Lost connection detection time difference

Question

Having a problem understanding why a broken JDBC connection is detected under Windows much quicker than it is under Linux.
I have an process that opens a connection to MSSQL using JDBC, a Stored Procedure is then executed and the result set
processed, this execute step is repeated every 60 seconds or so. If any exception is thrown then resources are closed, sleep
for 60 seconds, the connection is reopened (if fails then sleeps and repeats), once reopened the process continues. There is logging throughout
the above steps.
On Windows Vista 64bit Sun Java platform the process runs fine, if I deliberately break the connection (stop MSSQL service), the process
logs a problems within 60 seconds and starts handling the error, when the MSSQL comes back then within 60 seonds the connection is remade
and the process continues as expected.
Under Ubuntu 10 64bit both Sun and OpenJDK the process runs fine, until the connectionis broken, as it takes about 17 minutes for the exception to be thrown
for the broken connection (stopped MSSQL) to be detected, as opposed to under Windows taking seconds. Once detected then the process works as expected, retrying
and reconnecting as it should.
The libs and code used are the same the only difference between the windows and linux versions are the logging path parameter, have tried with different versions of Java
and with different MSSQL JDBC drivers.
What could be the reason for the time difference in the lost connection detection between Windows and Linux? Any help tips would be most welcome.
Thanks.

Answer 1

The connection is only made once and used throughout, it is only closed if there is a problem. This runs absolutely fine with logging for weeks, the execute and processing take millisecondsIt is pointless to keep the connection in that situation. Create it every time.
Doing that might even fix your problem. But regardless you should still not keep it.
Have jujst checked the timings and if the db connection is broken the on Linux (Ubuntu) the exception generated by step 4. takes up to 15 minutes 35 seconds to move on to step 7. on Windows though it is instant, there is a log message output prior to the execute statement this outputs then the exeception logged.You have an existing connection.
Then you attempt a request which I am guess takes 17 minutes to timeout. (That happens because you are testing by stopping the server.)
That is the problem.
First verify the connection url. It might have timeout values in it.
If not then the next most likely cause is that someone changed the OS (not java) TCP timeout values. Bad idea if they did. And the only way to fix is to change them back to something reasonable. You can code around this by using two threads. One that sets a timer and the other that runs the request. If the timer fires then attempt to cancel the request (see the jdbc api.)