altcodepro-polydb-python 2.5.19__tar.gz → 2.5.20__tar.gz

{altcodepro_polydb_python-2.5.19/src/altcodepro_polydb_python.egg-info → altcodepro_polydb_python-2.5.20}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: altcodepro-polydb-python
-Version: 2.5.19
+Version: 2.5.20
 Summary: Production-ready multi-cloud database abstraction layer with connection pooling, retry logic, and thread safety
 Author: AltCodePro
 Project-URL: Homepage, https://github.com/altcodepro/polydb-python

{altcodepro_polydb_python-2.5.19 → altcodepro_polydb_python-2.5.20}/pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "setuptools.build_meta"
 
 [project]
 name = "altcodepro-polydb-python"
-version = "2.5.19"
+version = "2.5.20"
 description = "Production-ready multi-cloud database abstraction layer with connection pooling, retry logic, and thread safety"
 readme = { file = "README.md", content-type = "text/markdown" }
 requires-python = ">=3.11"

{altcodepro_polydb_python-2.5.19 → altcodepro_polydb_python-2.5.20/src/altcodepro_polydb_python.egg-info}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: altcodepro-polydb-python
-Version: 2.5.19
+Version: 2.5.20
 Summary: Production-ready multi-cloud database abstraction layer with connection pooling, retry logic, and thread safety
 Author: AltCodePro
 Project-URL: Homepage, https://github.com/altcodepro/polydb-python

{altcodepro_polydb_python-2.5.19 → altcodepro_polydb_python-2.5.20}/src/altcodepro_polydb_python.egg-info/SOURCES.txt

@@ -99,6 +99,7 @@ tests/test_azure_table_unpack_entity.py
 tests/test_blockchain.py
 tests/test_blockchain_overflow_guard.py
 tests/test_cloud_factory.py
+tests/test_factory_wiring.py
 tests/test_field_encryption.py
 tests/test_gcp.py
 tests/test_kafka.py

{altcodepro_polydb_python-2.5.19 → altcodepro_polydb_python-2.5.20}/src/polydb/__init__.py

@@ -4,7 +4,7 @@ PolyDB - Enterprise Cloud-Independent Database Abstraction
 Full LINQ support, field-level audit, cache, soft delete, overflow storage
 """
 
-__version__ = "2.5.18"
+__version__ = "2.5.20"
 
 from .aio import AsyncDatabaseFactory, AsyncPolyDB
 from .audit.context import AuditContext

{altcodepro_polydb_python-2.5.19 → altcodepro_polydb_python-2.5.20}/src/polydb/adapters/DynamoDBAdapter.py

@@ -196,6 +196,13 @@ class DynamoDBAdapter(NoSQLKVAdapter):
     def _blob_key(self, model: type, pk: str, rk: str, checksum: str) -> str:
         return f"overflow/{self._table_name(model)}/{pk}/{rk}/{checksum}.json"
 
+    def _check_overflow(self, data: JsonDict):  # type: ignore[override]
+        """DynamoDB overflows inside `_put_raw` (`_maybe_overflow_to_s3`, into this adapter's own bucket), never in the
+        base class. The base whole-record overflow replaced the record with a stub pointing at a key in the separate
+        object store, which `_resolve_overflow` (it reads this adapter's bucket) can never find -- the record was lost
+        on read. Returning the data untouched leaves the one working path in charge."""
+        return data, None
+
     def _maybe_overflow_to_s3(
         self, model: type, pk: str, rk: str, payload: JsonDict
     ) -> Optional[JsonDict]:

{altcodepro_polydb_python-2.5.19 → altcodepro_polydb_python-2.5.20}/src/polydb/base/NoSQLKVAdapter.py

@@ -291,7 +291,9 @@ class NoSQLKVAdapter:
     ) -> Tuple[List[JsonDict], Optional[str]]:
         # Basic implementation - override per provider
         offset = int(continuation_token) if continuation_token else 0
-        results = self.query(model, query, limit=page_size + 1)
+        # Fetch through the end of THIS page (+1 to learn whether another follows) and drop what earlier pages already
+        # returned -- reading from the start every time made every page after the first a repeat of the first.
+        results = self.query(model, query, limit=offset + page_size + 1)[offset:]
 
         has_more = len(results) > page_size
         if has_more:

{altcodepro_polydb_python-2.5.19 → altcodepro_polydb_python-2.5.20}/src/polydb/databaseFactory.py

@@ -19,7 +19,7 @@ from dataclasses import dataclass, field
 from datetime import datetime
 from typing import Any, Callable, Dict, List, Optional, Set, Tuple, Union
 
-from tenacity import retry, stop_after_attempt, wait_exponential
+from tenacity import retry, retry_if_exception, stop_after_attempt, wait_exponential
 
 from .adapters.PostgreSQLAdapter import PostgreSQLAdapter
 from .audit.context import AuditContext
@@ -32,6 +32,7 @@ from .errors import AdapterConfigurationError
 from .models import PageRequest, PageResult
 from .monitoring import HealthCheck, MetricsCollector, PerformanceMonitor
 from .query import QueryBuilder
+from .retry import _is_non_retryable
 from .security import DataMasking, FieldEncryption
 from .types import JsonDict, Lookup, ModelMeta
 
@@ -53,6 +54,31 @@ _DEFAULT_RETRY = retry(
     reraise=True,
 )
 
+_TRANSIENT_MARKERS = (
+    "timeout", "timed out", "connection reset", "connection refused", "connection closed", "connection aborted",
+    "server closed the connection", "temporarily unavailable", "too many requests", "throttl", "service unavailable",
+    "serverbusy", "503", "could not connect",
+)
+
+
+def _is_transient(exc: BaseException) -> bool:
+    if _is_non_retryable(exc):
+        return False
+    if isinstance(exc, (ConnectionError, TimeoutError)):
+        return True
+    s = str(exc).lower()
+    return any(m in s for m in _TRANSIENT_MARKERS)
+
+
+def _TRANSIENT_RETRY(fn):
+    return retry(
+        retry=retry_if_exception(_is_transient),
+        wait=wait_exponential(multiplier=0.5, min=0.5, max=6),
+        stop=stop_after_attempt(3),
+        reraise=True,
+    )(fn)
+
+
 _UNIQUE_VIOLATION_MARKERS = (
     "23505",  # Postgres SQLSTATE
     "duplicate key value violates",  # Postgres message
@@ -140,6 +166,7 @@ def _extract_meta(model: Union[type, str]) -> ModelMeta:
                 provider=raw.get("provider"),
                 cache=raw.get("cache", False),
                 cache_ttl=raw.get("cache_ttl"),
+                encrypted_fields=tuple(raw.get("encrypted_fields") or ()),
             )
     return ModelMeta(storage="nosql", table=None, collection=None)
 
@@ -348,7 +375,42 @@ class DatabaseFactory:
         return result
 
     def _run(self, fn: Callable[[], Any]) -> Any:
-        return fn()
+        """Run one storage operation, retrying only TRANSIENT failures (connection loss, timeouts, throttling) with
+        exponential backoff when retries are enabled. Constraint violations, bad arguments and anything the retry
+        classifier marks non-retryable are raised at once -- a retry could not change their outcome."""
+        if not self._enable_retries:
+            return fn()
+        return _TRANSIENT_RETRY(fn)()
+
+    def _audit_write(
+        self, action: str, name: str, meta: ModelMeta, entity_id: Any, success: bool,
+        before: Optional[JsonDict], after: Optional[JsonDict], error: Optional[str],
+    ) -> None:
+        """Record a write in the audit log. Never raises: an audit failure must not turn a completed write into an
+        error. Encrypted fields are masked so the log never holds their plaintext."""
+        if not (self._enable_audit and self._audit):
+            return
+
+        def _mask(row: Optional[JsonDict]) -> Optional[JsonDict]:
+            if not row or not meta.encrypted_fields:
+                return row
+            return {k: ("[encrypted]" if k in meta.encrypted_fields else v) for k, v in row.items()}
+
+        try:
+            before_m, after_m = _mask(before), _mask(after)
+            changed = (
+                sorted(k for k in set(after_m or {}) | set(before_m or {}) if (before_m or {}).get(k) != (after_m or {}).get(k))
+                if before_m is not None and after_m is not None else None
+            )
+            if isinstance(entity_id, dict):
+                entity_id = entity_id.get("id")
+            self._audit.record(
+                action=action, model=name, entity_id=None if entity_id is None else str(entity_id),
+                storage_type=meta.storage, provider=meta.provider or "default", success=success,
+                before=before_m, after=after_m, error=error, changed_fields=changed,
+            )
+        except Exception as exc:  # noqa: BLE001
+            logger.warning("Audit write failed (non-fatal): %s", exc)
 
