Merge branch 'main' into fix/local-only-format

Author: j-hartshorn

sqlmesh/cli/main.py

@@ -635,16 +635,22 @@ def invalidate(ctx: click.Context, environment: str, **kwargs: t.Any) -> None:
     is_flag=True,
     help="Cleanup snapshots that are not referenced in any environment, regardless of when they're set to expire",
 )
+@click.option(
+    "--force-delete",
+    is_flag=True,
+    help="Delete expired environment and snapshot state records even when the physical table or view drops fail. "
+    "Any objects that could not be dropped become orphaned and must be removed manually.",
+)
 @click.pass_context
 @error_handler
 @cli_analytics
-def janitor(ctx: click.Context, ignore_ttl: bool, **kwargs: t.Any) -> None:
+def janitor(ctx: click.Context, ignore_ttl: bool, force_delete: bool, **kwargs: t.Any) -> None:
     """
     Run the janitor process on-demand.
 
     The janitor cleans up old environments and expired snapshots.
     """
-    ctx.obj.run_janitor(ignore_ttl, **kwargs)
+    ctx.obj.run_janitor(ignore_ttl, force_delete=force_delete, **kwargs)
 
 
 @cli.command("destroy")

sqlmesh/core/context.py

@@ -890,12 +890,12 @@ def _has_environment_changed() -> bool:
         return completion_status
 
     @python_api_analytics
-    def run_janitor(self, ignore_ttl: bool) -> bool:
+    def run_janitor(self, ignore_ttl: bool, force_delete: bool = False) -> bool:
         success = False
 
         if self.console.start_cleanup(ignore_ttl):
             try:
-                self._run_janitor(ignore_ttl)
+                self._run_janitor(ignore_ttl, force_delete=force_delete)
                 success = True
             finally:
                 self.console.stop_cleanup(success=success)
@@ -2899,24 +2899,43 @@ def _destroy(self) -> bool:
 
         return True
 
-    def _run_janitor(self, ignore_ttl: bool = False) -> None:
+    def _run_janitor(self, ignore_ttl: bool = False, force_delete: bool = False) -> None:
         current_ts = now_timestamp()
+        failures: t.List[str] = []
 
         # Clean up expired environments by removing their views and schemas
-        self._cleanup_environments(current_ts=current_ts)
+        failures.extend(
+            self._cleanup_environments(current_ts=current_ts, force_delete=force_delete)
+        )
 
-        delete_expired_snapshots(
-            self.state_sync,
-            self.snapshot_evaluator,
-            current_ts=current_ts,
-            ignore_ttl=ignore_ttl,
-            console=self.console,
-            batch_size=self.config.janitor.expired_snapshots_batch_size,
+        failures.extend(
+            delete_expired_snapshots(
+                self.state_sync,
+                self.snapshot_evaluator,
+                current_ts=current_ts,
+                ignore_ttl=ignore_ttl,
+                force_delete=force_delete,
+                console=self.console,
+                batch_size=self.config.janitor.expired_snapshots_batch_size,
+            )
         )
         self.state_sync.compact_intervals()
 
-    def _cleanup_environments(self, current_ts: t.Optional[int] = None) -> None:
+        if failures:
+            failure_string = "\n  - ".join(failures)
+            summary = f"Janitor completed with failures:\n  {failure_string}"
+            if force_delete:
+                summary += "\nState records have been deleted, but the underlying objects may still exist in the database.\nPlease investigate and clean up manually the above if necessary."
+            if self.config.janitor.warn_on_delete_failure:
+                self.console.log_warning(summary)
+            else:
+                raise SQLMeshError(summary)
+
+    def _cleanup_environments(
+        self, current_ts: t.Optional[int] = None, force_delete: bool = False
+    ) -> t.List[str]:
         current_ts = current_ts or now_timestamp()
+        failures: t.List[str] = []
 
         expired_environments_summaries = self.state_sync.get_expired_environments(
             current_ts=current_ts
@@ -2926,15 +2945,20 @@ def _cleanup_environments(self, current_ts: t.Optional[int] = None) -> None:
             expired_env = self.state_reader.get_environment(expired_env_summary.name)
 
             if expired_env:
-                cleanup_expired_views(
-                    default_adapter=self.engine_adapter,
-                    engine_adapters=self.engine_adapters,
-                    environments=[expired_env],
-                    warn_on_delete_failure=self.config.janitor.warn_on_delete_failure,
-                    console=self.console,
+                failures.extend(
+                    cleanup_expired_views(
+                        default_adapter=self.engine_adapter,
+                        engine_adapters=self.engine_adapters,
+                        environments=[expired_env],
+                        console=self.console,
+                    )
                 )
 
-        self.state_sync.delete_expired_environments(current_ts=current_ts)
+        # we want to retry on the next janitor pass if drops failed, unless
+        # force_delete is set in which case we purge state records regardless
+        if not failures or force_delete:
+            self.state_sync.delete_expired_environments(current_ts=current_ts)
+        return failures
 
     def _try_connection(self, connection_name: str, validator: t.Callable[[], None]) -> None:
         connection_name = connection_name.capitalize()

sqlmesh/core/janitor.py

@@ -16,16 +16,16 @@
     RowBoundary,
     ExpiredBatchRange,
 )
-from sqlmesh.utils.errors import SQLMeshError
 
 
 def cleanup_expired_views(
     default_adapter: EngineAdapter,
     engine_adapters: t.Dict[str, EngineAdapter],
     environments: t.List[Environment],
-    warn_on_delete_failure: bool = False,
     console: t.Optional[Console] = None,
-) -> None:
+) -> t.List[str]:
+    failures: t.List[str] = []
+
     expired_schema_or_catalog_environments = [
         environment
         for environment in environments
@@ -85,10 +85,8 @@ def get_adapter(gateway_managed: bool, gateway: t.Optional[str] = None) -> Engin
                 console.update_cleanup_progress(expired_view)
         except Exception as e:
             message = f"Failed to drop the expired environment view '{expired_view}': {e}"
-            if warn_on_delete_failure:
-                logger.warning(message)
-            else:
-                raise SQLMeshError(message) from e
+            logger.warning(message)
+            failures.append(message)
 
