fix: eliminate double S3 read and close file handles in repair_metadata

During repair_metadata, each wheel was read from S3 twice: once for
content metadata extraction and again for metadata artifact creation.
Now the temp file from the first read is reused for the second,
halving S3 reads and temp file I/O for wheels.

Also explicitly closes artifact.file after each iteration to release
S3 buffer memory immediately instead of waiting for GC.

JIRA: PULP-1573

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

Author: decko

pulp_python/app/tasks/repair.py

@@ -1,4 +1,5 @@
 import logging
+import os
 from collections import defaultdict
 from gettext import gettext as _
 from itertools import groupby
@@ -119,18 +120,35 @@ def repair_metadata(content: QuerySet[PythonPackageContent]) -> tuple[int, set[s
                 .first()
                 .artifact
             )
-            new_data = artifact_to_python_content_data(package.filename, main_artifact, domain)
+            # For wheels, keep the temp file so we can reuse it for metadata artifact
+            is_wheel = package.filename.endswith(".whl")
+            if is_wheel:
+                new_data, temp_path = artifact_to_python_content_data(
+                    package.filename, main_artifact, domain, keep_temp_file=True
+                )
+            else:
+                new_data = artifact_to_python_content_data(
+                    package.filename, main_artifact, domain
+                )
+                temp_path = None
+
             total_metadata_repaired += update_metadata_artifact_if_needed(
                 package,
                 new_data.get("metadata_sha256"),
                 main_artifact,
                 metadata_batch,
                 pkgs_metadata_not_repaired,
+                temp_path=temp_path,
             )
             total_repaired += update_package_if_needed(
                 package, new_data, batch, set_of_update_fields
             )
 
+            # Clean up temp file and close artifact file handle
+            if temp_path and os.path.exists(temp_path):
+                os.unlink(temp_path)
+            main_artifact.file.close()
+
         # For on-demand content, we expect that:
         # 1. PythonPackageContent always has correct name and version
         # 2. RemoteArtifact always has correct sha256
@@ -240,6 +258,7 @@ def update_metadata_artifact_if_needed(
     main_artifact: Artifact,
     metadata_batch: list[tuple],
     pkgs_metadata_not_repaired: set[str],
+    temp_path: str | None = None,
 ) -> int:
     """
     Repairs metadata artifacts for wheel packages by creating missing metadata artifacts
@@ -252,6 +271,7 @@ def update_metadata_artifact_if_needed(
         main_artifact: The main package artifact used to generate metadata.
         metadata_batch: List of tuples for batch processing (updated in-place).
         pkgs_metadata_not_repaired: Set of package PKs that failed repair (updated in-place).
+        temp_path: Path to already-extracted temp wheel file, avoids re-reading from S3.
 
     Returns:
         Number of repaired metadata artifacts (only when batch is flushed at BULK_SIZE).
@@ -266,13 +286,13 @@ def update_metadata_artifact_if_needed(
 
     # Create missing
     if not cas:
-        metadata_batch.append((package, main_artifact))
+        metadata_batch.append((package, main_artifact, temp_path))
     # Fix existing
     elif new_metadata_sha256 != original_metadata_sha256:
         ca = cas.first()
         metadata_artifact = ca.artifact
         if metadata_artifact is None or (metadata_artifact.sha256 != new_metadata_sha256):
-            metadata_batch.append((package, main_artifact))
+            metadata_batch.append((package, main_artifact, temp_path))
 
     if len(metadata_batch) == BULK_SIZE:
         not_repaired = _process_metadata_batch(metadata_batch)
@@ -289,16 +309,21 @@ def _process_metadata_batch(metadata_batch: list[tuple]) -> set[str]:
     and their corresponding ContentArtifacts.
 
     Args:
-        metadata_batch: List of (package, main_artifact) tuples.
+        metadata_batch: List of (package, main_artifact, temp_path) tuples.
 
     Returns:
         Set of package PKs for which metadata artifacts could not be created.
     """
     not_repaired = set()
     content_artifacts = []
 
-    for package, main_artifact in metadata_batch:
-        metadata_artifact = artifact_to_metadata_artifact(package.filename, main_artifact)
+    for package, main_artifact, temp_path in metadata_batch:
+        if temp_path and os.path.exists(temp_path):
+            metadata_artifact = artifact_to_metadata_artifact_from_path(
+                package.filename, temp_path
+            )
+        else:
+            metadata_artifact = artifact_to_metadata_artifact(package.filename, main_artifact)
         if metadata_artifact:
             ca = ContentArtifact(
                 artifact=metadata_artifact,