Hi Geoff
I'm truly sorry for asking so many questions. I only write after trying a
lot of different options, looking through source, test cases and examples.
It is tough going when one doesn't have as much familiarity as you.
Here's a paraphrase of what I've configured....
Application
*service1*
name "hts"
entity - an API server based on VanillaJavaApp (Jetty under the
hood) that has a very long initialization phase and make take a minute or
more to be ready. It has a simple /up url for testing whether it is
ready to start processing. As soon as it returns a 200 status and this:
{"server_version":"1.0.3 built 07-02-2019"} , it is ready:
* service2*
entity: DynamicCluster of about 20 instances of
Another VanillaJavaApp client application that works its
way through a chunk of a large data set sending requests to the above
server, beginning as soon as it is ready.
This client app exits if the server is not present via the
/up command mentioned above. So it should not even be started until then
*What I'd like to do, is to retard the spawning of the cluster until the
server side is ready (not just running)*
*My strategy was to create an HttpRequestSensor in service1 that would poll
that /up command*
brooklyn.initializers:
- type: org.apache.brooklyn.core.sensor.http.HttpRequestSensor
brooklyn.config:
period: 20000
name: ctakesReady
targetType: java.lang.String
URI:
$brooklyn:formatString("http://localhost:%s/up",$brooklyn:config("port")
)
suppressDuplicates: true
jsonPath: "/server_version"
*Then to create a variable in the server entity's map, which transformed
the sensor result into a boolean-like value*
server.isUp:
regexReplacement:
- $brooklyn:sensor("ctakesReady")
- ".*"
- "true"
*And finally, in the dynamicCluster's startup to add a latch*
id: client_cluster
type: org.apache.brooklyn.entity.group.DynamicCluster
brooklyn.config:
latch.start:
$brooklyn:component("hts").attributeWhenReady("server.isUp")
cluster.initial.size: 2
dynamiccluster.memberspec:
$brooklyn:entitySpec:
......
I've tested all the bits in Brooklyn separately: I can launch the server,
I can create the cluster, I can get that cluster to talk to a server I
launch manually, but I just can't tie it all together
In the logs I can see that the HttpRequestSensor task is scheduling each
20s but I can't tell if it is doing what I want. I can't tell what
happens when the HttpRequestSensor receives a connection refused... which
it will do for the first minute or two.
The only thing I can tell is I'm getting the following error, maybe because
it is trying to access ctakesReady before the sensor has a chance to create
it.
2019-08-03T12:37:27,005 DEBUG 123 o.a.b.c.f.Poller [er-OjlhJH1g-3908]
PollJob for org.apache.brooklyn.feed.http.HttpFeed$2@6fb72eb8 handling
java.lang.NullPointerException using
org.apache.brooklyn.core.feed.DelegatingPollHandler@19d3dd8a[ctakesReady @
paxy3zlxqx <- http[ctakesReady]]
2019-08-03T12:37:27,005 DEBUG 123 o.a.b.c.f.AttributePollHandler
[er-OjlhJH1g-3908] Read of VanillaJavaAppImpl{id=paxy3zlxqx}->Sensor:
ctakesReady (java.lang.String) gave exception (in grace period):
java.lang.NullPointerException
......
2019-08-03T12:38:07,011 WARN 123 o.a.b.c.f.AttributePollHandler
[er-OjlhJH1g-3907] Read of VanillaJavaAppImpl{id=paxy3zlxqx}->Sensor:
ctakesReady (java.lang.String) gave exception (grace period expired,
occurring for 40s 6ms): java.lang.NullPointerException
I feel as if there should be a way to create that HTTP sensor with a
starting value. Also I'm sure there may be a better way to link the
success of the HTTPSensor to the latch without the regex transformation,
but I just can't find a good example. All of the examples just use a
handy boolean attribute value that's already present in the entity
definition.
There's need to solve my problem completely - but if you can point me in
the right direction, that would be great.
Many thanks
Peter