     def _is_sql(self, meta: ModelMeta, override: Optional[EngineOverride] = None) -> bool:
         if override and override.force_sql:
@@ -432,9 +494,12 @@ class DatabaseFactory:
                 with monitor as m:
                     result = self._run(_op)
                     m.rows_affected = 1
-                    return result
-            return self._run(_op)
-        except Exception:
+            else:
+                result = self._run(_op)
+            self._audit_write("create", name, meta, entity_id, True, None, after_plain, None)
+            return result
+        except Exception as exc:
+            self._audit_write("create", name, meta, entity_id, False, None, None, str(exc))
             raise
 
     # ──────────────────────────────────────────────────────────
@@ -467,14 +532,26 @@ class DatabaseFactory:
         projected = bool(fields or omit)
         sql_columns = select_columns(fields, omit, columns)
 
+        drop_deleted = False
         if self._soft_delete and not include_deleted:
-            query = self._apply_soft_delete_filter(query)
+            if self._is_sql(meta, engine_override):
+                query = self._apply_soft_delete_filter(query)
+            else:
+                # A NoSQL equality filter on `deleted_at = None` matches nothing when the attribute was never written
+                # (Azure Table, DynamoDB), so filter the returned rows instead.
+                drop_deleted = True
 
         adapters = self._adapters_for(model, meta, engine_override)
         use_external_cache = (
             self._enable_cache and self._cache and getattr(meta, "cache", False) and not projected
         )
         encrypted_fields = getattr(meta, "encrypted_fields", [])
+        # The cache key must cover everything that changes the result: the query, the window, and the session variables
+        # (e.g. the tenant a row-level-security policy filters by) -- otherwise two callers share an entry.
+        cache_key_query = {
+            "q": query or {}, "limit": limit, "offset": offset,
+            "sv": sorted((session_vars or {}).items()), "e": bool(engine_override),
+        }
 
         def _op() -> List[JsonDict]:
             if self._is_sql(meta, engine_override):
@@ -497,22 +574,27 @@ class DatabaseFactory:
                 nosql_kwargs: Dict[str, Any] = {}
                 if projected:
                     nosql_kwargs = {"fields": fields, "omit": omit}
+                skip = offset or 0
                 raw = adapters.nosql.query(
-                    cls, query=query, limit=limit, no_cache=no_cache or bool(use_external_cache),
-                    **nosql_kwargs,
+                    cls, query=query, limit=(limit + skip) if limit else None,
+                    no_cache=no_cache or bool(use_external_cache), **nosql_kwargs,
                 )
+                if skip:
+                    raw = raw[skip:]  # NoSQL adapters take no offset: read through it, then drop the skipped rows
+            if drop_deleted:
+                raw = [r for r in raw if not r.get("deleted_at")]
             if self.encryption and encrypted_fields:
                 raw = [self.encryption.decrypt_fields(r, encrypted_fields) for r in raw]
             if self._cache and use_external_cache and not no_cache:
                 ttl = cache_ttl or getattr(meta, "cache_ttl", 300)
                 try:
-                    self._cache.set(name, query or {}, raw, ttl)
+                    self._cache.set(name, cache_key_query, raw, ttl)
                 except Exception as _ce:
                     logger.warning("Cache set failed (non-fatal): %s", _ce)
             return raw
 
         if self._cache and use_external_cache and not no_cache:
-            cached = self._cache.get(name, query or {})
+            cached = self._cache.get(name, cache_key_query)
             if cached is not None:
                 return cached
 
