Bug 1200

Summary:	Handle Leak
Product:	TAO	Reporter:	Lukas Gruetzmacher <gruetzmacher>
Component:	ORB	Assignee:	Nanbor Wang <bala>
Status:	RESOLVED FIXED
Severity:	normal
Priority:	P3
Version:	1.2.1
Hardware:	x86
OS:	Windows NT
Bug Depends on:
Bug Blocks:	1277
Attachments:	Example to reproduce the descibed behavoir

Description Lukas Gruetzmacher 2002-05-07 01:42:28 CDT

Running the example - which I'll attach following - you'll see growing Handles 
and Threads:
Start the server.
Start the Clients (up to 50 at the same time).

The server increase the count of Handles uninterrupted (may depend on the 
system load).
The clients starting with a number of Threads and Handles (2 and 69). After 
some minutes some of them increase it's resources to 3 and 75. Why they have a 
different behavoir in face of identical code?

May be the behavoir of the clients are explainable. But the server has a handle 
leak in my eyes.

Comment 1 Lukas Gruetzmacher 2002-05-07 01:43:45 CDT

Created attachment 114 [details]
Example to reproduce the descibed behavoir

Comment 2 Nanbor Wang 2002-05-13 07:39:01 CDT

Lukas-

I dont think this is a problem. I had seen you report a few days back but 
forgot to reply.. 

Here is what is happening
- When multiple clients (with multiple threads) are started, the client starts 
opening up connections on demand. 1.2.1 uses Muxed connection support. When a 
thread searches for an already opened up connection and if it is busy, it 
immediately opens up another connection. The number of handles used up when 
clients are connected *may* not be a good test.

You should allow all the clients to exit and see whether there are handles open 
(when the server still runs). If there are handles used then we have a problem. 
The $TAO_ROOT/tests/Cache_Growth_Test tests for a similar situation. 

Please let us know if you disagree or you feel something else is happening, 
which I am missing

Thanks
Bala

Comment 3 Nanbor Wang 2002-05-13 07:39:46 CDT

I am just resolving the bug. If you dont agree please feel free to reopen the 
bug.

Comment 4 Nanbor Wang 2002-05-13 07:42:03 CDT

Marking it as verified which looks better

Comment 5 Lukas Gruetzmacher 2002-05-13 09:01:05 CDT

I've done just now the following steps:
Server started:
MEM      VIRT     HANDLES    THREADS
4816     2244     71         11
10 Clients connected - after 30 minutes:
6048     3496     255        21
All Clients killed
5316     2668     114        11
10 Clients connected - after 5 minutes:
6076     3484     236        21
after 10 minutes:
6184     3596     266        21
All Clients killed
5332     2684     115        11

So I see a leak here on memory and handels.

Comment 6 Nanbor Wang 2002-05-13 09:19:33 CDT

Lukas-

Thanks! I am reopening the bug! May be you could ask OCI to look into this. If 
not, we will look into this when we get time!

Thanks

Comment 7 Nanbor Wang 2002-06-26 19:25:04 CDT

Assigning the bug to me

Comment 8 Nanbor Wang 2002-06-26 19:33:40 CDT

Accpeting the bug

Comment 9 Nanbor Wang 2002-07-06 22:13:14 CDT

Lukas-

Could you write a simple example and attach it? Looks to me that the Win32 
specific stuff is putting me off. Coudl you please write a test as simple as in 
$TAO_ROOT/tests ?

I have some time and I can investigate this.

Comment 10 Carlos O'Ryan 2002-08-10 21:56:54 CDT

Resource leaks are blockers.

Comment 11 Nanbor Wang 2003-01-15 12:18:25 CST

As far as we know after running some of the tests like Bug_1020_Regression,
Bug_1361_Regression etc. we are confident that this is fixed. Hence I am closing
this bug.