6:16 p.m.
Well, I was about to say "you could do this using rule inheritance" but there is
a caveat, and possibly a bug. It almost works. :)
Try this:
rule "BaseStream1Criteria"
enabled false
when
m : Message( foo == "BAR" ) from entry-point "Stream1"
then
System.out.println( "BaseStream1Criteria" );
end
rule "SubRule1" extends "BaseStream1Criteria"
when
Message( status == Message.HELLO ) from m;
then
System.out.println( "SubRule1 " );
end
rule "BaseStream2Criteria"
enabled false
when
m : Message( foo == "BAS" ) from entry-point "Stream2"
then
System.out.println( "BaseStream2Criteria" );
end
rule "SubRule2" extends "BaseStream2Criteria"
when
Message( status == Message.HELLO ) from m;
then
System.out.println( "SubRule2 " );
end
Basically, define one rule for each entry point that has the global criteria for that
entry point. Then extend that rule.
The caveat is that it seems you can't use the events matched in the super-rule in the
RHS of the subrule. (Try putting the object "m" matched into the action of
either of the subrules. It neither validates in eclipse nor compiles at runtime, drools
5.0.1.) If this isn't fixed in trunk (I can't check at the moment) I'll open
a jira.
Code attached. When you run it, this should be the output:
SubRule2
SubRule1
--- On Mon, 11/23/09, Barry Kaplan <groups1(a)memelet.com> wrote:
From: Barry Kaplan <groups1(a)memelet.com>
Subject: [rules-users] Does Session effeciently filter unused facts, or...
To: rules-users(a)lists.jboss.org
Date: Monday, November 23, 2009, 9:12 PM
... should I put a filter before insert().
I will have lots of events coming from hardware devices,
only some of which
be applicable to a given session instance. I've tried
extracting the
declared types from the packages, but that doesn't include
classes used in
patterns which where only imported. So determine the actual
set of event
classes used by a given set of packages seems not to be so
easy.
I've drilled down to EntryPointNode.assertObject......
--
View this message in context:
http://old.nabble.com/Does-Session-effeciently-filter-unused-facts%2C-or....
Sent from the drools - user mailing list archive at
Nabble.com.
_______________________________________________
rules-users mailing list
rules-users(a)lists.jboss.org
https://lists.jboss.org/mailman/listinfo/rules-users
12:47 a.m.
New subject: Does Session effeciently filter unused facts, or...
Greg, I'm not following how this is related to my question. Probably I was
unclear.
Suppose I have 1000 event types/classes, but rules that use only 50 types in
LHS patterns. I will be receiving events of random types at say 100hz. The
question is whether I should just insert the events regardless of whether
any rule actually uses the instances' event type (because the rete-oo will
filter out the unused types very quickly) or whether I should add a filter
before sesssion.insert(..) to let only those types of events that are
actually used by a rule.
My second problem/question then becomes: How to determine which
types/classes are actually used in LHS patterns for a given package? By this
I mean, how can I write code to make this determination.
--
View this message in context:
http://old.nabble.com/Does-Session-effeciently-filter-unused-facts%2C-or....
Sent from the drools - user mailing list archive at Nabble.com.
1:07 a.m.
New subject: Does Session effeciently filter unused facts, or...
Thinking a bit more about it, if the rules author does not care about a
given event type, he will probably not even declare that event type with
@role( event ), meaning the engine will not discard it automatically. By
default the engine will keep this facts around until retracted because at
some point in the future the user might dynamically add rules that match
them, what is clearly not your case...
So, for your problem, I believe pre-filtering is probably your best
option with released versions of Drools. For future versions, I think we
should add a kbase configuration option to tell the engine what to do in
case of facts that don't match any rule: either keep them (as it does today)
or discard them right away. What do you think?
Edson
2009/11/24 Barry Kaplan <groups1(a)memelet.com>
Greg, I'm not following how this is related to my question. Probably I was
unclear.
