Some Eclipse Foundation services are deprecated, or will be soon. Please ensure you've read this important communication.

Bug 344336

Summary:	Headless mode not flushing CSV logger
Product:	[Technology] STEM	Reporter:	Matthew Davis <mattadav>
Component:	Simulation	Assignee:	Matthew Davis <mattadav>
Status:	CLOSED FIXED	QA Contact:
Severity:	major
Priority:	P3
Version:	unspecified
Target Milestone:	1.2.0
Hardware:	All
OS:	Mac OS X - Carbon (unsup.)
Whiteboard:

Description Matthew Davis

2011-04-29 14:43:20 EDT

Per report on newsgroup:

http://www.eclipse.org/forums/index.php?t=rview&goto=667489&th=208356#msg_667489

Headless mode does not properly flush output to the logger.

Investigating...

Comment 1 Matthew Davis

2011-04-29 14:43:51 EDT

Assigning to myself

Comment 2 Matthew Davis

2011-04-29 15:51:11 EDT

Problem identified.

STEM uses a 10 MB write buffer on the loggers, as in many situations the CSV logger is out biggest bottleneck.  There is no forced flushes while a simulation runs, it only flushes when the buffer fills up (which happens in 10 MB blocks).

The problem with headless vs. GUI is simple.  The two modes use different log wrappers that listen to simulation events.  The GUI mode flushes the buffers on simulation pause and simulation stop.  Headless mode only flushes (and closes the writer) on simulation stop.  

When a simulation completes its sequence, it enters into the PAUSED state.  In GUI, this is fine, as it flushes the output then.  For headless mode, it's not fine, because Java exits immediately after all jobs finish and never sends a STOP event.

So, there are two bugs here:

1.  Headless mode needs to send a STOP event to all simulations when they complete (which is OK, because they will never be used again for that instance).
2.  The CustomCSVLogger (which headless mode uses) should flush the buffers on PAUSE.

A third bug deals more generally with the headless mode, as there's no way to STOP a simulation.  If you forcefully exit (control-c or kill) the JVM before a simulation finishes (which is possible on unbounded sims), there is potential (likely) data loss in the logs.  Will examine this one more in depth.

Comment 3 Matthew Davis

2011-04-29 16:27:31 EDT

Fix committed to trunk at revision 1694.  Merged to 1.2.0 branch at 1695.