Declaration accepts auto-populated tables that declare their own primary-key attributes

オープン
#1,576 コメント 3 件 リアクション 0 件 担当者 0 名 GitHub で見る

メンテナーはふだん 6 日以内に返信

まだ誰も着手していません。

評価

難易度
3/5
見積もり時間
1〜2日
初心者へのやさしさ
52/100
issue の種類
機能追加
明瞭さ
おおむね明確
活発さ
活発
技術スタック
mysql, python
領域
databases

調査の方向性

The check already exists in two places: _get_fk_derived_pk_attrs in jobs.py and introduces_dimension in diagram.py, so start by reading how each decides which primary-key attributes come from foreign keys. Then find where a Computed or Imported class is declared and add the same test there, emitting a warning that names the attributes and links explanation/entity-integrity.md. Done means the warning fires at declaration, Part tables are not flagged, and the diagram and the jobs table agree with the new check. The raise behind a config flag is optional and needs a maintainer decision on the setting name.

索引モデルが issue の本文から書いたものです。

説明

Summary

Computed and Imported tables may declare primary-key attributes of their own, beyond their foreign keys. DataJoint accepts them at declaration without a warning; the only signal comes later, and only when job reservation is used: the jobs table logs Ignoring non-FK primary key attribute '<attr>' in jobs table for <table>. Job granularity will be degraded. The documentation says such a table should not exist: a new dimension in an auto-populated table belongs in a part table (explanation/entity-integrity.md, "Autopopulated tables introduce dimensions only through part tables"). #1523 states the same rule in passing: an auto-populated table introduces no new dimension, only its parts may. Declaration does not check it, and the resulting table behaves badly in a way that is easy to miss.

Reproduction

DataJoint 2.3.5.dev2 (master), MySQL 8.0:

import datajoint as dj

schema = dj.Schema("scratch_dimrule")
calls = []

@schema
class Image(dj.Manual):
    definition = """
    image_id : int32
    ---
    n_blobs : int32          # stands in for what detection would find
    """

@schema
class Blob(dj.Computed):
    definition = """
    -> Image
    blob_idx : int32         # declared by the auto-populated table itself
    ---
    x : float32
    """

    def make(self, key):
        calls.append(key["image_id"])
        n = (Image & key).fetch1("n_blobs")
        self.insert(dict(key, blob_idx=i, x=float(i)) for i in range(n))

Image.insert([dict(image_id=1, n_blobs=2), dict(image_id=2, n_blobs=0)])
Blob.populate()
Blob.populate()
Blob.populate()
print("rows:", len(Blob()))
print("make() calls by image_id:", calls)
print("remaining in key_source - Blob:", len(Blob.key_source - Blob))

Output:

rows: 2
make() calls by image_id: [1, 2, 2, 2]
remaining in key_source - Blob: 1

The declaration raises no error and emits no warning, and neither does populate() with the default reserve_jobs=False. Blob.primary_key is ['image_id', 'blob_idx'].

With job reservation the table does draw a warning, from the jobs table rather than from declaration. Blob.populate(reserve_jobs=True), Blob.jobs, and Blob.jobs.refresh() each log:

Ignoring non-FK primary key attribute 'blob_idx' in jobs table for `scratch_dimrule`.`__blob`. Job granularity will be degraded.

That message (jobs.py, _get_fk_derived_pk_attrs) describes a consequence for job bookkeeping, not the modeling problem, and a pipeline that never reserves jobs never sees it.

Why it matters

The table has no row that stands for the computation as a whole, only rows for its pieces. A key whose computation legitimately yields zero pieces (image 2 above: no blobs found) leaves nothing behind. It stays in key_source - Blob and is recomputed on every populate(), indefinitely; with reserve_jobs=True it is re-queued each time. progress() reports it as remaining forever. The same holds for any downstream table that depends on Blob: it cannot tell "no blobs" from "not computed yet".

The documented pattern has no such problem. The master records the computation (num_blobs = 0), the part holds the pieces, and both are inserted in one transaction by the master's make():

@schema
class Detection(dj.Computed):
    definition = """
    -> Image
    ---
    num_blobs : int32
    """

    class Blob(dj.Part):
        definition = """
        -> master
        blob_idx : int32
        ---
        x : float32
        """

Proposal

At declaration, when a Computed or Imported table's primary key contains attributes that do not come from its foreign keys (the same condition the jobs table already detects):

  • warn, naming the attributes and pointing to the part-table pattern; and
  • optionally, raise behind a config setting (e.g. config.declaration.strict_dimensions = True), for teams that want the rule enforced.

The check is the same one dj.Diagram already makes to underline a table that introduces a dimension (introduces_dimension in diagram.py), so the two would agree: an auto-populated table would never be underlined, and a Part table could be.

Existing tables declared this way keep working. The warning is emitted when the class is declared or activated, not on every query. The jobs table's message could then point to the same explanation, or be folded into it.

主要言語
Python
スター
198
フォーク
103
平均マージ
6日 1時間
マージ済み PR(30日)
10

環境構築

はじめの一歩

  1. issue を最後まで読み、次にプロジェクトのコントリビューションガイドを読みます。
  2. 着手することを issue にコメントします — 二人が同じ作業をするのを防げます。
  3. リポジトリをフォークし、ブランチを切って変更します。
  4. issue 番号を参照したプルリクエストを送ります。

datajoint/datajoint-python のほかの issue

datajoint/datajoint-python の issue をすべて見る

似ている issue

Python の issue をもっと見る

新しい issue をメールで受け取る

初心者向けの GitHub issue を短くまとめたダイジェスト。