Yes, what Svet said is exactly right. I want to know the status of a
boolean sensor which can usefully be set to false. I think the behaviour is
currently is rather surprising - well, it surprised me when I realised my
blueprint didn't work because I was waiting on a false value!
I'd be happy with adding extra arguments to `attributeWhenReady` so that we
could retain backward compatibility. Something like the following:
1. These are equivalent, and work the same way as the existing DSL function
(see first example in the list)
$brooklyn:entity("whatever").attributeWhenReady("service.isUp")
$brooklyn:entity("whatever").attributeWhenReady:
- $brooklyn:sensor("service.isUp")
$brooklyn:entity("whatever").attributeWhenReady:
- service.isUp
$brooklyn:entity("whatever").attributeWhenReady:
- sensor: service.isUp
$brooklyn:entity("whatever").attributeWhenReady:
- sensor: $brooklyn:sensor("service.isUp")
2. This uses the new mechanism in
https://github.com/apache/brooklyn-server/pull/291 to create an object from
a factory method that provides the readiness predicate
$brooklyn:entity("whatever").attributeWhenReady:
- sensor: $brooklyn:sensor("cluster.first")
ready:
$brooklyn:object:
- type: org.apache.brooklyn.predicates.EntityPredicates
factoryMethod.name: equalTo
factoryMethod.args: true
3. This uses existing $brooklyn:object syntax to create a readiness
predicate from a known classname that has a no-arg ctor
$brooklyn:entity("whatever").attributeWhenReady:
- sensor: $brooklyn:sensor("cluster.first")
ready:
$brooklyn:object:
- type: org.apache.brooklyn.predicates.EntityPredicates.NotNull
$brooklyn:entity("whatever").attributeWhenReady:
- sensor: $brooklyn:sensor("cluster.first")
ready:
$brooklyn:object:
- type: org.apache.brooklyn.predicates.EntityPredicates.EqualTo
object.fields:
equalTo: true
4. Here we use a new DSL function that understands how to create predicate
objects
$brooklyn:entity("whatever").attributeWhenReady:
- sensor: $brooklyn:sensor("cluster.first")
ready: $brooklyn:predicate:equalTo(true)
5. Here we allows string constants to describe some well known predicates
that would be looked up from an internal list
$brooklyn:entity("whatever").attributeWhenReady:
- sensor: $brooklyn:sensor("cluster.first")
ready: "notNul"l
$brooklyn:entity("whatever").attributeWhenReady:
- sensor: $brooklyn:sensor("cluster.first")
ready: "isTrue"
$brooklyn:entity("whatever").attributeWhenReady:
- sensor: $brooklyn:sensor("cluster.first")
ready: "immediate"
As you can see, the new functionality we want will require some changes to
the DSL no matter what, although #291 is promising in terms of allowing us
to create predicates on the fly in the DSL. I'm not sure about an explicit
`$brooklyn:predicate` addition, it seems far too specific. I think the
current functionality is too restrictive though, since it would require a
new class for every predicate, structured as a Java bean with appropriate
setters. One solution is to use the mechanism employed by enrichers, where
they accept strings indicating the function to be applied, "average", "sum"
and so on, and allow "notNull" or "immediate" as an argument for the
readiness predicate.
I'm not sure about overloading this into a description of the expression,
as Alex suggests. If we start to allow "= false", "> 4", "equalTo
'RUNNING'" and so on then we will probably end up writing a parser for some
limited DSL grammar for the predicate clauses, and this doesn't seem to be
in the spirit of the blueprint language. If we want that, why not go the
JSR-223 route, and accept embedded code as the predicate declaration:
6. Use JSR-223 to embed the code for the predicate
$brooklyn:entity("whatever").attributeWhenReady:
- sensor: $brooklyn:sensor("service.state")
ready.lang: "JavaScript"
ready.code: |
function accept(input) {
if (input == "RUNNING") return true;
if (input.startsWith("STOP") return true;
return false;
}
But, I think 2., 3. and 5. above are the way to go, assuming #291 is merged
and we can use its new syntax for objects.
Cheers,
Andrew.
On Mon, 8 Aug 2016 at 09:56 Svetoslav Neykov <
[email protected]> wrote:
> Initial reason for the change is doing "attributeWhenReady" on boolean
> sensor with "false" as the expected value. That blocks indefinitely with
> the current implementation.
>
> Using an enricher wouldn't allow for customizing the timeout - currently
> "indefinite" when starting and 1 min when stopping. We could extend
> "attributeWhenReady" with a timeout argument only (if needed) and defer the
> rest to an enricher.
