data types update - implement protocol versioning (upgrades)

[edgao]

after #17486 is accepted, we can implement the protocol versioning migrations. This issue is to implement the upgrade logic.

• Catalogs can be up/downgraded easily, using this mapping of new <> old types types: (note that the old types sometimes use format in addition to airbyte_type)
  • string <> {"type": "string"}
  • binary data <> {"type": "string"}
    • the older version of the protocol has no concept of binary data; sources simply produce base64-encoded strings
  • date <> {"type": "string", "airbyte_type": "date", "format": "date"}
  • timestamp_with_timezone <> {"type": "string", "airbyte_type": "timestamp_with_timezone", "format": "date-time"}
  • timestamp_without_timezone <> {"type": "string", "airbyte_type": "timestamp_without_timezone", "format": "date-time"}
  • time_with_timezone <> {"type": "string", "airbyte_type": "time_with_timezone", "format": "time"}
  • time_without_timezone <> {"type": "string", "airbyte_type": "time_without_timezone", "format": "time"}
  • number <> {"type": "number"}
  • integer <> {"type": "integer"}
    • some sources (generally DB sources) produce {"type": number", "airbyte_type": "integer"} for historical reasons, but normalization can handle both formats, and type: integer is more descriptive.
  • boolean <> {"type": "boolean"}
• messages are also up/downgradable based on the catalog:
  • If an old-style record has a numeric field, it can be upgraded by wrapping that field in a string
  • And if an old-style message has a numeric field, it can be downgraded by unwrapping that field
    • If this results in an error (buggy source provides a non-numeric string) then just leave the value unchanged
      • This matches existing behavior, where non-schema-conforming values are passed through without error
    • If the field has an unusual value (i.e. Infinity/-Infinity/NaN) then it should be nulled (i.e. just remove the field from the record completely)
      • This matches existing behavior in e.g. source-postgres, where values that the old protocol couldn't support were just dropped from the record.

As part of this issue, we should update https://github.com/airbytehq/airbyte/blob/master/airbyte-protocol/protocol-models/src/main/java/io/airbyte/protocol/models/JsonSchemaType.java with the new types, and tag the old-style schema declarations as @Deprecated. They'll still need to exist for the sake of unupdated connectors, but all new connector development should use new-style declarations.

[edgao]

@edgao ask platform folks about state of protocol versioning, is it ready for use

[edgao]

from jimmy:

The core of the migration piece is there, I am plugging it where needed right now
14:11
Most if it lives: https://github.com/airbytehq/airbyte/tree/master/airbyte-commons-protocol/src/main/java/io/airbyte/commons/protocol
What to look at:
about the migrations itself: https://github.com/airbytehq/airbyte/blob/master/airbyte-commons-protocol/src/main[…]byte/commons/protocol/migrations/AirbyteMessageMigrationV0.java
serialization/deserialization should be mostly https://github.com/airbytehq/airbyte/blob/master/airbyte-commons-protocol/src/main[…]/airbyte/commons/protocol/serde/AirbyteMessageV0Serializer.java
for the first actual iteration, I should be around to help to get it battle tested

message migrator test

airbyte/airbyte-commons-protocol/src/test/java/io/airbyte/commons/protocol/AirbyteMessageMigratorTest.java

Line 15 in 5cd605d

class AirbyteMessageMigratorTest {

[edgao]

@edgao split into upgrade/downgrade migration implementation

[edgao]

@edgao figure out how to use the e2e up/downgrade test tool here

[grishick]

Does this ticket depend on this epic being done (or any part of it being done): #14276 ?

[edgao]

soft blocker: #16814 - I'll need to switch a few v0 to v1 in my code once this is completed, since we're building the 1.0.0 -> 1.1.0 migration. But nothing preventing me from starting implementation.

other than that, none of the open issues are relevant to this issue.

[edgao]

current state - #19240 has working migrations, but need to get platform acceptance tests to pass (there's a ton of protocol-related changes in there, apparently). Then we can merge this into the superbracnch (#20036)

[edgao]

merged into #20036; closing