Fix large exports failing

CHANGES/5375.bugfix

@@ -0,0 +1 @@
+Added Pulp side batching to fix large exports that were failing due to changes in psycopg.

pulpcore/app/importexport.py

@@ -9,6 +9,7 @@
 from django.db.models.query import QuerySet
 
 from pulpcore.app.apps import get_plugin_config
+from pulpcore.app.models.content import Artifact
 from pulpcore.app.models.progress import ProgressReport
 from pulpcore.app.models.repository import Repository
 from pulpcore.app.modelresource import (
@@ -50,25 +51,21 @@ def _write_export(the_tarfile, resource, dest_dir=None):
             temp_file.write("[")
 
             def process_batch(batch):
-                dataset = resource.export(batch)
+                model = resource.queryset.model
+                queryset = model.objects.filter(pk__in=batch)
+                dataset = resource.export(queryset)
                 # Strip "[" and "]" as we are writing the dataset in batch
                 temp_file.write(dataset.json.lstrip("[").rstrip("]"))
 
-            batch = []
-            needs_comma = False
-            for item in resource.queryset.iterator(chunk_size=EXPORT_BATCH_SIZE):
-                batch.append(item)
-                if needs_comma:
-                    # Write "," if not last loop
-                    temp_file.write(", ")
-                    needs_comma = False
-
-                if len(batch) >= EXPORT_BATCH_SIZE:
-                    process_batch(batch)
-                    batch.clear()
-                    needs_comma = True
+            first_loop = True
+            resource_pks = resource.queryset.values_list("pk", flat=True)
+            for offset in range(0, len(resource_pks), EXPORT_BATCH_SIZE):
+                batch = resource_pks[offset : offset + EXPORT_BATCH_SIZE]
 
-            if batch:
+                if not first_loop:
+                    temp_file.write(", ")
+                else:
+                    first_loop = False
                 process_batch(batch)
 
             temp_file.write("]")
@@ -102,39 +99,48 @@ def export_versions(export, version_info):
     export.tarfile.addfile(info, io.BytesIO(version_json))
 
 
-def export_artifacts(export, artifacts):
+def export_artifacts(export, artifact_pks):
     """
     Export a set of Artifacts, ArtifactResources, and RepositoryResources
 
     Args:
         export (django.db.models.PulpExport): export instance that's doing the export
-        artifacts (django.db.models.Artifacts): List of artifacts in all repos being exported
+        artifact_pks (django.db.models.Artifacts): List of artifact_pks in all repos being exported
 
     Raises:
         ValidationError: When path is not in the ALLOWED_EXPORT_PATHS setting
     """
-    data = dict(message="Exporting Artifacts", code="export.artifacts", total=len(artifacts))
+    data = dict(message="Exporting Artifacts", code="export.artifacts", total=len(artifact_pks))
     with ProgressReport(**data) as pb:
         pb.BATCH_INTERVAL = 5000
 
         if settings.DEFAULT_FILE_STORAGE != "pulpcore.app.models.storage.FileSystem":
             with tempfile.TemporaryDirectory(dir=".") as temp_dir:
-                for artifact in pb.iter(artifacts.only("file").iterator()):
-                    with tempfile.NamedTemporaryFile(dir=temp_dir) as temp_file:
-                        # TODO: this looks like a memory usage threat
-                        # TODO: it's also probably horrificaly slow, going one-by-one over the net
-                        # TODO: probably we could skip the temp file entirely and add
-                        #       artifact.file.read() directly to the tarfile with tarfile.addfile()
-                        temp_file.write(artifact.file.read())
-                        temp_file.flush()
-                        artifact.file.close()
-                        export.tarfile.add(temp_file.name, artifact.file.name)
+                for offset in range(0, len(artifact_pks), EXPORT_BATCH_SIZE):
+                    batch = artifact_pks[offset : offset + EXPORT_BATCH_SIZE]
+                    batch_qs = Artifact.objects.filter(pk__in=batch).only("file")
+
+                    for artifact in pb.iter(batch_qs.iterator()):
+                        with tempfile.NamedTemporaryFile(dir=temp_dir) as temp_file:
+                            # TODO: this looks like a memory usage threat
+                            # TODO: it's probably very slow, going one-by-one over the net
+                            # TODO: probably we could skip the temp file entirely and add
+                            #       artifact.file.read() directly to the tarfile with
+                            #       tarfile.addfile()
+                            temp_file.write(artifact.file.read())
+                            temp_file.flush()
+                            artifact.file.close()
+                            export.tarfile.add(temp_file.name, artifact.file.name)
         else:
-            for artifact in pb.iter(artifacts.only("file").iterator()):
-                export.tarfile.add(artifact.file.path, artifact.file.name)
+            for offset in range(0, len(artifact_pks), EXPORT_BATCH_SIZE):
+                batch = artifact_pks[offset : offset + EXPORT_BATCH_SIZE]
+                batch_qs = Artifact.objects.filter(pk__in=batch).only("file")
+
+                for artifact in pb.iter(batch_qs.iterator()):
+                    export.tarfile.add(artifact.file.path, artifact.file.name)
 
     resource = ArtifactResource()
-    resource.queryset = artifacts
+    resource.queryset = Artifact.objects.filter(pk__in=artifact_pks)
     _write_export(export.tarfile, resource)
 
     resource = RepositoryResource()

pulpcore/app/tasks/export.py

@@ -25,7 +25,7 @@
     RepositoryVersion,
     Task,
 )
-from pulpcore.app.models.content import Artifact, ContentArtifact
+from pulpcore.app.models.content import ContentArtifact
 from pulpcore.app.serializers import PulpExportSerializer
 
 from pulpcore.app.util import compute_file_hash, Crc32Hasher
@@ -509,7 +509,7 @@ def _do_export(pulp_exporter, tar, the_export):
 
     # Export the top-level entities (artifacts and repositories)
     # Note: we've already handled "what about incrementals" when building the 'artifacts' list
-    export_artifacts(the_export, Artifact.objects.filter(pk__in=artifact_pks))
+    export_artifacts(the_export, list(artifact_pks))
     del artifact_pks
 
     # Export the repository-version data, per-version