@@ -677,9 +759,12 @@ class DatabaseFactory:
                 with monitor as m:
                     result = self._run(_op)
                     m.rows_affected = 1
-                    return result
-            return self._run(_op)
-        except Exception:
+            else:
+                result = self._run(_op)
+            self._audit_write("update", name, meta, entity_id, True, before, after_plain, None)
+            return result
+        except Exception as exc:
+            self._audit_write("update", name, meta, entity_id, False, before, None, str(exc))
             raise
 
     # ──────────────────────────────────────────────────────────
@@ -737,9 +822,12 @@ class DatabaseFactory:
                 with monitor as m:
                     result = self._run(_op)
                     m.rows_affected = 1
-                    return result
-            return self._run(_op)
-        except Exception:
+            else:
+                result = self._run(_op)
+            self._audit_write("upsert", name, meta, (after_plain or {}).get('id'), True, None, after_plain, None)
+            return result
+        except Exception as exc:
+            self._audit_write("upsert", name, meta, None, False, None, None, str(exc))
             raise
 
     # ──────────────────────────────────────────────────────────
@@ -850,9 +938,12 @@ class DatabaseFactory:
                 with monitor as m:
                     result = self._run(_op)
                     m.rows_affected = 1
-                    return result
-            return self._run(_op)
-        except Exception:
+            else:
+                result = self._run(_op)
+            self._audit_write("delete", name, meta, entity_id, True, before, None, None)
+            return result
+        except Exception as exc:
+            self._audit_write("delete", name, meta, entity_id, False, before, None, str(exc))
             raise
 
     # ──────────────────────────────────────────────────────────
@@ -906,8 +997,12 @@ class DatabaseFactory:
         name = _model_name(model)
         meta = _extract_meta(model)
 
+        drop_deleted = False
         if self._soft_delete and not include_deleted:
-            query = self._apply_soft_delete_filter(query)
+            if self._is_sql(meta, engine_override):
+                query = self._apply_soft_delete_filter(query)
+            else:
+                drop_deleted = True
 
         adapters = self._adapters_for(model, meta, engine_override)
         encrypted_fields = getattr(meta, "encrypted_fields", [])
@@ -924,6 +1019,8 @@ class DatabaseFactory:
                     else type(name, (), {"__polydb__": meta.__dict__})
                 )
                 raw, token = adapters.nosql.query_page(cls, query, page_size, continuation_token)
+                if drop_deleted:
+                    raw = [r for r in raw if not r.get("deleted_at")]
             if self.encryption and encrypted_fields:
                 raw = [self.encryption.decrypt_fields(r, encrypted_fields) for r in raw]
             return raw, token

{altcodepro_polydb_python-2.5.19 → altcodepro_polydb_python-2.5.20}/src/polydb/schema.py

@@ -131,6 +131,27 @@ class SchemaBuilder:
 
         return sql
 
+    def to_add_missing_columns(self, table_name: str) -> List[str]:
+        """`ALTER TABLE ... ADD COLUMN IF NOT EXISTS` for every non-key column. `CREATE TABLE IF NOT EXISTS` never adds a
+        column that was declared after the table was first created; appending these to the same migration makes an edited
+        column list reach existing tables. Added columns are always nullable (existing rows have no value) and never
+        UNIQUE; a declared default is kept."""
+        validate_table_name(table_name)
+        keys = set(self.primary_keys or [])
+        out: List[str] = []
+        for col in self.columns:
+            if col.name in keys or col.primary_key:
+                continue
+            parts = [validate_column_name(col.name)]
+            if col.type == ColumnType.VARCHAR and col.max_length:
+                parts.append(f"VARCHAR({col.max_length})")
+            else:
+                parts.append(col.type.value)
+            if col.default is not None:
+                parts.append(_render_default(col.default))
+            out.append(f"ALTER TABLE {table_name} ADD COLUMN IF NOT EXISTS {' '.join(parts)};")
+        return out
+
     def to_create_indexes(self, table_name: str) -> List[str]:
         """Generate CREATE INDEX statements"""
         statements = []
@@ -185,15 +206,31 @@ class MigrationManager:
 
         # Check if already applied
         existing = self.sql.execute(
-            "SELECT version FROM polydb_migrations WHERE version = %s", [version], fetch_one=True
+            "SELECT version, checksum FROM polydb_migrations WHERE version = %s", [version], fetch_one=True
         )
 
