[guisea]

Problem

When DirectAdmin pushes multiple zones within the same second, the save worker was processing them in separate 1/1 batches instead of one grouped batch.

Root cause: The batch drain loop (get_nowait()) only worked if backend processing was slow enough to act as an accidental debounce gate. With fast backends, item 1 would finish and hit Empty before items 2 and 3 from the same burst had landed in the queue — each item then opened its own batch.

Fix

Adds BATCH_DEBOUNCE_SECONDS = 0.5 to worker.py. After the blocking get() fires on the first item, the worker sleeps 500 ms so the rest of the burst can accumulate. The existing drain loop then processes all of them in a single batch.

Also fixes from app.utils import … → from directdnsonly.app.utils import … (short-form import was unresolvable without a runtime sys.path hack, which blocked writing worker tests).

Tests (TDD)

New tests/test_worker.py:

• test_burst_items_land_in_single_batch — 3 items arriving within the debounce window → asserts exactly 1 batch logged as 3/3
• test_items_outside_debounce_window_start_new_batch — items separated by more than the window → asserts 2 separate batches
• test_preloaded_queue_still_batches — 5 pre-loaded items → still drains as one batch

Test plan

• All 156 tests pass (pytest tests/ -v)
• Observe logs on next DA zone push burst — should see 📦 Batch complete — N/N with N > 1 instead of repeated 1/1

[guisea]

Additional fixes pushed to this branch

Two more issues found while tracing a Let's Encrypt DNS-01 challenge failure (both cluster nodes received the push, but LE still failed).

Reload logs promoted to INFO

NSD and BIND reload success was logged at DEBUG — invisible at the default INFO level. Changed to INFO so reload events appear in production without needing DADNS_LOG_LEVEL=debug.

Reload failure now triggers retry

reload_zone() return value was silently discarded in _process_single_backend and _delete_single_backend. A failed reload returned True to the caller, the batch reported success, and NSD kept serving stale zone data with no log evidence. Reload failure now returns False, emitting a warning and scheduling a retry via the existing backoff queue.

Test mock corrected

reload_zone.return_value was None (falsy) in the mock backend fixture — this bug was invisible because the return value was ignored. Fixed to True; the two debounce tests now correctly validate the full success path.