Suppose I have 1000 event types/classes, but rules that use only 50 types
in
LHS patterns. I will be receiving events of random types at say 100hz. The
question is whether I should just insert the events regardless of whether
any rule actually uses the instances' event type (because the rete-oo will
filter out the unused types very quickly) or whether I should add a filter
before sesssion.insert(..) to let only those types of events that are
actually used by a rule.
My second problem/question then becomes: How to determine which
types/classes are actually used in LHS patterns for a given package? By
this
I mean, how can I write code to make this determination.
--
View this message in context:
http://old.nabble.com/Does-Session-effeciently-filter-unused-facts%2C-or....
Sent from the drools - user mailing list archive at Nabble.com.
_______________________________________________
rules-users mailing list
rules-users(a)lists.jboss.org
https://lists.jboss.org/mailman/listinfo/rules-users
--
Edson Tirelli
JBoss Drools Core Development
JBoss by Red Hat @ www.jboss.com
4:55 a.m.
New subject: Does Session effeciently filter unused facts, or...
Ah, yes Edson. I didn't even think of that. That will probably force my hand
to filter.
But that then leaves me with how to determine the set of classes used in the
rules. Doing it by textual inspection is not viable since rules can be added
as part of the application (by "users').
--
View this message in context:
http://old.nabble.com/Does-Session-effeciently-filter-unused-facts%2C-or....
Sent from the drools - user mailing list archive at Nabble.com.
5:40 a.m.
New subject: Does Session effeciently filter unused facts, or...
In current Drools versions, you will need to use some "internal APIs" for
that, but we can probably add a higher level method call for the next
version.
The LHS of a rule is a tree of RuleConditionElements. What you need to do
is navigate the tree recursively using the getNestedElements() method and
looking for Patterns. In pseudo algorithm:
***************
Package pkg = ...; // obtain a package from the package builder or from the
rulebase
Set<Class<?>> usedClasses = new HashSet<Class<?>>();
for( Rule rule : pkg.getRules() ) {
RuleConditionElement lhs = rule.getLhs();
collectUsedClasses( usedClasses, lhs );
}
**************
public void collectUsedClasses( Set<Class<?>> usedClasses,
RuleConditionElement rce ) {
if( rce instanceof Pattern ) {
ObjectType ot = ((Pattern)rce).getObjectType();
if( ot instanceof ClassObjectType ) {
usedClasses.add( ((ClassObjectType)ot).getClassType() );
}
}
for( RuleConditionElement child : rce.getNestedElements() ) {
collectUsedClasses( usedClasses, child );
}
}
******************
I wrote the above code directly into the e-mail, so there might be some
mistakes, but the idea is there... so you can probably figure out, based on
that.
Edson
2009/11/24 Barry Kaplan <groups1(a)memelet.com>
Ah, yes Edson. I didn't even think of that. That will probably force my
hand
to filter.
But that then leaves me with how to determine the set of classes used in
the
rules. Doing it by textual inspection is not viable since rules can be
added
as part of the application (by "users'). So that pointer on how to
determine
used fact types would be much appreciated.
--
View this message in context:
http://old.nabble.com/Does-Session-effeciently-filter-unused-facts%2C-or....
Sent from the drools - user mailing list archive at Nabble.com.
_______________________________________________
rules-users mailing list
rules-users(a)lists.jboss.org
https://lists.jboss.org/mailman/listinfo/rules-users
--
Edson Tirelli
JBoss Drools Core Development
JBoss by Red Hat @ www.jboss.com
5:02 a.m.
New subject: Does Session effeciently filter unused facts, or...
An option like that would be ideal. However it would also be nice to have the
ability to determine the fact types so filtering pre
fact-handle-creation/insertion can be done as well.
In the meantime, for our current design at least, it will almost always be
the case that we can scale out if we overload drools. ie, Have a separate
thread/process per device (or subset), as the rules/cep processing is
hierarchical. Higher levels will be consuming events generated by lower
levels, but the rate of creation will decrease up the hierarchy.
