...
The recursively
method applies input records to its op
argument. The results are then both emitted as a result of recursively
and also fed back in to the op
KStream
.The op
KStream
Restrictions:
- op cannot be
UnaryOperator.identity
, or an equivalent function that simply returns its argument unmodified - this would produce an infinite recursive loop, since there's no opportunity refine the output to break out of the loop. op
MUST "terminate"; that is, it must have some condition which eventually prevents further recursion of a record. In our example here, the terminating condition is thejoin
, since the root node of our graph will have noparent
, so thejoin
will produce no output for the root node.- We can attempt to detect "definitely non-terminating" arguments by failing to detect operations that can cause the stream to terminate (e.g.
filter
,join
,flatMap
, etc.) in the process graph produced by the function. - We cannot guarantee that a function that includes terminating operations (
filter
,join
,flatMap
, etc.) actually terminates.
- We can attempt to detect "definitely non-terminating" arguments by failing to detect operations that can cause the stream to terminate (e.g.
Automatic Repartitioning
If the op
argument applies a key-changing operation (as it does in our example above), a repartition
topic may be automatically created. The optional Produced
argument can be provided to customize repartitioning behaviour. This argument is ignored if a repartition
topic is not necessary.
...
Repartitioning is required only when the op
KStream
requires repartitioning and the current KStream
does not require repartitioning. The reason for this is that if both streams require repartitioning, then the current stream is guaranteed to automatically repartition records before any operation that requires repartitioning. In the above example, recursively
would not include a repartition topic, because join
already includes one.
Restrictions:
...
.
Implementation
In KStreamImpl
, implementation is fairly simple:
...