jrmccluskey commented on a change in pull request #15743:
URL: https://github.com/apache/beam/pull/15743#discussion_r732062482
##########
File path: sdks/go/pkg/beam/core/runtime/exec/window.go
##########
@@ -96,3 +96,23 @@ func (w *WindowInto) Down(ctx context.Context) error {
func (w *WindowInto) String() string {
return fmt.Sprintf("WindowInto[%v]. Out:%v", w.Fn, w.Out.ID())
}
+
+// WindowMapper defines an interface maps windows from a main input window
space
+// to windows from a side input window space. Used during side input
materialization.
+type WindowMapper interface {
+ MapWindow(w typex.Window) (typex.Window, error)
+}
+
+type windowMapper struct {
+ wfn *window.Fn
+}
+
+func (f *windowMapper) MapWindow(w typex.Window) (typex.Window, error) {
+ candidates := assignWindows(f.wfn, w.MaxTimestamp())
+ if len(candidates) == 0 {
+ return nil, fmt.Errorf("failed to map main input window to side
input window with WindowFn %v", f.wfn.String())
+ }
+ // Return latest candidate window in terms of event time (only relevant
for sliding windows)
+ // Sliding windows append the latest window first in assignWindows.
+ return candidates[0], nil
Review comment:
So the Python comment of "last candidate" also generates them backwards.
Got it. Explicitly calling out behavior in terms of event time is much clearer.
The current implementation of getting the first element in the slice matches
the comment and intent from what I understood at the time, but fixing it isn't
a big issue (apart from updating the tests to reflect that difference)
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: [email protected]
For queries about this service, please contact Infrastructure at:
[email protected]
