fix(tests): object-store retention-prune tests no longer expire with the wall clock (#1066)

CHANGELOG.md

@@ -82,6 +82,8 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 
 ### Fixed
 
+- **Object-store history retention tests no longer expire with the wall clock** ([#1066](https://github.com/drt-hub/drt/issues/1066)): prune/cap fixtures now derive ordered run timestamps from the test-run time instead of hardcoding August 2026 dates, preventing clean `main` CI from changing its expected removed count as the calendar advances.
+
 - **BigQuery, Airtable, and Klaviyo now pass `drt validate --check-connection` with their documented least-privilege credentials** ([#1059](https://github.com/drt-hub/drt/issues/1059)): all three connectors were excluded from #1049's structural `ConnectionTestable` dispatch because their probes needed broader API scope than each connector's own docs recommend as the minimal write-only credential. Fixed with a real least-privilege probe per API, verified against each vendor's current documentation rather than guessed: BigQuery's `test_connection()` now calls `get_table()` (needs only `bigquery.tables.get`, already included in the `roles/bigquery.dataEditor` role that grants the documented `bigquery.tables.updateData`) instead of running a `SELECT 1` query job (needs the separate, broader `bigquery.jobs.create`). Airtable's now calls `GET /v0/meta/whoami`, Airtable's own documented scope-free identity endpoint, instead of reading a record (`data.records:read`). Klaviyo has no scope-free identity endpoint, so it still probes `GET /accounts/` but now also accepts Klaviyo's documented `403` + `code: "permission_denied"` response as proof of a valid (if more narrowly scoped) key, rather than treating every non-2xx as a failure — any other error still raises normally. All three are back in the normal `ConnectionTestable` dispatch; no exclusion list needed.
 
 - **Two more `drt run --dry-run --diff` preview correctness bugs, both pre-existing before #1044** ([#1061](https://github.com/drt-hub/drt/issues/1061), [#1062](https://github.com/drt-hub/drt/issues/1062)): (1) the tracked-mirror preview ignored `sync.mirror.scope` entirely — a run touching one tenant/parent could preview another tenant's stale keys as deletions, contradicting what the real `_finalize_mirror_tracked()` actually does. Now filters prior state to the scopes this run's source records actually produced, matching the real finalizer, across all five SQL dialects; unscoped tracked mirror is unaffected. (2) `compute_diff()`'s replace-mode full-table scan calls `fetch_rows(..., columns=[])`, and Postgres/MySQL/Snowflake/ClickHouse's readers built result dicts via `dict(zip([], row))` — always `{}` — silently collapsing every destination row into one keyless entry and severely understating a real `replace` run's deletions (Databricks was already fixed for this in #1044). All four now derive column names from query metadata (`cursor.description` / ClickHouse's `column_names`) when `columns` is empty, matching Databricks' existing fix.

tests/unit/test_state_objectstore.py

@@ -4,7 +4,7 @@
 
 import logging
 import threading
-from datetime import datetime, timezone
+from datetime import datetime, timedelta, timezone
 from pathlib import Path
 from unittest.mock import patch
 
@@ -257,12 +257,16 @@ def test_history_append_non_precondition_error_warns_without_retry(
 def test_remote_history_applies_entry_cap_only_during_prune() -> None:
     client = MemoryObjectClient()
     store = ObjectStoreHistoryStore(client, max_entries=2)
-    for day in range(1, 4):
-        store.append(_history(started_at=f"2026-08-0{day}T00:00:00+00:00"))
+    now = datetime.now(timezone.utc)
+    timestamps = [
+        (now - timedelta(days=days_ago)).isoformat() for days_ago in (3, 2, 1)
+    ]
+    for started_at in timestamps:
+        store.append(_history(started_at=started_at))
 
     assert len(store.read("s", limit=20)) == 3
     assert store.prune("s", retention_days=30) == 1
-    assert [entry.started_at[9] for entry in store.read("s")] == ["3", "2"]
+    assert [entry.started_at for entry in store.read("s")] == timestamps[::-1][:2]
 
 
 def test_prune_caps_by_started_at_not_append_order() -> None:
@@ -274,14 +278,18 @@ def test_prune_caps_by_started_at_not_append_order() -> None:
     """
     client = MemoryObjectClient()
     store = ObjectStoreHistoryStore(client, max_entries=2)
-    # Appended oldest-started-at last, as if a later-starting run's batch
+    now = datetime.now(timezone.utc)
+    oldest = (now - timedelta(days=3)).isoformat()
+    middle = (now - timedelta(days=2)).isoformat()
+    newest = (now - timedelta(days=1)).isoformat()
+    # Append out of started_at order, as if a later-starting run's batch
     # observer flushed and completed before an earlier-starting run did.
-    store.append(_history(started_at="2026-08-03T00:00:00+00:00"))
-    store.append(_history(started_at="2026-08-01T00:00:00+00:00"))
-    store.append(_history(started_at="2026-08-02T00:00:00+00:00"))
+    store.append(_history(started_at=newest))
+    store.append(_history(started_at=oldest))
+    store.append(_history(started_at=middle))
 
     assert store.prune("s", retention_days=30) == 1
-    assert [entry.started_at[9] for entry in store.read("s")] == ["3", "2"]
+    assert [entry.started_at for entry in store.read("s")] == [newest, middle]
 
 
 def test_history_reads_all_syncs_newest_first_and_prunes_old_entries() -> None: