6:21 p.m.
Hello
I have been watching the developments of Hawkular, and am really impressed!
I have a few questions about Hawkular.
Hawkular seems to be primarily focused on server metrics and monitoring.
I am, however, interested in exploring the possibility of using it for
monitoring and alerting of specific business events. For example, lets say
I had an online shop and I sold my widgets via affiliates. Would it be
possible to capture the sales event, persist and display metrics - so the
affiliate partner could login and see their activities? And to send such
events to Hawkular, is there a java client?
When using the server agent, are captured events sent async?
Thanks
--
Anton Hughes
3:26 a.m.
Hi Anton
It sounds like from your scenario, you are just interested in reporting a specific
business event, in which case I believe it is possible to just create your own metric with
data points, using the metrics REST API - however I'll let the members of the metrics
team comment more on that.
If you are interested in recording more information about your business transaction, then
you may want to consider Hawkular BTM. Its currently possible to capture the 'call
trace' of a business transaction instance and analyse the information using Kibana -
you can also extract relevant properties from the txn (e.g. affiliate) and use that
information to focus any queries.
The intention is to integrate at the backend with Hawkular Metrics and Alerts, but this
isn't in place at the moment.
Regards
Gary
----- Original Message -----
4:34 a.m.
----- Original Message -----
BTM uses ByteMan to instrument a java app and the information is then reported to the
server via a REST API - so it would also be possible for apps to directly report the
business transaction info to the server via the same REST API if they wished.
A demo of the latest release can be found here:
https://www.youtube.com/watch?v=vCIgC3QuJTI
Currently business transaction config details are stored in json files, but I'm
currently working on a UI for the next release which should make it a bit easier to work
with.
Not currently. Before BTM could store information in metrics, this issue
https://issues.jboss.org/browse/HWKMETRICS-253 would need to be resolved.
Regards
Gary
7:29 a.m.
Hi Anton,
You're right that at the moment server monitoring is the focus and the
only significant feed is the Wildfly feed. But the APIs should be there
to develop a custom feed if you like, to generate your own resources,
metrics and alert definitions. Also, depending on your needs, you
could find that your main needs are metrics and alerting, maybe BTM as
well, and those components can both be used in a "standalone" mode,
independent of the Hawkular framework. REST APIs are the typical way to
interact with Hawkular. Hawkular Alerts has no built-in UI, I'm not
sure if Metrics comes with any visualization outside of Hawkular.
On 10/26/2015 4:26 AM, Gary Brown wrote:
3:31 a.m.
It does, two solutions:
[1] use hawkular-charts library (preferred)
[2] use Grafana (Metrics has an API adaptor for InfluxDB v0.8 protocol)
[1] https://github.com/hawkular/hawkular-charts
[2] http://www.hawkular.org/docs/components/metrics/grafana_integration.html
Le 27/10/2015 13:29, Jay Shaughnessy a écrit :
I'm not sure if Metrics comes with any visualization outside of
Hawkular.
11:49 a.m.
On 27 October 2015 at 13:29, Jay Shaughnessy <jshaughn(a)redhat.com> wrote:
Thanks.
I wasnt aware that the hawkular components could be run separately.
The BTM module looks interesting, however I am concerned about maintenance.
For example, in the configuration example on
http://www.hawkular.org/docs/components/btm/index.html, if I were to change
the switchyard service names it would break - and I assume there would be
no compile time errors? Therefore, using a strongly typed hawkular client
to communicate with hawkular would be preferred.
With Metrics, is it possible to add custom data? To illustrate, I have the
following use-case:
*log events of interest - *as a product owner I would like to use all user
events of interest so that the user and system auditors can see what and
when a user did something. So, we would log "user John, with user id 1234
logged in at wed 25 oct 17:46. So we would be logging a custom event, and
storing custom data.
Thanks
--
Anton Hughes
12:16 p.m.
Hi Anton
Good point. The reason it is done this way is to give administrators control over what is
monitored at any particular time, rather than having something that needs to be defined at
development time.
At present, if a business transaction is recognised (i.e. by the URI), then it will be
named and processed as defined by its configuration. But as you point out, if something
changes in the service (i.e. its URI is changed), then there is potential for the
administrators to miss this, and subsequently receive no further information about the
business transaction.
At the moment there is a high level property that indicates that any instrumented
activity, that is not associated with a named business txn, should be ignored - by default
this is not enabled, so the server will receive business txn fragments that may or may not
be associated with a business transaction name - this config option enables the unnamed
fragments to be filtered out, but this could potentially miss the case where a service has
had its URI changed, and so will go from being named, to being ignored.
One alternative would be to use the 'exclusion' filter approach to filter out any
fragments that are definitely not of interest, so if a URI change occurred, the fragments
would be reported - but just not have a business transaction name. We could use this fact
to pick up the fact that unnamed fragments are being reported, which could direct an
administrator to fix the problem or add it to the exclusions if no longer of interest.
Although a typed API may still work best in your situation, would the approach outlined
above address your concern?
Regards
Gary
7:45 a.m.
On 28 October 2015 at 18:16, Gary Brown <gbrown(a)redhat.com> wrote:
Yes, I can see that is a good reason. And I think, so long as there is a
traditional in-code way, then you can have options for both parties. I do
see a trend - as part of the devops movement - where developers take on
more operations roles, and ops does more work in code - that the in-code
approach is likely to be favored by many.
