2_ENV_CHECK(env, AXIS2_CRTICAL_FAILURE);
if(socket < 0)
{
AXIS2_ERROR_SET(env->error, AXIS2_ERROR_INVALID_SOCKET,
AXIS2_FAILURE);
return AXIS2_FAILURE;
}
shutdown(socket, AXIS2_SHUT_WR);
axis2_network_handler_set_sock_option(env, socket, SO_RCVTIMEO, 1);
AXIS2_LOG_DEBUG(env->log, AXIS2_LOG_SI,
"\n[axis2_network_handler_close_socket, before recv, i=%ld]", i);
i = recv(socket, buf, 32, 0);
AXIS2_LOG_DEBUG(env->log, AXIS2_LOG_SI,
"\n[axis2_network_handler_close_socket, after recv, i=%ld]", i);
AXIS2_CLOSE_SOCKET(socket);
return AXIS2_SUCCESS;
}
---
2) recompile ( make axis2_util ) and copy axis2_util.dll to Axis2 libdir
3) run axis_http_server.exe with loglevel 4
4) run echooo.exe (source echooo.c is applyed) and wait more then 2
minutes before exit
5) watch over axis2.log
What do you deem the reason would be? If I remember correct, there was
this problem with the Java client, which lead to the decision to
introduce a timeout. However 2 minutes is too long, there must be some
constant in the system causing this IMHO, which needs to be found and
fixed.
Please raise a Jira on this.
I believe the problem occurs only in the windows platform because in
other (Linux and Unix) systems, socket obeys the
axis2_network_handler_set_sock_option(env, socket, SO_RCVTIMEO, 1) call
and sets the receive time out to 1 ms. So ideally recv(socket, buf, 32,
0) command should only spend 1ms maximum, but I am not sure how Windows
behaves with socket timeout setting. Can some one working in Windows
platform look into this ? If we can't fix the timeout setting option we
need to change the above piece of code (may be we can remove the recv
function and try out). IIRC, removing this will work perfectly with
axis2c client/server scenario, but may cause problems in interop'ing
with Java client/server.
Rather than removing this, I would prefer to find out how to get it
working on windows. Because inter operating with Java is important.
Samisa...
---------------------------------------------------------------------
To unsubscribe, e-mail: [EMAIL PROTECTED]
For additional commands, e-mail: [EMAIL PROTECTED]
