bvolpato commented on code in PR #24979:
URL: https://github.com/apache/beam/pull/24979#discussion_r1253182086
##########
sdks/python/apache_beam/io/gcp/bigquery.py:
##########
@@ -1515,11 +1526,41 @@ def process(self, element, *schema_side_inputs):
if not self.with_batched_input:
row_and_insert_id = element[1]
+ row_byte_size = get_deep_size(row_and_insert_id)
+
+ # send large rows that exceed BigQuery insert limits to DLQ
+ if row_byte_size >= self.max_insert_payload_size:
+ row_mb_size = row_byte_size / 1_000_000
+ max_mb_size = self.max_insert_payload_size / 1_000_000
+ error = (
+ f"Received row with size {row_mb_size}MB that exceeds "
+ f"the maximum insert payload size set ({max_mb_size}MB).")
+ return [
+ pvalue.TaggedOutput(
+ BigQueryWriteFn.FAILED_ROWS_WITH_ERRORS,
+ GlobalWindows.windowed_value(
+ (destination, row_and_insert_id[0], error))),
+ pvalue.TaggedOutput(
+ BigQueryWriteFn.FAILED_ROWS,
+ GlobalWindows.windowed_value(
+ (destination, row_and_insert_id[0])))
+ ]
+
+ buffer_byte_size_with_row = (
+ row_byte_size if not self._rows_buffer[destination] else
+ row_byte_size + get_deep_size(self._rows_buffer[destination]))
Review Comment:
Can't this be expensive?
We already had `get_deep_size` for every element before, perhaps we could
also accumulate it here instead of doing `get_deep_size` on the entire array
again?
##########
sdks/python/apache_beam/io/gcp/bigquery.py:
##########
@@ -1515,11 +1526,41 @@ def process(self, element, *schema_side_inputs):
if not self.with_batched_input:
row_and_insert_id = element[1]
+ row_byte_size = get_deep_size(row_and_insert_id)
+
+ # send large rows that exceed BigQuery insert limits to DLQ
+ if row_byte_size >= self.max_insert_payload_size:
+ row_mb_size = row_byte_size / 1_000_000
+ max_mb_size = self.max_insert_payload_size / 1_000_000
+ error = (
+ f"Received row with size {row_mb_size}MB that exceeds "
+ f"the maximum insert payload size set ({max_mb_size}MB).")
+ return [
+ pvalue.TaggedOutput(
+ BigQueryWriteFn.FAILED_ROWS_WITH_ERRORS,
+ GlobalWindows.windowed_value(
+ (destination, row_and_insert_id[0], error))),
+ pvalue.TaggedOutput(
+ BigQueryWriteFn.FAILED_ROWS,
+ GlobalWindows.windowed_value(
+ (destination, row_and_insert_id[0])))
+ ]
+
+ buffer_byte_size_with_row = (
+ row_byte_size if not self._rows_buffer[destination] else
+ row_byte_size + get_deep_size(self._rows_buffer[destination]))
Review Comment:
Can't this be _potentially_ really expensive?
We already had `get_deep_size` for every element before, perhaps we could
also accumulate it here instead of doing `get_deep_size` on the entire array
again?
--
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]
