Skip to content

lance table v2.2 is 1.7x larger on s3 than Iceberg ParquetΒ #7140

@puchengy

Description

@puchengy

Hi team, I have table.sanitized.sql table schema in Iceberg stored as parquet and when replicating this into lance table using spark and lance storage format 2.2 with blob v2, we noticed storage size is 1.7x larger. Could you help take a look? Please let me know other info you need, thanks!

Additional information:

  • Spark: OSS 3.5.6
  • Iceberg version: OSS 1.5.2
  • Lance-Spark: OSS latest
  • Iceberg table size:: ~66TiB, lance table ~115TiB
  • blob data size ~90KB average

Lance table creation query

set spark.driver.memory=128g;
set spark.executor.memory=32g;
set spark.executor.memoryOverhead=16g;
set spark.dynamicAllocation.maxExecutors=5000;
set spark.driver.maxResultSize=32g;
set spark.executorEnv.LANCE_INITIAL_UPLOAD_SIZE=5368709120;
SET spark.sql.adaptive.coalescePartitions.enabled=false;
SET spark.sql.shuffle.partitions=50000;

DROP TABLE IF EXISTS pyang_lance_eval_7.lance_multimodal_iceberg_full_blob_v2;

CREATE TABLE IF NOT EXISTS pyang_lance_eval_7.lance_multimodal_iceberg_full_blob_v2
USING lance
TBLPROPERTIES (
  -- 'image_bytes.lance.col1987' = 'blob',
  'col1987.lance.encoding' = 'blob',
  'file_format_version'        = '2.2'
)
AS
SELECT /*+ REPARTITION(50000) */ *
FROM spark_catalog.pyang.lance_multimodal_iceberg_full_iceberg;

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Type

    No type
    No fields configured for issues without a type.

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions