[GCP] Optimize BigQuery TriggerCopyJobs performance for WRITE_APPEND

sdks/python/apache_beam/io/gcp/bigquery_file_loads.py

@@ -491,6 +491,7 @@ class TriggerCopyJobs(beam.DoFn):
   """
 
   TRIGGER_DELETE_TEMP_TABLES = 'TriggerDeleteTempTables'
+  MAX_SOURCES_PER_COPY_JOB = 1200
 
   def __init__(
       self,
@@ -528,96 +529,98 @@ def process(
       self, element_list, job_name_prefix=None, unused_schema_mod_jobs=None):
     if isinstance(element_list, tuple):
       # Allow this for streaming update compatibility while fixing BEAM-24535.
-      self.process_one(element_list, job_name_prefix)
-    else:
-      for element in element_list:
-        self.process_one(element, job_name_prefix)
+      element_list = [element_list]
 
-  def process_one(self, element, job_name_prefix):
-    destination, job_reference = element
+    if not element_list:
+      return
+    if not self.bq_io_metadata:
+      self.bq_io_metadata = create_bigquery_io_metadata(self._step_name)
 
-    copy_to_reference = bigquery_tools.parse_table_reference(destination)
+    first_destination = element_list[0][0]
+    copy_to_reference = bigquery_tools.parse_table_reference(first_destination)
     if copy_to_reference.projectId is None:
       copy_to_reference.projectId = vp.RuntimeValueProvider.get_value(
           'project', str, '') or self.project
 
-    copy_from_reference = bigquery_tools.parse_table_reference(destination)
-    copy_from_reference.tableId = job_reference.jobId
-    if copy_from_reference.projectId is None:
-      copy_from_reference.projectId = vp.RuntimeValueProvider.get_value(
-          'project', str, '') or self.project
-
-    _LOGGER.info(
-        "Triggering copy job from %s to %s",
-        copy_from_reference,
-        copy_to_reference)
-
-    wait_for_job, write_disposition = (
-      self._determine_write_disposition(copy_to_reference))
-
-    if not self.bq_io_metadata:
-      self.bq_io_metadata = create_bigquery_io_metadata(self._step_name)
-
-    project_id = (
-        copy_to_reference.projectId
-        if self.load_job_project_id is None else self.load_job_project_id)
-    copy_job_name = '%s_%s' % (
-        job_name_prefix,
-        _bq_uuid(
-            '%s:%s.%s' % (
-                copy_from_reference.projectId,
-                copy_from_reference.datasetId,
-                copy_from_reference.tableId)))
-    job_reference = self.bq_wrapper._insert_copy_job(
-        project_id,
-        copy_job_name,
-        copy_from_reference,
-        copy_to_reference,
-        create_disposition=self.create_disposition,
-        write_disposition=write_disposition,
-        job_labels=self.bq_io_metadata.add_additional_bq_job_labels())
-
-    if wait_for_job:
-      self.bq_wrapper.wait_for_bq_job(job_reference, sleep_duration_sec=10)
-    self.pending_jobs.append(
-        GlobalWindows.windowed_value((destination, job_reference)))
+    copy_from_references = []
+    for destination, job_reference in element_list:
+      copy_from_reference = bigquery_tools.parse_table_reference(destination)
+      copy_from_reference.tableId = job_reference.jobId
+      if copy_from_reference.projectId is None:
+        copy_from_reference.projectId = vp.RuntimeValueProvider.get_value(
+            'project', str, '') or self.project
+      copy_from_references.append(copy_from_reference)
 
