Posted it a few days ago, would be nice if someone
could have a look. I assume there is a wait missing, right ?
If so the fix is simple and we can include it in 2.3.0.
Jörn
Jörn Kottmann wrote:
Hi everyone,
to put some load on one of our UIMA AS system I wrote
a small tool which uses the client API to send CASes to
a processing pipeline.
This tool has an endless loop which waits for work
to arrive and then sends it to the processing pipeline.
It basically looks like this:
text = getNextDocument() // returns plain text which should be analyzed
CAS cas = uimASEngine.getCAS();
cas.setDocumentText(text)
uimaAsEngine.sendCAS(cas)
When the tool is started there is usually lots of work waiting
and all available CASes are sent out to the service. It takes a while
until the first CAS is completly processed. Now the
tool gets a new text and waits for a new CAS inside getCAS().
Now I wonder why waiting for a CAS inside getCAS() puts 95 to 100 %
CPU load
on one core on my client machine.
I looked at the place where it waits in
BaseUIMAAsynchronousEngineCommon_impl
around line 536 in getCAS().
Here is the wait loop:
while (running) {
try {
// Wait until the CAS producer adds the CAS to the
CasQueueEntry and
// signals CAS availability.
entry.getSemaphore().acquire();
if (entry.getCas() == null) {
continue;
} else {
return entry.getCas();
}
} finally {
entry.getSemaphore().release();
} } // while
For me it looks like that if entry.getCAS() returns null it immediately
tries again and polls entry.getCAS() as often as possible.
Jörn
