Open-API: TableRequirements should use union of subclasses #10434

anuragmantri · 2024-06-03T20:42:20Z

Since we are using the discriminator field, the type property is inherited from TableRequirements. This PR removed redundant type properties in child classes.

anuragmantri · 2024-06-03T20:42:49Z

@flyrain could you please take a look?

Fokko · 2024-06-03T20:45:17Z

Thanks for raising this @anuragmantri

Looking at the changes in the code, it will remove the inheritance:

diff --git a/open-api/rest-catalog-open-api.py b/open-api/rest-catalog-open-api.py
index e13aeee..46c7eca 100644
--- a/open-api/rest-catalog-open-api.py
+++ b/open-api/rest-catalog-open-api.py
@@ -361,10 +361,10 @@ class RemovePartitionStatisticsUpdate(BaseUpdate):
 
 
 class TableRequirement(BaseModel):
-    type: str
+    pass
 
 
-class AssertCreate(TableRequirement):
+class AssertCreate(BaseModel):
     """
     The table must not already exist; used for create transactions
     """
@@ -372,7 +372,7 @@ class AssertCreate(TableRequirement):
     type: Literal['assert-create']
 
 
-class AssertTableUUID(TableRequirement):
+class AssertTableUUID(BaseModel):
     """
     The table UUID must match the requirement's `uuid`
     """
@@ -381,7 +381,7 @@ class AssertTableUUID(TableRequirement):
     uuid: str
 
 
-class AssertRefSnapshotId(TableRequirement):
+class AssertRefSnapshotId(BaseModel):
     """
     The table branch or tag identified by the requirement's `ref` must reference the requirement's `snapshot-id`; if `snapshot-id` is `null` or missing, the ref must not already exist
     """
@@ -391,7 +391,7 @@ class AssertRefSnapshotId(TableRequirement):
     snapshot_id: int = Field(..., alias='snapshot-id')
 
 
-class AssertLastAssignedFieldId(TableRequirement):
+class AssertLastAssignedFieldId(BaseModel):
     """
     The table's last assigned column id must match the requirement's `last-assigned-field-id`
     """
@@ -400,7 +400,7 @@ class AssertLastAssignedFieldId(TableRequirement):
     last_assigned_field_id: int = Field(..., alias='last-assigned-field-id')
 
 
-class AssertCurrentSchemaId(TableRequirement):
+class AssertCurrentSchemaId(BaseModel):
     """
     The table's current schema id must match the requirement's `current-schema-id`
     """
@@ -409,7 +409,7 @@ class AssertCurrentSchemaId(TableRequirement):
     current_schema_id: int = Field(..., alias='current-schema-id')
 
 
-class AssertLastAssignedPartitionId(TableRequirement):
+class AssertLastAssignedPartitionId(BaseModel):
     """
     The table's last assigned partition id must match the requirement's `last-assigned-partition-id`
     """
@@ -4[18](https://github.com/apache/iceberg/actions/runs/9356980269/job/25755651734?pr=10434#step:7:19),7 +418,7 @@ class AssertLastAssignedPartitionId(TableRequirement):
     last_assigned_partition_id: int = Field(..., alias='last-assigned-partition-id')
 
 
-class AssertDefaultSpecId(TableRequirement):
+class AssertDefaultSpecId(BaseModel):
     """
     The table's default spec id must match the requirement's `default-spec-id`
     """
@@ -4[27](https://github.com/apache/iceberg/actions/runs/9356980269/job/25755651734?pr=10434#step:7:28),7 +427,7 @@ class AssertDefaultSpecId(TableRequirement):
     default_spec_id: int = Field(..., alias='default-spec-id')
 
 
-class AssertDefaultSortOrderId(TableRequirement):
+class AssertDefaultSortOrderId(BaseModel):
     """
     The table's default sort order id must match the requirement's `default-sort-order-id`
     """

flyrain · 2024-06-03T21:31:57Z

We don't have to use inheritance here, but I'm OK with it.
If we choose the inheritance, we will also need to remove the type properties in each subclasses.

anuragmantri · 2024-06-03T22:49:55Z

Thanks @Fokko and @flyrain - I think we should keep the inheritance and remove the duplicates in sub types. I updated this PR to do that.

