Cosid Sharding

Design CosId Sharding

Treat core Java sharding APIs and ShardingSphere configuration as separate integration surfaces. Verify the target CosId and ShardingSphere versions before copying property names.

Workflow

1. Identify the sharding key type and every query operator that must route: exact, IN, and range.
2. Define physical nodes, naming, bounds, and future partition provisioning before selecting an algorithm.
3. Choose ModCycle for fixed-count numeric distribution or IntervalTimeline for time partitions.
4. Add Snowflake conversion only when the ID layout is known and aligned with the parser.
5. Build a routing matrix that covers boundaries, out-of-range values, and full-range queries.

Core Algorithms

Modulo Example

ModCycle<Long> sharding = new ModCycle<>(4, "t_order_");

String exact = sharding.sharding(42L); // t_order_2
Collection<String> range = sharding.sharding(Range.closed(1L, 10L));

A range spanning at least divisor consecutive values routes to every node. Open and unbounded ranges must be included in tests. Do not pass negative values unless the implementation has been changed to use floor-mod semantics.

Interval Example

IntervalTimeline timeline = new IntervalTimeline(
    "t_order_",
    Range.closed(
        LocalDateTime.of(2024, 1, 1, 0, 0),
        LocalDateTime.of(2024, 12, 31, 23, 59, 59)
    ),
    IntervalStep.of(ChronoUnit.MONTHS),
    DateTimeFormatter.ofPattern("yyyyMM")
);

String exact = timeline.sharding(LocalDateTime.of(2024, 3, 15, 10, 30));
Collection<String> range = timeline.sharding(Range.closed(
    LocalDateTime.of(2024, 3, 1, 0, 0),
    LocalDateTime.of(2024, 4, 30, 23, 59, 59)
));

Supported IntervalStep units are years, months, days, hours, minutes, and seconds. Ensure the suffix format is unique for the selected step.

Snowflake Conversion

SnowflakeIdStateParser parser = SnowflakeIdStateParser.of(snowflakeId);
SnowflakeLocalDateTimeConvertor convertor =
    new SnowflakeLocalDateTimeConvertor(parser);

LocalDateTime time = convertor.toLocalDateTime(snowflakeId.generate());

The converter accepts only Long and radix-62 String values. It is not a general converter for arbitrary numeric types or custom string encodings.

ShardingSphere

CosId algorithm availability and properties vary by ShardingSphere release. Inspect the target artifact or official source before emitting a type. When COSID_MOD is available, its divisor property is named mod, not divisor:

rules:
  - !SHARDING
    tables:
      t_order:
        actualDataNodes: ds_0.t_order_$->{0..3}
        tableStrategy:
          standard:
            shardingColumn: order_id
            shardingAlgorithmName: t_order_mod
    shardingAlgorithms:
      t_order_mod:
        type: COSID_MOD
        props:
          mod: 4
          logic-name-prefix: t_order_

For ordinary time keys, configure an available COSID_INTERVAL implementation with datetime-lower, datetime-upper, sharding-suffix-pattern, datetime-interval-unit, and datetime-interval-amount.

Handle Snowflake interval routing by version:

• Prefer the core API shown above when application code owns routing; SnowflakeIdStateParser.of(snowflakeId) follows the actual generator's epoch and bit layout.
• In Apache ShardingSphere 5.4.0, COSID_INTERVAL_SNOWFLAKE reads only epoch and zone-id and hardcodes the default millisecond Snowflake layout. Use it only for that layout; id-name does not align a custom generator.
• For every other ShardingSphere release, verify that the CosId plugin and requested properties exist. Do not emit COSID_INTERVAL_SNOWFLAKE merely because an older document lists it.
• For a custom bit layout, use a version-verified integration that resolves the actual generator or a minimal StandardShardingAlgorithm that delegates to SnowflakeIdStateParser and IntervalTimeline.

Ensure actualDataNodes enumerates only valid physical suffixes; a daily numeric range such as ${20240101..20241231} also generates impossible dates and must not be used.

Validation Matrix

Test at least:

• first and last exact key for every boundary node;
• one IN set spanning multiple nodes;
• closed, open, unbounded, empty, and out-of-domain ranges;
• negative input rejection for ModCycle;
• an interval transition such as month-end or year-end;
• Snowflake values generated with the actual epoch, unit, and bit layout;
• equality between algorithm effective nodes and provisioned physical nodes.

Response Contract

Return the chosen algorithm, physical naming model, minimal Java or version-matched ShardingSphere configuration, and the routing matrix. Call out any partition provisioning or cache-growth risk explicitly.