tango-controls-bot
commented
14 years ago Hola Sergi,
I did some tests this morning and effectively there is one case where it seems there is a memory leak. This case is the following: A device command takes more than 3 seconds (default timeout). I Think this is what happens in your case for Modbus device not able to communicate with their "IpHost". After the timeout exception, the client re-send the command. As command 1 is still in execution (timeout is a local exception on the client side), the server now has 2 commands (the old one still in its execution) and the new one (waiting for device monitor). From Valgrind, the leak comes from the omniORB layer. I will check this with Duncan and let you know what happens.
Cheers
Manu
Original comment by: taurel
Hi all,
After putting all (75) our Modbus device servers in a single machine we detected that the memory usage was continuously increasing due to Modbus devices.
After some investigation we detected that the cause were those Modbus device servers who were unable to connect with its IpHost and were throwing timeout exceptions for any command executed.
The problem seems related to the C++ command exception itself, as no increase in memory usage has been observed in other modbus devices. It's easy to reproduce executing a faulty asynchronous command at a regular basis.
Sergi Rubio Manrique
Electronics & Controls Engineer
CELLS - ALBA Synchrotron - Computing Division Carretera BP 1413, de Cerdanyola del Vallès a Sant Cugat del Vallès 08290 Cerdanyola del Vallès (Barcelona) Phone: +34 93 592 43 87
http://www.cells.es/Members/srubio / sergi.rubio@cells.es
Reported by: sergirubio
Original Ticket: tango-cs/bugs/320