-        if existing:
-            return False
-
         # Calculate checksum
         checksum = hashlib.sha256(up_sql.encode()).hexdigest()
 
+        if existing:
+            stored = existing.get("checksum") if isinstance(existing, dict) else None
+            if not stored or stored == checksum:
+                return False
+            # The DDL under this version was edited (a column added, an index changed). Re-apply it -- the statements
+            # this codebase generates are idempotent (IF NOT EXISTS) -- and remember the new checksum. A failure here
+            # must not stop a deployment from starting: keep the old checksum so the next start tries again.
+            try:
+                self.sql.execute(up_sql)
+                self.sql.execute(
+                    "UPDATE polydb_migrations SET checksum = %s WHERE version = %s", [checksum, version]
+                )
+                return True
+            except Exception as exc:  # noqa: BLE001
+                import logging
+
+                logging.getLogger(__name__).warning("Re-applying edited migration %s failed (kept old): %s", version, exc)
+                return False
+
         try:
             # Execute migration
             self.sql.execute(up_sql)

{altcodepro_polydb_python-2.5.19 → altcodepro_polydb_python-2.5.20}/src/polydb/types.py

@@ -24,6 +24,8 @@ class ModelMeta:
     provider: Optional[str] = None
     cache: bool = False
     cache_ttl: Optional[int] = None
+    # Fields stored encrypted when the factory has encryption enabled (a tuple: the dataclass is frozen/hashable).
+    encrypted_fields: Tuple[str, ...] = ()
 