@flyrain this pattern of discriminator field present also as child field is used in 3 other places

ContentFile: data-file, position-delete-file etc
BaseUpdate: update-uuid, upgrade-format-version etc
ViewRequirement: assert-view-uuid

Should we remove child fields and move the descriptor fields to parent all these places to be consistent?

flyrain

+1 Thanks @anuragmantri for working on it. We can remove the duplication for the other schema in different PRs.

open-api/rest-catalog-open-api.yaml

Fokko · 2024-06-04T06:41:18Z

open-api/rest-catalog-open-api.py

    The table must not already exist; used for create transactions
    """

-    type: Literal['assert-create']


I'm sad to see this go, but it looks like it is a limitation of the open-api generator, since it could also get this information from the mapping field. I would love to get to know what others think

It looks like we're using it in an odd way here. The example here also does not use inheritance but returns an union:

class Cat(BaseModel): pet_type: Literal['cat'] meows: int class Dog(BaseModel): pet_type: Literal['dog'] barks: float class Lizard(BaseModel): pet_type: Literal['reptile', 'lizard'] scales: bool class Model(BaseModel): pet: Union[Cat, Dog, Lizard] = Field(..., discriminator='pet_type') n: int

It is just that we try to add this field in there. I think the following is more correct:

diff --git a/open-api/rest-catalog-open-api.yaml b/open-api/rest-catalog-open-api.yaml index f8f68d06df..b85207e57b 100644 --- a/open-api/rest-catalog-open-api.yaml +++ b/open-api/rest-catalog-open-api.yaml @@ -2589,18 +2589,6 @@ components: type: object required: - type - properties: - type: - type: "string" - enum: - - "assert-create" - - "assert-table-uuid" - - "assert-ref-snapshot-id" - - "assert-last-assigned-field-id" - - "assert-current-schema-id" - - "assert-last-assigned-partition-id" - - "assert-default-spec-id" - - "assert-default-sort-order-id" discriminator: propertyName: type mapping: @@ -2612,6 +2600,15 @@ components: assert-last-assigned-partition-id: '#/components/schemas/AssertLastAssignedPartitionId' assert-default-spec-id: '#/components/schemas/AssertDefaultSpecId' assert-default-sort-order-id: '#/components/schemas/AssertDefaultSortOrderId' + oneOf: + - $ref: '#/components/schemas/AssertCreate' + - $ref: '#/components/schemas/AssertTableUUID' + - $ref: '#/components/schemas/AssertRefSnapshotId' + - $ref: '#/components/schemas/AssertLastAssignedFieldId' + - $ref: '#/components/schemas/AssertCurrentSchemaId' + - $ref: '#/components/schemas/AssertLastAssignedPartitionId' + - $ref: '#/components/schemas/AssertDefaultSpecId' + - $ref: '#/components/schemas/AssertDefaultSortOrderId' AssertCreate: allOf: @@ -2636,6 +2633,7 @@ components: The table branch or tag identified by the requirement's `ref` must reference the requirement's `snapshot-id`; if `snapshot-id` is `null` or missing, the ref must not already exist required: + - type - ref - snapshot-id properties: @@ -2651,8 +2649,12 @@ components: description: The table's last assigned column id must match the requirement's `last-assigned-field-id` required: + - type - last-assigned-field-id properties: + type: + type: string + enum: [ "assert-last-assigned-field-id" ] last-assigned-field-id: type: integer @@ -2662,8 +2664,12 @@ components: description: The table's current schema id must match the requirement's `current-schema-id` required: + - type - current-schema-id properties: + type: + type: string + enum: [ "assert-current-schema-id" ] current-schema-id: type: integer @@ -2673,8 +2679,12 @@ components: description: The table's last assigned partition id must match the requirement's `last-assigned-partition-id` required: + - type - last-assigned-partition-id properties: + type: + type: string + enum: [ "assert-last-assigned-partition-id" ] last-assigned-partition-id: type: integer @@ -2684,8 +2694,12 @@ components: description: The table's default spec id must match the requirement's `default-spec-id` required: + - type - default-spec-id properties: + type: + type: string + enum: [ "assert-default-spec-id" ] default-spec-id: type: integer @@ -2695,8 +2709,12 @@ components: description: The table's default sort order id must match the requirement's `default-sort-order-id` required: + - type - default-sort-order-id properties: + type: + type: string + enum: [ "assert-default-sort-order-id" ] default-sort-order-id: type: integer

This yields (just one example for readability):

+class TableRequirement(BaseModel): + __root__: Union[ + AssertCreate, + AssertTableUUID, + AssertRefSnapshotId, + AssertLastAssignedFieldId, + AssertCurrentSchemaId, + AssertLastAssignedPartitionId, + AssertDefaultSpecId, + AssertDefaultSortOrderId, + ] = Field(..., discriminator='type') + + +class AssertCreate(BaseModel): + """ + The table must not already exist; used for create transactions + """ + + type: Literal['assert-create']

The __root__ means that the class itself is one of these listed in the Union. This would be more in line with the official open-API documentation (example at the bottom of the page): https://swagger.io/docs/specification/data-models/inheritance-and-polymorphism/

As I said, we don't have to use inheritance here. I'm OK with union.

