Details
-
Bug
-
Status: Open
-
Major
-
Resolution: Unresolved
-
1.6.x
-
None
-
Windows XP
Description
Hello,
In my company, we use a homemade Windows service to launch processes on remote
servers (one service per server). A central scheduling program connects to the
remote service and then fires jobs on the machine.
This way, backup of our SVN repository is scheduled and the command
"svnadmin.exe dump D:\SVN_repos > D:\SVN_backups\myfile.dmp" is launched
through this Windows service.
While testing and stressing this solution, I noticed svnadmin.exe has been
killed during its dump when I was logging off the server ("2010-01-05
16:23:12,682 TRACE [Service] svnadmin: Caught signal" in the log). After
searching a bit on my favourite search engine, I realized that, on Windows,
every process (even services!) receives a CTRL_LOGOFF_EVENT, and if this signal
is not caught by the program, process is killed. Each service on Windows, event
started with LocalSystem account receives a CTRL_LOGOFF_EVENT (cf
http://msdn.microsoft.com/en-us/library/aa376876%28VS.85%29.aspx). As I
understand a program must implement a HandlerRoutine Callback function to ignore
such events (http://msdn.microsoft.com/en-us/library/ms683242%28VS.85%29.aspx).
What I could suggest is to add a parameter to svnadmin to reduce OS signals
usage [like -Xrs flag for several JVMs].
Thanks in advance for your help,
Nicolas
PS : I entered Windows XP as OS but in fact it's Windows 2003 (not in the combo
list)
Original issue reported by nhannebicque