 
 @runtime_checkable

altcodepro_polydb_python-2.5.20/tests/test_factory_wiring.py

@@ -0,0 +1,247 @@
+"""DatabaseFactory wiring: encrypted_fields reach the factory, writes are audited, transient failures retry, and the
+read-cache key covers the window and session variables. Unit level: no database, fake adapters."""
+import types
+
+import pytest
+
+from polydb.databaseFactory import DatabaseFactory, _extract_meta
+from polydb.security import FieldEncryption
+
+
+class _Sql:
+    def __init__(self):
+        self.rows, self.fail_times, self.calls = {}, 0, 0
+
+    def insert(self, table, data, session_vars=None):
+        self.calls += 1
+        if self.fail_times:
+            self.fail_times -= 1
+            raise TimeoutError("connection timed out")
+        self.rows[data["id"]] = dict(data)
+        return dict(data)
+
+    def select(self, table, query=None, limit=None, offset=None, session_vars=None, fields=None):
+        self.calls += 1
+        return [dict(r) for r in self.rows.values()]
+
+    def update(self, table, entity_id, data, session_vars=None):
+        eid = entity_id["id"] if isinstance(entity_id, dict) else entity_id
+        self.rows[eid].update(data)
+        return dict(self.rows[eid])
+
+    def delete(self, table, entity_id, session_vars=None):
+        self.rows.pop(entity_id["id"] if isinstance(entity_id, dict) else entity_id, None)
+        return {"deleted": True}
+
+
+class _Cache:
+    def __init__(self):
+        self.store, self.keys = {}, []
+
+    def get(self, model, query):
+        self.keys.append(query)
+        return self.store.get((model, repr(sorted(query.items()))))
+
+    def set(self, model, query, value, ttl=None):
+        self.store[(model, repr(sorted(query.items())))] = value
+
+    def invalidate(self, model, query=None):
+        self.store.clear()
+
+
+class _Audit:
+    def __init__(self):
+        self.records = []
+
+    def record(self, **kw):
+        self.records.append(kw)
+
+
+def _factory(sql, *, audit=None, cache=None, encryption=None, retries=True):
+    f = object.__new__(DatabaseFactory)
+    f._enable_retries, f._enable_audit, f._audit = retries, audit is not None, audit
+    f._enable_cache, f._cache, f._soft_delete = cache is not None, cache, False
+    f.encryption, f.metrics = encryption, None
+    f._adapters_for = lambda model, meta, override=None: types.SimpleNamespace(sql=sql, nosql=None)
+    return f
+
+
+class Note:
+    __polydb__ = {"storage": "sql", "table": "notes", "encrypted_fields": ["secret"], "cache": True, "cache_ttl": 60}
+
+
+def test_encrypted_fields_reach_the_model_metadata():
+    assert _extract_meta(Note).encrypted_fields == ("secret",)
+    assert _extract_meta(type("X", (), {"__polydb__": {"storage": "sql", "table": "t"}})).encrypted_fields == ()
+
+
+def test_encryption_runs_for_a_model_that_declares_encrypted_fields(monkeypatch):
+    monkeypatch.setenv("POLYDB_ENCRYPTION_KEY", __import__("base64").b64encode(b"k" * 32).decode())
+    sql = _Sql()
+    try:
+        enc = FieldEncryption()
+    except Exception:
+        pytest.skip("FieldEncryption needs a configured key in this environment")
+    f = _factory(sql, encryption=enc)
+    out = f.create(Note, {"id": "1", "secret": "plain"})
+    assert sql.rows["1"]["secret"] != "plain"      # stored encrypted
+    assert out["secret"] == "plain"                # returned decrypted
+
+
+def test_writes_are_audited_and_encrypted_values_are_masked():
+    sql, audit = _Sql(), _Audit()
+    f = _factory(sql, audit=audit)
+    f.create(Note, {"id": "1", "secret": "plain", "title": "a"})
+    kinds = [r["action"] for r in audit.records]
+    assert kinds == ["create"] and audit.records[0]["success"] is True
+    assert audit.records[0]["after"]["secret"] == "[encrypted]" and audit.records[0]["after"]["title"] == "a"
+
+
+def test_an_audit_failure_never_breaks_the_write():
+    class Boom:
+        def record(self, **kw):
+            raise RuntimeError("audit store down")
+    sql = _Sql()
+    assert _factory(sql, audit=Boom()).create(Note, {"id": "1"})["id"] == "1"
+
+
+def test_transient_errors_are_retried_and_constraint_errors_are_not(monkeypatch):
+    import polydb.databaseFactory as df
+    monkeypatch.setattr(df, "_TRANSIENT_RETRY", lambda fn: __import__("tenacity").retry(
+        retry=df.retry_if_exception(df._is_transient), stop=df.stop_after_attempt(3), reraise=True)(fn))
+    sql = _Sql()
+    sql.fail_times = 2
+    _factory(sql).create(Note, {"id": "1"})
+    assert sql.calls == 3
+
+    calls = []
+    f = _factory(sql)
+
+    def op():
+        calls.append(1)
+        raise ValueError('duplicate key value violates unique constraint "x"')
+    with pytest.raises(ValueError):
+        f._run(op)
+    assert len(calls) == 1
+
+
+def test_retries_can_be_turned_off():
+    f = _factory(_Sql(), retries=False)