> As for "option 2" we could fix it in a backwards compatible way:
> * introduce "attributeWhenTruthy", alias to "attributeWhenReady"
> * introduce "attributeWhenPresent" with a notNull && notEmpty check
> * deprecate "attributeWhenReady"
>
> Svet.
>
>
> > On 7.08.2016 г., at 15:03, Alex Heneveld <
> [email protected]> wrote:
> >
> > You could add an isSet enricher then wait on it. I prefer customisation
> in
> > enrichers rather than extending the dsl.
> >
> > What is the use case? Everywhere I've seen whenReady is the most natural.
> >
> > Best
> > Alex
> >
> > On 5 Aug 2016 17:07, "Aled Sage" <[email protected]> wrote:
> >
> >> Hi all,
> >>
> >> Motivated by Andrew's recent PR [1] and Svet's suggestion, I'd like to
> >> propose a change to the semantics/arguments to
> >> $brooklyn:attributeWhenReady().
> >>
> >> _*Current Behaviour*_
> >>
> >> _Simple example_
> >> If you write in a YAML blueprint something like:
> >>
> >> brooklyn.config:
> >> myurl:
> $brooklyn:component("db").attributeWhenReady("datastore.url")
> >>
> >> then when you attempt to get the config value for "myurl" it will block
> >> until the "datastore.url" sensor value is available on the entity with
> id
> >> "db".
> >>
> >> _"Groovy truth" example_
> >> The attributeWhenReady actually blocks until groovy truth is satisfied
> >> (boolean is true, collection is non-empty, string is non-empty, etc).
> >>
> >> You can therefore write:
> >>
> >> brooklyn.config:
> >> launch.latch:
> >> $brooklyn:component("db").attributeWhenReady("service.isUp")
> >>
> >> and it will block until the db entity's service.isUp sensor is true.
> >>
> >> Unfortunately this means it's impossible to wait for a boolean attribute
> >> to be non-null (i.e. stop waiting when false or true).
> >>
> >> _Java API Functionality_
> >> Comparing the yaml DSL with what can be done in DependentConfiguration
> >> [2], there is a lot we can't do:
> >>
> >> * customize the "readiness" function (i.e. what sensor values mean
> >> "ready")
> >> * define "abort conditions" (e.g. stop waiting if the entity
> >> transitions to some other state)
> >> * specify a timeout
> >> * give a default value for if there is a timeout, or if the entity is
> >> removed while we're waiting.
> >>
> >>
> >> _*Proposal: option 1*_
> >>
> >> My favoured option is to make attributeWhenReady more configurable.
> >>
> >> For example, we could support:
> >>
> >> brooklyn.config:
> >> launch.latch: $brooklyn:component("db").attributeWhenReady:
> >> sensor: "service.isUp"
> >> readiness: $brooklyn:predicates.equalTo("true")
> >> timeout: 10m
> >>
> >> Maybe we'd support overloading of in-lining the args (relying on
> parameter
> >> order of [sensor, readiness, timeout]), such as:
> >>
> >> brooklyn.config:
> >> launch.latch:
> >> $brooklyn:component("db").attributeWhenReady("service.isUp",
> >> $brooklyn:predicates.equalTo("true"), 10m)
> >>
> >> I think I'm ok with leaving the default as groovy-truth semantics, as
> long
> >> as we make the documentation clearer for that.
> >>
> >> However, this proposal might well be hard to implement. It would likely
> >> require significant change to how we parse the $brooklyn expressions
> (if my
> >> recollection of this code is correct).
> >>
> >> _*Proposal: option 2*_
> >>
> >> An alternative suggested by Svet in a comment on [1]:
> >>
> >> Should we fix attributeWhenReady instead? It's confusing to have it
> >> wait indefinitely on false.
> >> Having another attribute, very similar to attributeWhenReady
> >> complicates things even further.
> >>
> >> Suggest renaming attributeWhenRedy to attributeWhenTruth and
> >> changing attributeWhenReady to wait on notNull && notEmpty.
> >>
> >> That sounds like a good idea - it feels much more explicit, and is very
> >> simple to implement.
> >>
> >> However, it would break backwards compatibility (e.g. for those using
> >> `attributeWhenReady("service.isUp")`.
> >>
> >> If we think the semantics are wrong, then we should definitely change it
> >> while we're still on version 0.x rather than ignoring it.
> >>
> >> Aled
> >>
> >> [1] https://github.com/apache/brooklyn-server/pull/282
> >> [2] https://github.com/apache/brooklyn-server/blob/rel/apache-
> >> brooklyn-0.9.0/core/src/main/java/org/apache/brooklyn/core/
> >> sensor/DependentConfiguration.java#L669
> >>
> >>
>
> --
Andrew Kennedy ; Founder clocker.io project ; @grkvlt ; Cloudsoft
