Hi Flavio,
we have rather modest goals - so currently we don't plan to handle batch
(although in theory it could be done). We also don't even think about
visualizing and editing existing programs - in generality this could be
pretty difficult.
We only let users define pretty simple diagrams (in particular they have
to be trees, not DAGs) to be able to handle them properly (although we
do have mechanism to plug-in custom Flink code - it just have to be
coded separately as part of the model)
thanks,
maciek
On 04/09/2017 15:43, Flavio Pompermaier wrote:
Thanks for sharing this nice tool maciek. Does it handle both batch
and Streaming? Is it able to visualize also an existing Flink program?
Best,
Flavio
On Mon, Sep 4, 2017 at 3:03 PM, Maciek Próchniak <m...@touk.pl
<mailto:m...@touk.pl>> wrote:
Hello,
we would like to announce availability of TouK Nussknacker - tool
for creating Flink processes with GUI.
It's our attempt to bring Flink a bit closer to analysts and
business people - by letting them design processes with MS
Visio-like tool ;)
Of course not every Flink process can be created by drawing
diagrams, but we wanted to make it possible for analyst to design
process that does filtering, enrichment or simple aggregations.
Our idea is pretty simple:
- first, data model and additional services have to be developed
and packaged into jar (this is the jar that is used by Flink jobs)
- this is the part where coding skills are required
- users can create new processes by drawing diagrams - the nodes
in toolbox are based on the data model. This includes e.g.
creating filters with SPEL (Spring Expression Language) expressions.
- the process is deployed to Flink cluster as jar with model and
diagram serialized as JSON
We make it easier to work with our diagrams by integration with
Grafana (each process registers quite detailed metrics about it's
behaviour), letting users test their processes via GUI with Flink
mini-cluster and various other goodies (subprocesses, basic code
completion, export to PDF etc.)
So far we have one quite large (>30 processes, >100k/s total
throughput in all processes) production deployment and a few more
in progress. Our main use case so far is RTM (Real Time Marketing)
and fraud detection in telco, banking and similar industries.
We think that our tool can be of interest mainly for more
traditional enterprises, that do not have large development teams,
but have competent analysts in their BI departments.
Nussknacker is released under ASL 2.0 (however we use React
heavily so beware of patents clause...)
Of course, we'd like to hear your opinions.
The code is here: https://github.com/touk/nussknacker
<https://github.com/touk/nussknacker>
The quickstart is here:
https://touk.github.io/nussknacker/Quickstart.html
<https://touk.github.io/nussknacker/Quickstart.html>
And our blog post is here:
https://touk.pl/blog/2017/09/04/touk-nussknacker-using-apache-flink-made-easier-for-analysts-and-business/
<https://touk.pl/blog/2017/09/04/touk-nussknacker-using-apache-flink-made-easier-for-analysts-and-business/>
I'll also be talking about Nussknacker next week at Flink Forward
-
https://berlin.flink-forward.org/kb_sessions/touk-nussknacker-creating-flink-jobs-with-gui/
<https://berlin.flink-forward.org/kb_sessions/touk-nussknacker-creating-flink-jobs-with-gui/>
- hope to see you there :)
thanks,
maciek próchniak
TouK