+    calls = []
+
+    def op():
+        calls.append(1)
+        raise TimeoutError("timeout")
+    with pytest.raises(TimeoutError):
+        f._run(op)
+    assert len(calls) == 1
+
+
+def test_the_read_cache_key_covers_session_variables_and_the_window():
+    sql, cache = _Sql(), _Cache()
+    sql.rows["1"] = {"id": "1"}
+    f = _factory(sql, cache=cache)
+    f.read(Note, {"x": 1}, session_vars={"app.tenant_id": "a"})
+    f.read(Note, {"x": 1}, session_vars={"app.tenant_id": "b"})   # must NOT be served tenant a's entry
+    f.read(Note, {"x": 1}, session_vars={"app.tenant_id": "a"}, limit=5)
+    assert sql.calls == 3
+    f.read(Note, {"x": 1}, session_vars={"app.tenant_id": "a"})   # same caller, same query: a hit
+    assert sql.calls == 3
+
+
+# ---------------- reported adapter findings ----------------------------------------------------------------------------
+
+def test_nosql_query_page_advances_instead_of_repeating_the_first_page():
+    from polydb.base.NoSQLKVAdapter import NoSQLKVAdapter
+
+    class A(NoSQLKVAdapter):
+        def _query_raw(self, model, filters, limit, select=None):
+            rows = [{"id": str(i)} for i in range(10)]
+            return rows[:limit] if limit else rows
+
+    a = A()
+    first, token = a.query_page(object, {}, 4)
+    second, token2 = a.query_page(object, {}, 4, token)
+    third, token3 = a.query_page(object, {}, 4, token2)
+    assert [r["id"] for r in first] == ["0", "1", "2", "3"]
+    assert [r["id"] for r in second] == ["4", "5", "6", "7"]
+    assert [r["id"] for r in third] == ["8", "9"] and token3 is None
+
+
+def test_dynamodb_record_over_the_limit_is_not_replaced_by_a_stub():
+    from polydb.adapters.DynamoDBAdapter import DynamoDBAdapter
+
+    data = {"id": "1", "blob": "x" * 500_000}
+    store, key = DynamoDBAdapter._check_overflow(object.__new__(DynamoDBAdapter), data)
+    assert store is data and key is None
+
+
+def test_soft_deleted_rows_are_hidden_on_nosql_even_when_the_attribute_was_never_written():
+    class _NoSql:
+        def query(self, cls, query=None, limit=None, no_cache=False, **kw):
+            return [{"id": "1"}, {"id": "2", "deleted_at": "2026-01-01"}, {"id": "3", "deleted_at": None}]
+
+    class Doc:
+        __polydb__ = {"storage": "nosql"}
+
+    f = _factory(None)
+    f._soft_delete = True
+    f._adapters_for = lambda model, meta, override=None: types.SimpleNamespace(sql=None, nosql=_NoSql())
+    assert [r["id"] for r in f.read(Doc, {})] == ["1", "3"]
+    assert len(f.read(Doc, {}, include_deleted=True)) == 3
+
+
+def test_nosql_read_honours_offset():
+    seen = {}
+
+    class _NoSql:
+        def query(self, cls, query=None, limit=None, no_cache=False, **kw):
+            seen["limit"] = limit
+            return [{"id": str(i)} for i in range(10)][:limit]
+
+    class Doc:
+        __polydb__ = {"storage": "nosql"}
+
+    f = _factory(None)
+    f._adapters_for = lambda model, meta, override=None: types.SimpleNamespace(sql=None, nosql=_NoSql())
+    out = f.read(Doc, {}, limit=3, offset=4)
+    assert [r["id"] for r in out] == ["4", "5", "6"] and seen["limit"] == 7
+
+
+def test_edited_migration_ddl_is_reapplied_and_unchanged_ddl_is_not():
+    import hashlib
+
+    from polydb.schema import MigrationManager
+
+    class _S:
+        def __init__(self):
+            self.sql_log, self.stored = [], {}
+
+        def execute(self, sql, params=None, fetch_one=False, **kw):
+            self.sql_log.append(sql)
+            if sql.startswith("SELECT version"):
+                v = params[0]
+                return {"version": v, "checksum": self.stored[v]} if v in self.stored else None
+            if sql.startswith("UPDATE polydb_migrations"):
+                self.stored[params[1]] = params[0]
+            return None
+
+        def insert(self, table, data):
+            self.stored[data["version"]] = data["checksum"]
+
+    s = _S()
+    m = object.__new__(MigrationManager)
+    m.sql = s
+    assert m.apply_migration("v1", "n", "CREATE TABLE a (x int);") is True
+    assert m.apply_migration("v1", "n", "CREATE TABLE a (x int);") is False                 # unchanged
+    assert m.apply_migration("v1", "n", "CREATE TABLE a (x int); ALTER TABLE a ADD COLUMN IF NOT EXISTS y int;") is True
+    assert s.stored["v1"] == hashlib.sha256(b"CREATE TABLE a (x int); ALTER TABLE a ADD COLUMN IF NOT EXISTS y int;").hexdigest()
+
+
+def test_schema_builder_emits_add_column_statements_for_non_key_columns():
+    from polydb.schema import Column, ColumnType, SchemaBuilder
+
+    b = SchemaBuilder()
+    b.add_column(Column(name="id", type=ColumnType.VARCHAR, nullable=False, primary_key=True))
+    b.add_column(Column(name="title", type=ColumnType.TEXT, nullable=False))
+    assert b.to_add_missing_columns("notes") == ["ALTER TABLE notes ADD COLUMN IF NOT EXISTS title TEXT;"]