fix(memory-monitor): use FailureType.transient_error for OOM fail-fast

OOM pressure near the cliff is a transient condition (workload/slice shape
dependent) rather than an internal CDK logic error, so a retry with fresh
process memory is meaningful. Flip failure_type from system_error to
transient_error so orchestrators can retry appropriately.

Co-Authored-By: patrick.nilan@airbyte.io <patrick.nilan@airbyte.io>

Author: devin-ai-integration[bot]

airbyte_cdk/utils/memory_monitor.py

@@ -126,7 +126,7 @@ class MemoryMonitor:
     (default 5000) to 100 messages to narrow the race window near OOM.
 
     **Fail-fast:** Raises ``AirbyteTracedException`` with
-    ``FailureType.system_error`` when *both*:
+    ``FailureType.transient_error`` when *both*:
 
     1. Cgroup usage >= 95% of the container limit (container is near OOM-kill)
     2. Anonymous memory >= 85% of *current cgroup usage* (most of the charged
@@ -296,7 +296,7 @@ def check_memory_usage(self) -> None:
                             f"Thresholds: cgroup >= {int(_CRITICAL_THRESHOLD * 100)}%, "
                             f"anon share of usage >= {int(_ANON_SHARE_OF_USAGE_THRESHOLD * 100)}%."
                         ),
-                        failure_type=FailureType.system_error,
+                        failure_type=FailureType.transient_error,
                     )
                 elif not self._critical_logged:
                     self._critical_logged = True

unit_tests/utils/test_memory_monitor.py

@@ -387,7 +387,7 @@ def test_raises_when_cgroup_critical_and_anon_share_of_usage_above_threshold() -
     ):
         with pytest.raises(AirbyteTracedException) as exc_info:
             monitor.check_memory_usage()
-    assert exc_info.value.failure_type == FailureType.system_error
+    assert exc_info.value.failure_type == FailureType.transient_error
     assert "critical threshold" in (exc_info.value.message or "")
     assert "96%" in (exc_info.value.message or "")
     assert "anon share of usage" in (exc_info.value.internal_message or "")