We currently use the New Relic agent, and it is my understanding that Jboss
( possibly Wildfly too) can only run one agent at a time? It would be
difficult to convince my team to drop New Relic for Hawkular. And, as
mentioned in the initial thread, our use for Hawkular is more about
monitoring and capturing in code.
I know that, at this stage at least, Hawkular is very focus on server and
system monitoring. Should you wish to explore further the possiblity of
monitoring/capturing and alerting from more user-defined metrics and
events, you may wish to consider using something like the jbpm/drools
modeller. The idea being, a hawkular admin could define the model, then -
again possibly re-use from drools - alert rules could be described using
the user-defined model.
--
Anton Hughes
8:12 a.m.
On 10/28/2015 12:49 PM, Anton Hughes wrote:
Metrics and Alerts can both be used outside of the Hawkular framework so
really you can store any metric you like, or alert on basically any data
you like. As for Events, the next release of Hawkular Alerts (0.6.0)
will include a new Events feature that you may find interesting.
Whereas Alerts are relatively rare, typically involve human interaction,
and run through a simple life-cycle; Events are likely much more
numerous, representing any sort of happening that a client wants to
persist. The interesting thing about Events in HK-Alerts is that they
can be inserted directly via API or can be generated via Trigger, like
an Alert. And Events can also be used as Trigger conditions, to
contribute to further Alert or Event generation.
8:16 a.m.
On 29 October 2015 at 14:12, Jay Shaughnessy <jshaughn(a)redhat.com> wrote:
Thanks Jay - this sounds really cool!
I have heard a few times now that hawkular components can be used outside
of the hawkular framework. What exactly is the hawkular framework? As an
outsider I am learning about Hawkular and its features. There is good
documentation on the features, but the underlying framework, not so much.
Also, regarding documentation, I could not find how to store any 'metric'
or data. Specifically, I am looking to store not just a metric but a pojo.
--
Anton Hughes
8:34 a.m.
Anton, yes, it can be a little confusing. The Hawkular project is an
end-to-end monitoring and management tool focused on Red Hat software.
Today it basically offers a Wildfly agent for discovering and managing
app servers, their hosted apps, and all of the things that make up those
apps. What is can handle grows with every release. Hawkular leverages
a bunch of components to perform that job. There is HK-Inventory to
represent a network of inventories resources (like an app server, a
datasource, a jvm, etc), HK-Metrics as a Cassandra-backed time-series
store, HK-Alerts as a Drools-backed alerting tool, HK-Accounts as a
KeyCloak backed multi-tenant/auth/authz tool, HK-Console for UI, HK-Bus
for a comm backbone, etc..
Some of the HK components, namely HK-Metrics and HK-Alerts support
standalone deployment outside of Hawkular. They are named
Hawkular-Metrics and Hawkular-Alerts because they have been developed as
part of the Hawkular project, but they can be used independently. Hope
that helps...
On 10/29/2015 9:16 AM, Anton Hughes wrote:
8:16 a.m.
Coming back to my original question - and based on some further thinking
and reading of the Hawkular website, I have the following thoughts.
On the Hawkular website, it is written:
For who ?
I've highlighted the general area that I am most interested - and I think
many others would be too.
Please take a quick look at
http://martinfowler.com/eaaDev/EventSourcing.html. Event Sourcing places
emphasis on events of interest - in the Shipping example in this link the
interesting events are:
- Ship Arrives
- Ship Departs
To be able store (store metrics?) and react (trigger alert) in this example
would be very beneficial in many situations.
I hope this helps to illustrates my use-case.
On 29 October 2015 at 14:34, Jay Shaughnessy <jshaughn(a)redhat.com> wrote:
--
Anton Hughes
2:26 a.m.
----- Original Message -----
Hi Anton,
Thanks for giving feedback and illustrate this interesting use case.
We are finishing the Events feature inside the Hawkular Alerts component and I think that
it can work for your requeriments.
Once development is in a state to be used I will add some examples following your use case
(and publish them on the blog).
I hope that might helps.
Lucas
2:39 a.m.
Hi Anton,
We have integrated the Events feature in hawkular-alerts master branch:
https://github.com/hawkular/hawkular-alerts
You can follow the hawkular-alerts doc to build and test the component:
http://www.hawkular.org/docs/components/alerts/index.html
We will release a 0.6.0.Final version shortly.
I have prepared some "HelloWorld" examples to show how to use Events on this
repo:
https://github.com/lucasponce/hawkular-examples/tree/master/events
I want alto to publish a blog post once the release is out with more demo but I didn't
want to delay my response to this topic without any feedback.
The examples are using plain bash scripts over the REST API to show basic features.
Don't hesitate to ask or give feedback, here, github, irc or event JIRA if you find
something broken.
Thanks,
Lucas
----- Original Message -----
4:41 a.m.
Thanks Lucas!
This looks really great!
I will give this some solid testing and let you know how it goes.
One nice feature would be to add swagger documentation to your rest api.
However, your current documentation looks really good!
Kind regards
On 11 November 2015 at 09:39, Lucas Ponce <lponce(a)redhat.com> wrote:
--
Anton Hughes
5:57 a.m.
----- Original Message -----
We have some documentation
http://www.hawkular.org/docs/rest/rest-alerts.html
but this is a point we want to improve when we have some slot between releases.
If there is something wrong/broken, please, don't hesitate to comment (or if it's
clear, feel free to open a JIRA and we'll work on it).
Thanks,
Lucas
3427
days inactive
3447
days old
19 comments
5 participants
participants (5)
-
Anton Hughes -
Gary Brown -
Jay Shaughnessy -
Lucas Ponce -
Thomas Segismont