googleapis / googleapis/google-cloud-python
google-cloud-ndb: Expando dynamic property updates are not persisted by put()
- 主要言語
- Python
- スター
- 5.4k
- フォーク
- 1.8k
- 平均マージ
- 3日 4時間
- マージ済み PR(30日)
- 122
説明
`google-cloud-ndb==2.3.0`, Python 3.11.
If I assign a dynamic property on an Expando and `put()`, the first write sticks. Changing that same property later and `put()` again looks fine (no error; in-process it even reads as the new value) but a refetch from datastore still has the old number.
Declared properties on the class (e.g. `FloatProperty`) update correctly. This also worked on the old App Engine ndb.
```python
from google.cloud import ndb
class Item(ndb.Expando):
name = ndb.StringProperty()
# ndb client context already active
ent = Item(name="x")
ent.extra = 2.00
key = ent.put()
ent = key.get()
ent.extra = 9.99
ent.put()
ent = key.get()
print(ent.extra) # 2.00, expected 9.99
```
`Expando.__setattr__` looks like the cause. If the name is already in `_properties` it calls `super().__setattr__`. For a class-level Property that hits the descriptor and updates `_values`. For a dynamic name there is no descriptor on the class, so it lands in `__dict__` and `put()` serializes the stale `_values`.
Is this expected? I couldn't find it documented. The old runtime always went through `_set_value` for non-class names.
Opened here because https://github.com/googleapis/python-ndb is archived.
コントリビューションガイド
調査の方向性
Start with google-cloud-ndb's Expando.__setattr__ entry point and follow its _properties, _values, and _set_value handling. Reproduce the two-put example, then add or update a regression test showing that changing a dynamic property persists after refetch while declared properties continue to update correctly.
索引モデルが issue の本文から書いたものです。
評価
- 技術スタック
- google-cloud, python
- 領域
- backend, databases
- issue の種類
- バグ
- 難易度
- 3/5
- 見積もり時間
- 1〜2日
- 活発さ
- 活発
- 明瞭さ
- おおむね明確
- 初心者へのやさしさ
- 70/100