-  def _determine_write_disposition(self, copy_to_reference) -> tuple[bool, str]:
-    """
-    Determines the write disposition for a BigQuery copy job,
-     based on destination.
-
-    When the write_disposition for a job is WRITE_TRUNCATE, multiple copy jobs
-    to the same destination can interfere with each other, truncate data, and
-    write to the BigQuery table repeatedly. To prevent this, the first copy job
-    runs with the user's specified write_disposition, but subsequent jobs must
-    always use WRITE_APPEND. This ensures that subsequent copy jobs do not
-    clear out data appended by previous jobs.
-
-    Args:
-        copy_to_reference: The reference to the destination table.
-
-    Returns:
-        A tuple containing a boolean indicating whether to wait for the job to
-        complete and the write disposition to use for the job.
-    """
     full_table_ref = '%s:%s.%s' % (
         copy_to_reference.projectId,
         copy_to_reference.datasetId,
         copy_to_reference.tableId)
-    if full_table_ref not in self._observed_tables:
-      write_disposition = self.write_disposition
-      wait_for_job = True
+    is_first_time = full_table_ref not in self._observed_tables
+    if is_first_time:
       self._observed_tables.add(full_table_ref)
-      Lineage.sinks().add(
-          'bigquery',
-          copy_to_reference.projectId,
-          copy_to_reference.datasetId,
-          copy_to_reference.tableId)
-    else:
-      wait_for_job = False
-      write_disposition = 'WRITE_APPEND'
-    return wait_for_job, write_disposition
+      if self.bq_io_metadata:
+        Lineage.sinks().add(
+            'bigquery',
+            copy_to_reference.projectId,
+            copy_to_reference.datasetId,
+            copy_to_reference.tableId)
+
+    # Split into chunks of MAX_SOURCES_PER_COPY_JOB
+    chunks = [
+        copy_from_references[i:i + self.MAX_SOURCES_PER_COPY_JOB]
+        for i in range(
+            0, len(copy_from_references), self.MAX_SOURCES_PER_COPY_JOB)
+    ]
+
+    copy_job_name_base = '%s_%s' % (
+        job_name_prefix,
+        _bq_uuid(
+            '%s:%s.%s' % (
+                copy_to_reference.projectId,
+                copy_to_reference.datasetId,
+                copy_to_reference.tableId)))
+
+    project_id = (
+        copy_to_reference.projectId
+        if self.load_job_project_id is None else self.load_job_project_id)
+
+    for i, chunk in enumerate(chunks):
+      if i == 0 and is_first_time:
+        write_disposition = self.write_disposition
+        # Wait inline only if we have multiple chunks and write disposition is WRITE_TRUNCATE or WRITE_EMPTY.
+        # This ensures the first chunk initializes the table, and subsequent chunks (WRITE_APPEND) append to it.
+        wait_for_job = (
+            self.write_disposition in ('WRITE_TRUNCATE', 'WRITE_EMPTY') and
+            len(chunks) > 1)
+      else:
+        write_disposition = 'WRITE_APPEND'
+        wait_for_job = False
+
+      chunk_job_name = copy_job_name_base
+      if len(chunks) > 1:
+        chunk_job_name = f"{copy_job_name_base}_{i}"
+
+      _LOGGER.info(
+          "Triggering copy job %s from %s to %s (write_disposition: %s)",
+          chunk_job_name, [str(r) for r in chunk],
+          copy_to_reference,
+          write_disposition)
+
+      job_reference = self.bq_wrapper._insert_copy_job(
+          project_id,
+          chunk_job_name,
+          chunk,
+          copy_to_reference,
+          create_disposition=self.create_disposition,
+          write_disposition=write_disposition,
+          job_labels=self.bq_io_metadata.add_additional_bq_job_labels()
+          if self.bq_io_metadata else None)
+
+      if wait_for_job:
+        self.bq_wrapper.wait_for_bq_job(job_reference, sleep_duration_sec=10)
+
+      self.pending_jobs.append(
+          GlobalWindows.windowed_value((first_destination, job_reference)))
 
   def finish_bundle(self):
     for windowed_value in self.pending_jobs:
@@ -855,7 +858,8 @@ def process(self, element):
       if latest_partition.can_accept(file_size):
         latest_partition.add(file_path, file_size)
       else:
-        partitions.append(latest_partition.files)
+        if latest_partition.files:
+          partitions.append(latest_partition.files)
         latest_partition = PartitionFiles.Partition(
             self.max_partition_size, self.max_files_per_partition)
         latest_partition.add(file_path, file_size)
@@ -1181,12 +1185,13 @@ def _load_data(
       # the truncation happens only once. See
       # https://github.com/apache/beam/issues/24535.
       finished_temp_tables_load_job_ids_list_pc = (
-          finished_temp_tables_load_job_ids_pc | beam.MapTuple(
+          finished_temp_tables_load_job_ids_pc
+          | beam.MapTuple(
               lambda destination, job_reference: (
-                  bigquery_tools.parse_table_reference(destination).tableId,
+                  bigquery_tools.get_hashable_destination(destination),
                   (destination, job_reference)))
           | beam.GroupByKey()
-          | beam.MapTuple(lambda tableId, batch: list(batch)))
+          | beam.MapTuple(lambda dest, batch: list(batch)))
     else:
       # Loads can happen in parallel.
       finished_temp_tables_load_job_ids_list_pc = (