I think we could start with a list of what Apache Eagle needed from
Topology lifecycle management and see if we do cross-post to both dev@ list
of Eagle and Storm to see if enough interest from Storm to work on it or
accept contributions from Eagle.

- Henry

On Fri, Jan 15, 2016 at 4:06 PM, Zhang, Edward (GDI Hadoop) <
[email protected]> wrote:

> I had a short discussion with Henry about this. We probably need discuss a
> more graceful way to tackle the problem of whether eagle does this or
> storm does this.
> Today we know that Storm also provides topology view/statistics features
> but does not have topology lifecycle management UI.
> But can we ask Storm team to build this topology lifecycle UI?
> Right now, can we make our implementation more pluggable and extensible so
> later on if Storm has this feature, we can use that directly (at least
> from API lever)
>
> Do you want to communicate with storm community about this or probably
> Eagle committer can build this feature and contribute back to storm?
>
> What do you think?
>
> Thanks
> Edward
>
>
>
> On 1/11/16, 14:45, "Edward Zhang" <[email protected]> wrote:
>
> >Thanks for the valuable comments, you are right in the high-level
> >observations :-)
> >
> >(I participated some offline discussion on this proposal) I think this
> >proposal is based on the requirements that Eagle not only monitors
> >security
> >events from hadoop but also monitors security events from other data
> >source, for example cassandra and even more requirements are from hadoop
> >native metric monitoring. In last 2 months, when we want to onboard new
> >diverse datasources(for example mongo db metrics, hadoop native metrics
> >etc.), we find it is impossible for user(mostly operations team) to
> >understand storm topology or write code for very simple metrics
> >ingestion/alert rules. For monitoring perspective, user usually wants
> >metric/log onboarding is as simple as turn-key operation.
> >
> >It is possible for Eagle developers to develop applications for each data
> >source, but it might be better for Eagle to support logs/metrics with some
> >general schema. User just needs some configurations to get new data source
> >flow into Eagle and create policy on-the-fly against the streaming data.
> >Maybe I am wrong, but Eagle looks is an application to Storm framework,
> >but
> >Eagle would be a framework to monitoring applications e.g. security or
> >other data activity.
> >
> >The point of Eagle to separate application and framework is very correct.
> >In Eagle source code, the application code and framework code are
> >separated
> >from the beginning. For those features like policy restore after machine
> >fails, DSL, aggregation etc, I think Eagle team should look for
> >contributing back to Storm if people agree that is what stream framework
> >should have.
> >
> >We also found that Eagle monitoring uses streaming framework but streaming
> >framework is not customized for monitoring. The gap between monitoring
> >platform and streaming framework has to be filled to make sure monitoring
> >is reliable. For example compared to real-time streaming analytics,
> >monitoring does not want to have any false alert or missing alert
> >especially for security event, which requires more processing semantics
> >than popular streaming framework provides. We will actively explore help
> >from streaming projects.
> >
> >Thanks
> >
> >Edward
> >
> >On Mon, Jan 11, 2016 at 10:55 AM, Julian Hyde <[email protected]> wrote:
> >
> >> Can I make a high-level observation? (And, although I¹m a mentor of this
> >> project, I¹m not speaking as a mentor, just someone who has built
> >>various
> >> database and streaming systems over the years.)
> >>
> >> I¹ve noticed that Eagle is taking on several problems that ‹
> >> architecturally speaking ‹ should be part of the underlying streaming
> >> system. This topology manager and also, a DSL declarative streaming
> >> queries, and making sure that streaming queries continue where they left
> >> off, even in the presence of failures of individual stream-processing
> >>nodes.
> >>
> >> These are very hard distributed systems problems, and they are
> >>horizontal
> >> problems that have nothing to do with Eagle¹s problem domain
> >>(security). It
> >> would be analogous to an application that is selling concert tickets
> >> deciding to develop HBase as part of the application.
> >>
> >> If the Eagle community wants to solve these problems, that¹s awesome,
> >>and
> >> you should go for it. Apache projects are great at pulling in people
> >>with
> >> diverse skills and when they gather momentum they can build some amazing
> >> technology. But I think Eagle should consider putting more architectural
> >> separation between your stream management technology and your
> >>application.
> >> You could do that by building separate modules (and testing them
> >> independently). Or you could contribute the functionality you need to
> >>the
> >> underlying system (e.g. Storm/Nimbus). Your project will run much more
> >> smoothly if you call out the hard problems you are trying to solve.
> >>
> >> And, as a side benefit, projects that have nothing to do with Hadoop
> >> security will be able to use (and test, and bug-fix) the technology you
> >>are
> >> developing.
> >>
> >> Julian
> >>
> >>
> >> > On Jan 11, 2016, at 8:57 AM, Hao Chen <[email protected]> wrote:
> >> >
> >> > Currently eagle is requiring user to manually manage topologies
> >> completely
> >> > independent of eagle components,  which is not very smooth for the
> >>user
> >> and
> >> > management experience end-to-end from on-boarding datasource, starting
> >> > topologies, defining policy and also monitoring policy and execution
> >> > status, so how do you think we manage everything in single place and
> >> > dynamically manage topology lifecycle like
> >> > starting/stopping/status/monitoring as well policy
> >> > creation/modification/monitoring all in eagle ui only? So that user
> >>don't
> >> > need to touch storm anymore except specifying where the nimbus is when
> >> > setting up eagle.
> >> >
> >> >
> >> > --
> >> >
> >> > Hao
> >>
> >>
>
>

Reply via email to