Edson Tirelli-4 wrote:
For future versions, I think we
should add a kbase configuration option to tell the engine what to do in
case of facts that don't match any rule: either keep them (as it does
today)
or discard them right away. What do you think?
--
View this message in context:
http://old.nabble.com/Does-Session-effeciently-filter-unused-facts%2C-or....
Sent from the drools - user mailing list archive at Nabble.com.
1 a.m.
New subject: Does Session effeciently filter unused facts, or...
Greg, I believe I fixed this bug in trunk a few weeks ago. Although I
confess I did not understood the intent of your rules... :)
Barry, you can certainly go either way. Rete is very efficient on
filtering fact types, but of course there is some overhead on fact handle
creation. If performance is acceptable for you by just inserting events and
letting the engine discard them automatically when not used, I would not
worry about. If you are having performance problems due to the volumes of
non-interesting events and you want to try pre-filtering, I can give you
some pointers on how to find the fact types used by your rules/queries. In
this case, we might even make a feature of that and have the "pre-filter"
being executed by the engine... ;)
Edson
2009/11/24 Greg Barton <greg_barton(a)yahoo.com>
Well, I was about to say "you could do this using rule
inheritance" but
there is a caveat, and possibly a bug. It almost works. :)
Try this:
rule "BaseStream1Criteria"
enabled false
when
m : Message( foo == "BAR" ) from entry-point
"Stream1"
then
System.out.println( "BaseStream1Criteria" );
end
rule "SubRule1" extends "BaseStream1Criteria"
when
Message( status == Message.HELLO ) from m;
then
System.out.println( "SubRule1 " );
end
rule "BaseStream2Criteria"
enabled false
when
m : Message( foo == "BAS" ) from entry-point
"Stream2"
then
System.out.println( "BaseStream2Criteria" );
end
rule "SubRule2" extends "BaseStream2Criteria"
when
Message( status == Message.HELLO ) from m;
then
System.out.println( "SubRule2 " );
end
Basically, define one rule for each entry point that has the global
criteria for that entry point. Then extend that rule.
The caveat is that it seems you can't use the events matched in the
super-rule in the RHS of the subrule. (Try putting the object "m" matched
into the action of either of the subrules. It neither validates in eclipse
nor compiles at runtime, drools 5.0.1.) If this isn't fixed in trunk (I
can't check at the moment) I'll open a jira.
Code attached. When you run it, this should be the output:
SubRule2
SubRule1
--- On Mon, 11/23/09, Barry Kaplan <groups1(a)memelet.com> wrote:
> From: Barry Kaplan <groups1(a)memelet.com>
> Subject: [rules-users] Does Session effeciently filter unused facts,
or...
> To: rules-users(a)lists.jboss.org
> Date: Monday, November 23, 2009, 9:12 PM
>
> ... should I put a filter before insert().
>
> I will have lots of events coming from hardware devices,
> only some of which
> be applicable to a given session instance. I've tried
> extracting the
> declared types from the packages, but that doesn't include
> classes used in
> patterns which where only imported. So determine the actual
> set of event
> classes used by a given set of packages seems not to be so
> easy.
>
> I've drilled down to EntryPointNode.assertObject......
> --
> View this message in context:
http://old.nabble.com/Does-Session-effeciently-filter-unused-facts%2C-or....
> Sent from the drools - user mailing list archive at
> Nabble.com.
>
> _______________________________________________
> rules-users mailing list
> rules-users(a)lists.jboss.org
> https://lists.jboss.org/mailman/listinfo/rules-users
>
_______________________________________________
rules-users mailing list
rules-users(a)lists.jboss.org
https://lists.jboss.org/mailman/listinfo/rules-users
--
Edson Tirelli
JBoss Drools Core Development
JBoss by Red Hat @ www.jboss.com
3:36 a.m.
New subject: Does Session effeciently filter unused facts, or...
The intent was to define a common set of filtering conditions for each combination of
event type and entry point, implemented by the super rule.
