[ 
https://issues.apache.org/jira/browse/BEAM-8537?focusedWorklogId=379612&page=com.atlassian.jira.plugin.system.issuetabpanels:worklog-tabpanel#worklog-379612
 ]

ASF GitHub Bot logged work on BEAM-8537:
----------------------------------------

                Author: ASF GitHub Bot
            Created on: 30/Jan/20 21:41
            Start Date: 30/Jan/20 21:41
    Worklog Time Spent: 10m 
      Work Description: robertwb commented on pull request #10375: [BEAM-8537] 
Provide WatermarkEstimator to track watermark
URL: https://github.com/apache/beam/pull/10375#discussion_r373202214
 
 

 ##########
 File path: sdks/python/apache_beam/runners/common.py
 ##########
 @@ -765,41 +978,63 @@ def _invoke_process_per_window(self,
       # ProcessSizedElementAndRestriction.
       self.threadsafe_restriction_tracker.check_done()
       deferred_status = self.threadsafe_restriction_tracker.deferred_status()
-      output_watermark = None
-      if self.watermark_estimator:
-        output_watermark = self.watermark_estimator.current_watermark()
       if deferred_status:
         deferred_restriction, deferred_watermark = deferred_status
         element = windowed_value.value
         size = self.signature.get_restriction_provider().restriction_size(
             element, deferred_restriction)
-        return ((
-            windowed_value.with_value(((element, deferred_restriction), size)),
-            output_watermark), deferred_watermark)
-    return None
+        if self.threadsafe_watermark_estimator:
+          output_watermark = (
+              self.threadsafe_watermark_estimator.current_watermark())
+          estimator_state = (
+              self.threadsafe_watermark_estimator.get_estimator_state())
+          return ((
+              windowed_value.with_value(
+                  ((element, (deferred_restriction, estimator_state)), size)),
+              output_watermark), deferred_watermark)
+        else:
+          return ((
+              windowed_value.with_value(
+                  ((element, deferred_restriction), size)), None),
+                  deferred_watermark)
+        return None
 
   def try_split(self, fraction):
     # type: (...) -> Optional[Tuple[SplitResultType, SplitResultType]]
-    if self.threadsafe_restriction_tracker and self.current_windowed_value:
-      # Temporary workaround for [BEAM-7473]: get current_watermark before
-      # split, in case watermark gets advanced before getting split results.
-      # In worst case, current_watermark is always stale, which is ok.
-      if self.watermark_estimator:
-        current_watermark = self.watermark_estimator.current_watermark()
-      else:
-        current_watermark = None
-      split = self.threadsafe_restriction_tracker.try_split(fraction)
+    restriction_tracker = self.threadsafe_restriction_tracker
+    current_windowed_value = self.current_windowed_value
+    if restriction_tracker and current_windowed_value:
+      current_watermark = None
+      # Make sure that the RestrictionTracker and WatermarkEstimator are locked
+      # together.
+      with self._synchronized_lock:
+        split = restriction_tracker.try_split(fraction)
+        if self.threadsafe_watermark_estimator:
+          current_watermark = (
+              
self.threadsafe_watermark_estimator.current_watermark_with_lock())
+          estimator_state = (self.threadsafe_watermark_estimator
+                             .get_estimator_state_with_lock())
       if split:
         primary, residual = split
         element = self.current_windowed_value.value
         restriction_provider = self.signature.get_restriction_provider()
         primary_size = restriction_provider.restriction_size(element, primary)
         residual_size = restriction_provider.restriction_size(element, 
residual)
-        return (
-            ((self.current_windowed_value.with_value((
-                (element, primary), primary_size)), None), None),
-            ((self.current_windowed_value.with_value((
-                (element, residual), residual_size)), current_watermark), 
None))
+        if self.threadsafe_watermark_estimator:
+          return (
+              ((self.current_windowed_value.with_value((
 
 Review comment:
   As mentioned in the typing PRs, these tuples of tuples of tuples are getting 
a bit unwieldy. Perhaps we could introduce a class (or at least named tuple)? 
Open question: should the primary the same type from the residual, or should it 
be a different type (given it can hold a watermark and delay time). 
 
----------------------------------------------------------------
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.
 
For queries about this service, please contact Infrastructure at:
[email protected]


Issue Time Tracking
-------------------

    Worklog Id:     (was: 379612)
    Time Spent: 7.5h  (was: 7h 20m)

> Provide WatermarkEstimatorProvider for different types of WatermarkEstimator
> ----------------------------------------------------------------------------
>
>                 Key: BEAM-8537
>                 URL: https://issues.apache.org/jira/browse/BEAM-8537
>             Project: Beam
>          Issue Type: Improvement
>          Components: sdk-py-core, sdk-py-harness
>            Reporter: Boyuan Zhang
>            Assignee: Boyuan Zhang
>            Priority: Major
>          Time Spent: 7.5h
>  Remaining Estimate: 0h
>
> This is a follow up for in-progress PR:  
> https://github.com/apache/beam/pull/9794.
> Current implementation in PR9794 provides a default implementation of 
> WatermarkEstimator. For further work, we want to let WatermarkEstimator to be 
> a pure Interface. We'll provide a WatermarkEstimatorProvider to be able to 
> create a custom WatermarkEstimator per windowed value. It should be similar 
> to how we track restriction for SDF: 
> WatermarkEstimator <---> RestrictionTracker 
> WatermarkEstimatorProvider <---> RestrictionTrackerProvider
> WatermarkEstimatorParam <---> RestrictionDoFnParam



--
This message was sent by Atlassian Jira
(v8.3.4#803005)

Reply via email to