     # Drop the schemas for the expired environments
     for engine_adapter, schema in schemas_to_drop:
@@ -102,10 +100,8 @@ def get_adapter(gateway_managed: bool, gateway: t.Optional[str] = None) -> Engin
                 console.update_cleanup_progress(schema.sql(dialect=engine_adapter.dialect))
         except Exception as e:
             message = f"Failed to drop the expired environment schema '{schema}': {e}"
-            if warn_on_delete_failure:
-                logger.warning(message)
-            else:
-                raise SQLMeshError(message) from e
+            logger.warning(message)
+            failures.append(message)
 
     # Drop any catalogs that were associated with a snapshot where the engine adapter supports dropping catalogs
     # catalogs_to_drop is only populated when environment_suffix_target is set to 'catalog'
@@ -117,10 +113,10 @@ def get_adapter(gateway_managed: bool, gateway: t.Optional[str] = None) -> Engin
                     console.update_cleanup_progress(catalog)
             except Exception as e:
                 message = f"Failed to drop the expired environment catalog '{catalog}': {e}"
-                if warn_on_delete_failure:
-                    logger.warning(message)
-                else:
-                    raise SQLMeshError(message) from e
+                logger.warning(message)
+                failures.append(message)
+
+    return failures
 
 
 def delete_expired_snapshots(
@@ -129,9 +125,10 @@ def delete_expired_snapshots(
     *,
     current_ts: int,
     ignore_ttl: bool = False,
+    force_delete: bool = False,
     batch_size: t.Optional[int] = None,
     console: t.Optional[Console] = None,
-) -> None:
+) -> t.List[str]:
     """Delete all expired snapshots in batches.
 
     This helper function encapsulates the logic for deleting expired snapshots in batches,
@@ -142,12 +139,14 @@ def delete_expired_snapshots(
         snapshot_evaluator: SnapshotEvaluator instance to clean up tables associated with snapshots.
         current_ts: Timestamp used to evaluate expiration.
         ignore_ttl: If True, include snapshots regardless of TTL (only checks if unreferenced).
+        force_delete: If True, delete snapshot state records even when physical table cleanup fails.
         batch_size: Maximum number of snapshots to fetch per batch.
         console: Optional console for reporting progress.
 
     Returns:
-        The total number of deleted expired snapshots.
+        List of failure messages so callers can surface them at the end of the janitor run.
     """
+    failures: t.List[str] = []
     num_expired_snapshots = 0
     for batch in iter_expired_snapshot_batches(
         state_reader=state_sync,
@@ -165,17 +164,32 @@ def delete_expired_snapshots(
             len(batch.expired_snapshot_ids),
             end_info,
         )
-        snapshot_evaluator.cleanup(
-            target_snapshots=batch.cleanup_tasks,
-            on_complete=console.update_cleanup_progress if console else None,
-        )
-        state_sync.delete_expired_snapshots(
-            batch_range=ExpiredBatchRange(
-                start=RowBoundary.lowest_boundary(),
-                end=batch.batch_range.end,
-            ),
-            ignore_ttl=ignore_ttl,
-        )
-        logger.info("Cleaned up expired snapshots batch")
-        num_expired_snapshots += len(batch.expired_snapshot_ids)
+        cleanup_succeeded = True
+        try:
+            snapshot_evaluator.cleanup(
+                target_snapshots=batch.cleanup_tasks,
+                on_complete=console.update_cleanup_progress if console else None,
+            )
+        except Exception as failed_drops:
+            message = f"Failed to clean up: {failed_drops}"
+            logger.warning(message)
+            failures.append(message)
+            cleanup_succeeded = False
+
+        if cleanup_succeeded or force_delete:
+            try:
+                state_sync.delete_expired_snapshots(
+                    batch_range=ExpiredBatchRange(
+                        start=RowBoundary.lowest_boundary(),
+                        end=batch.batch_range.end,
+                    ),
+                    ignore_ttl=ignore_ttl,
+                )
+                logger.info("Cleaned up expired snapshots batch")
+                num_expired_snapshots += len(batch.expired_snapshot_ids)
+            except Exception as e:
+                message = f"Failed to delete expired snapshot state records: {e}"
+                logger.warning(message)
+                failures.append(message)
     logger.info("Cleaned up %s expired snapshots", num_expired_snapshots)
+    return failures

sqlmesh/core/snapshot/evaluator.py

@@ -547,7 +547,7 @@ def cleanup(
             t.snapshot.snapshot_id: t.dev_table_only for t in filtered_targets
         }
         with self.concurrent_context():
-            concurrent_apply_to_snapshots(
+            errors, _ = concurrent_apply_to_snapshots(
                 [t.snapshot for t in filtered_targets],
                 lambda s: self._cleanup_snapshot(
                     s,
@@ -557,7 +557,11 @@ def cleanup(
                 ),
                 self.ddl_concurrent_tasks,
                 reverse_order=True,
+                raise_on_error=False,
             )
+        if errors:
+            errored_snapshots = "\n".join(f"  {e.node.name}: {e.__cause__}" for e in errors)
+            raise SQLMeshError(f"\n{errored_snapshots}")
 
     def audit(
         self,