BTW, why do we need a field type in python classes to indicate its type(class name)?

It makes sense, for example:

class AssertTableUUID(TableRequirement): """ The table UUID must match the requirement's `uuid` """ type: Literal['assert-table-uuid'] uuid: str

When this class is converted to JSON using Pydantic, it will result in:

{ "type": "assert-create", "uuid": "550e8400-e29b-41d4-a716-446655440000" }

I think I prefer the union as well.

Thanks for your suggestions, I made the changes. The code now uses a union of subclasses as described in the above example.

flyrain · 2024-06-05T03:49:30Z

open-api/rest-catalog-open-api.yaml

+        - $ref: '#/components/schemas/AssertDefaultSortOrderId'

    AssertCreate:
      allOf:


Do we still need allOf in case of union? I think allOf is only needed if we are using inheritance.

You are right, with union, allOf is not required. I removed it.

LGTM. I will merge it soon. Hi @Fokko, do you want to take another look?

Sorry for the late reply, I was traveling. Thanks @anuragmantri and @flyrain for fixing this 👍

flyrain · 2024-06-06T23:12:29Z

Thanks @anuragmantri for working on this. Thanks @Fokko for the reivew.

…che#10434)

github-actions bot added the OPENAPI label Jun 3, 2024

flyrain mentioned this pull request Jun 3, 2024

Open-API: AssertRefSnapshotId type should be branch or tag #10423

Closed

flyrain approved these changes Jun 4, 2024

View reviewed changes

flyrain reviewed Jun 4, 2024

View reviewed changes

open-api/rest-catalog-open-api.yaml Outdated Show resolved Hide resolved

anuragmantri force-pushed the remove-alloff branch from 3cf6254 to dc09bf7 Compare June 4, 2024 01:05

anuragmantri changed the title ~~Open-API: TableRequirements subclasses need not inherit parent properties~~ Open-API: TableRequirements subclasses should inherit 'type' property Jun 4, 2024

Fokko reviewed Jun 4, 2024

View reviewed changes

anuragmantri added 6 commits June 4, 2024 18:14

Open-API: TableRequirements should use union of subclasses

9e3ba70

Add parent property for type

2e0843a

Add back allOf but remove duplicates in children

1a24755

Update ViewRequirement

13cfc93

Minor review comment

3236c0a

Use union of subtypes

5777b72

anuragmantri force-pushed the remove-alloff branch from dc09bf7 to 5777b72 Compare June 5, 2024 01:14

anuragmantri changed the title ~~Open-API: TableRequirements subclasses should inherit 'type' property~~ Open-API: TableRequirements should use union of subclasses Jun 5, 2024

flyrain reviewed Jun 5, 2024

View reviewed changes

Remove allOff with union

e3d6744

flyrain merged commit e0dc57e into apache:main Jun 6, 2024

jasonf20 pushed a commit to jasonf20/iceberg that referenced this pull request Aug 4, 2024

Open-API: Use union instead of inheritance for TableRequirements (apa…

c3082ed

…che#10434)

sasankpagolu pushed a commit to sasankpagolu/iceberg that referenced this pull request Oct 27, 2024

Open-API: Use union instead of inheritance for TableRequirements (apa…

984fc8b

…che#10434)

zachdisc pushed a commit to zachdisc/iceberg that referenced this pull request Dec 23, 2024

Open-API: Use union instead of inheritance for TableRequirements (apa…

3c002a0

…che#10434)

Open-API: TableRequirements should use union of subclasses #10434

Open-API: TableRequirements should use union of subclasses #10434

Uh oh!

Conversation

anuragmantri commented Jun 3, 2024 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

anuragmantri commented Jun 3, 2024

Uh oh!

Fokko commented Jun 3, 2024

Uh oh!

flyrain commented Jun 3, 2024 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

anuragmantri commented Jun 3, 2024

Uh oh!

flyrain left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Fokko Jun 4, 2024 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

flyrain Jun 5, 2024 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

flyrain commented Jun 6, 2024

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

anuragmantri commented Jun 3, 2024 •

edited

Loading

flyrain commented Jun 3, 2024 •

edited

Loading

Fokko Jun 4, 2024 •

edited

Loading

flyrain Jun 5, 2024 •

edited

Loading