You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
* Move arg assignments to constructor
* Rename onChange params for legibility
* Reuse KStream builder for chainers (WIP for pipelines).
Pass KStream topology builder to chainers.
This is a step towarding allowing an entire pipeline to use the same builder,
and thereby be treated as a single KStream app (maybe with sub-topologies).
This change also cancels pipelines atomically, rather than cancelling steps
individually.
It is incomplete as joiners need the same behaviour.
* Minor formatting
* Reuse KStream builder for joiners
* Clear KStream state on shutdown
* Formatting
* Fix transformer tests for changed params with stream builder reuse
* Move MD5 hashing utils to object + use (hashed) pipeline name to enrich Kafka properties for pipeline
* Formatting + remove unused imports
* Move SeldonHeaders to dedicated file
* Refactor Alibi headers to drop & rename containing variable
* Formatting
* Refactor Kafka headers handling logic to subpackage
* Rename header transformers for intent
* Move pipeline topology & associated classes to separate file
* Refactor pipeline start/stop to methods
* Make pipeline metadata a private member
* Simplify pipeline cancellation/deletion & remove now-unused method
* Make pipeline a Kafka state listener instead of transformer implementations
This is because it is the pipeline itself which represents a KStream app,
and as only one state listener is permitted per KStream app, we cannot have
each transformer listening individually.
In any case, we are using this to determine when the pipeline is ready to
update the scheduler of this fact, thus it makes the most sense to check
for the pipeline as a whole rather than any subcomponent.
* Formatting of format strings
* Set pipeline as the state listener for its KStream app
* Remove redundant import
* Formatting + remove unused imports
* Remove start & stop methods from transformer interface & implementations
* Tidy imports
* Move PipelineTopology class to Kafka-related package
* Rename Transformer -> PipelineStep to better reflect usage and intent
Additionally, 'transformer' implies a 'transform' method, whereas this interface is much more a loose net around implementations.
* Remove unused test var
* Use engine in subscriber name & README instead of transformer
No description provided.
The text was updated successfully, but these errors were encountered: