fix: make sandbox cleaning parameters configurable via settings

Move hardcoded batch_size, delete_chunk_size, and semaphore concurrency
values into SandboxStoreSettings so they can be tuned per deployment.

Author: chrisburr

diracx-core/src/diracx/core/settings.py

@@ -276,6 +276,24 @@ class SandboxStoreSettings(ServiceSettingsBase):
     deletion). Default: 50.
     """
 
+    clean_batch_size: int = 50_000
+    """Number of sandbox candidates to select per batch during cleaning.
+
+    Each batch runs SELECT → S3 delete → DB delete sequentially. Default: 50000.
+    """
+
+    clean_delete_chunk_size: int = 1000
+    """Number of sandbox DB rows to delete per chunk during cleaning.
+
+    Smaller chunks mean shorter transactions and less lock contention. Default: 1000.
+    """
+
+    clean_max_concurrent_db_deletes: int = 10
+    """Maximum number of concurrent DB delete chunks during cleaning.
+
+    Controls parallelism of database DELETE operations. Default: 10.
+    """
+
     _client: S3Client = PrivateAttr()
 
     @contextlib.asynccontextmanager

diracx-logic/src/diracx/logic/jobs/sandboxes.py

@@ -201,8 +201,6 @@ async def insert_sandbox(
 async def clean_sandboxes(
     sandbox_metadata_db: SandboxMetadataDB,
     settings: SandboxStoreSettings,
-    *,
-    batch_size: int = 50000,
 ) -> int:
     """Delete sandboxes that are not assigned to any job.
 
@@ -214,12 +212,12 @@ async def clean_sandboxes(
     Args:
         sandbox_metadata_db: Database connection (not yet entered).
         settings: Sandbox store settings with S3 client.
-        batch_size: Number of sandboxes to process per batch.
 
     Returns:
         Total number of sandboxes deleted.
 
     """
+    batch_size = settings.clean_batch_size
     total_deleted = 0
     cursor = 0
     batch_num = 0
@@ -294,8 +292,8 @@ async def clean_sandboxes(
         # transaction to avoid locking millions of rows in a single DELETE).
         # Up to 10 chunks run concurrently via a semaphore.
         t0 = time.monotonic()
-        delete_chunk_size = 1000
-        sem = asyncio.Semaphore(10)
+        delete_chunk_size = settings.clean_delete_chunk_size
+        sem = asyncio.Semaphore(settings.clean_max_concurrent_db_deletes)
 
         async def _delete_chunk(chunk: list[int], _sem: asyncio.Semaphore = sem) -> int:
             async with _sem, sandbox_metadata_db:

docs/admin/reference/env-variables.md

@@ -157,6 +157,30 @@ Maximum number of connections in the S3 client connection pool.
 Higher values allow more parallel S3 requests (e.g. during bulk sandbox
 deletion). Default: 50.
 
+### `DIRACX_SANDBOX_STORE_CLEAN_BATCH_SIZE`
+
+*Optional*, default value: `50000`
+
+Number of sandbox candidates to select per batch during cleaning.
+
+Each batch runs SELECT → S3 delete → DB delete sequentially. Default: 50000.
+
+### `DIRACX_SANDBOX_STORE_CLEAN_DELETE_CHUNK_SIZE`
+
+*Optional*, default value: `1000`
+
+Number of sandbox DB rows to delete per chunk during cleaning.
+
+Smaller chunks mean shorter transactions and less lock contention. Default: 1000.
+
+### `DIRACX_SANDBOX_STORE_CLEAN_MAX_CONCURRENT_DB_DELETES`
+
+*Optional*, default value: `10`
+
+Maximum number of concurrent DB delete chunks during cleaning.
+
+Controls parallelism of database DELETE operations. Default: 10.
+
 ## OTELSettings
 
 Settings for the Open Telemetry Configuration.