That defines the expiration policy for ReconfigEvent, but it does not define the relationship between ReconfigEvent and DataReading. Imagine a timeline, dotted with ReconfigEvents... what your rule is saying is for each DataReading event, match it with all the "periods where there is no ReconfigEvent", as the lifetime of DataReading is infinity.
Maybe a simpler way of understanding it (without using the "not" for simplification) is think about this:
declare Man
@livesUpTo( 100 years )
end
rule "marriage"
when
$w : Woman( age > 18 )
$m : Man( )
then
// marry woman and man
end
The rule above defines that each woman will merry every existing man, even if each of them only lives for 100 years.
declare ReconfigEvent
@role( event )
@expires ( 5s )
end
would cover that, meaning that every ReconfigEvent would expire
after 5s, thus causing to no longer be needed by the engine.
Thanks,
Chris
On 7/25/2011 12:17 PM, Edson Tirelli wrote:
Chris,
I haven't tried your code, but just looking at the rule,
there is no temporal correlation between DataReading and
ReconfigEvent, so if I am not mistaken, that will create a
required interval of infinite time when the temporal reasoning
is applied, meaning the engine has to keep the readings in
memory forever.
Adding a temporal constraint on the ReconfigEvent pattern
on your rule or defining an explicit expiration policy for the
DataReadings are ways to work around that.
I am performing a simple test of injecting an event every 1
millisecond
like so:
for (int x = 0; x < 100000; x++){
DataReading dr = new DataReading("Reading " + x,
12.0f);
myStream.insert(dr);
ksession.fireAllRules();
Thread.sleep(1);
}
The rule that evaluates this is simple. It basically delays
"then" for
3s to see if a followup reading is inserted and makes sure
that no
ReconfigEvent is active(5s expiration).
So if a reading comes in and a followup reading is not
inserted within 3
seconds and there is is not an existing ReconfigEvent event
alive, then
it should output and insert a ReconfigEvent, essentially
disabling any
DataReading action on those events for the next 5 seconds or
so. This
all works just fine as expected.
My question is, how come I don't get memory back when all
100,000 of my
events have been inserted. Memory goes up slowly over the
course of
insertions, which I can understand, but once that loop is
finished,
memory never reduces, so essentially, the application will
eventually
run out of memory after some time. I should not have to
explicitly
remove/retract events should I? Shouldn't they be removed
from working
memory as soon as they are no longer viable? What should I
be doing to
reclaim memory from the session/knowledgebase?
I have included the full Main program here and the
Sample.drl file below it.
for (int x = 0; x < 100000; x++){
DataReading dr = new DataReading("Reading " + x,
12.0f);
myStream.insert(dr);
ksession.fireAllRules();
Thread.sleep(1);
declare ReconfigEvent
@role( event )
@expires ( 5s )
end
rule "Wait for follup reading or no config"
//lock-on-active
when
$dr: DataReading(reading > 10.0) from entry-point "My
Stream"
not(DataReading(reading < 10.0, this after[0s,3s]
$dr) from
entry-point "My Stream")
not(ReconfigEvent() from entry-point "My Stream")
then
System.err.println("Action: " + new
Date(System.currentTimeMillis()));
System.err.println("Data reading " + $dr.getName() + "
> 10");
ReconfigEvent rce = new ReconfigEvent();
entryPoints["My Stream"].insert( rce );