--- On Tue, 11/24/09, Edson Tirelli <ed.tirelli(a)gmail.com> wrote:
From: Edson Tirelli <ed.tirelli(a)gmail.com>
Subject: Re: [rules-users] Does Session effeciently filter unused facts, or...
To: "Rules Users List" <rules-users(a)lists.jboss.org>
Date: Tuesday, November 24, 2009, 6:00 PM
Greg, I believe I fixed this bug in trunk a few weeks
ago. Although I confess I did not understood the intent of
your rules... :)
Barry, you can certainly go either way. Rete is very
efficient on filtering fact types, but of course there is
some overhead on fact handle creation. If performance is
acceptable for you by just inserting events and letting the
engine discard them automatically when not used, I would not
worry about. If you are having performance problems due to
the volumes of non-interesting events and you want to try
pre-filtering, I can give you some pointers on how to find
the fact types used by your rules/queries. In this case, we
might even make a feature of that and have the
"pre-filter" being executed by the engine... ;)
Edson
2009/11/24 Greg Barton <greg_barton(a)yahoo.com>
Well, I was about to say "you could do this using rule
inheritance" but there is a caveat, and possibly a bug.
It almost works. :)
Try this:
rule "BaseStream1Criteria"
enabled false
when
m : Message( foo == "BAR"
) from entry-point "Stream1"
then
System.out.println(
"BaseStream1Criteria" );
end
rule "SubRule1" extends
"BaseStream1Criteria"
when
Message( status == Message.HELLO )
from m;
then
System.out.println( "SubRule1
" );
end
rule "BaseStream2Criteria"
enabled false
when
m : Message( foo == "BAS"
) from entry-point "Stream2"
then
System.out.println(
"BaseStream2Criteria" );
end
rule "SubRule2" extends
"BaseStream2Criteria"
when
Message( status == Message.HELLO )
from m;
then
System.out.println( "SubRule2
" );
end
Basically, define one rule for each entry point that has
the global criteria for that entry point. Then extend that
rule.
The caveat is that it seems you can't use the events
matched in the super-rule in the RHS of the subrule. (Try
putting the object "m" matched into the action of
either of the subrules. It neither validates in eclipse
nor compiles at runtime, drools 5.0.1.) If this isn't
fixed in trunk (I can't check at the moment) I'll
open a jira.
Code attached. When you run it, this should be the
output:
SubRule2
SubRule1
--- On Mon, 11/23/09, Barry Kaplan <groups1(a)memelet.com>
wrote:
> From: Barry Kaplan <groups1(a)memelet.com>
> Subject: [rules-users] Does Session effeciently
filter unused facts, or...
> To: rules-users(a)lists.jboss.org
> Date: Monday, November 23, 2009, 9:12 PM
>
> ... should I put a filter before insert().
>
> I will have lots of events coming from hardware
devices,
> only some of which
> be applicable to a given session instance. I've
tried
> extracting the
> declared types from the packages, but that doesn't
include
> classes used in
> patterns which where only imported. So determine the
actual
> set of event
> classes used by a given set of packages seems not to
be so
> easy.
>
> I've drilled down to
EntryPointNode.assertObject......
> --
> View this message in context:
http://old.nabble.com/Does-Session-effeciently-filter-unused-facts%2C-or....
> Sent from the drools - user mailing list archive at
> Nabble.com.
>
> _______________________________________________
> rules-users mailing list
> rules-users(a)lists.jboss.org
> https://lists.jboss.org/mailman/listinfo/rules-users
>
_______________________________________________
rules-users mailing list
rules-users(a)lists.jboss.org
https://lists.jboss.org/mailman/listinfo/rules-users
--
Edson Tirelli
JBoss Drools Core Development
JBoss by Red Hat @ www.jboss.com
-----Inline Attachment Follows-----
_______________________________________________
rules-users mailing list
rules-users(a)lists.jboss.org
https://lists.jboss.org/mailman/listinfo/rules-users
5259
days inactive
5260
days old
7 comments
3 participants
participants (3)
-
Barry Kaplan -
Edson Tirelli -
Greg Barton