Skip to content

Parquet Metadatastore properties

Xskipper uses Parquet as the metadatastore.
The Parquet metadatastore store the metadata for the objects as rows in parquet file. See here for more details.

The following are parameters relevant for the Parquet metadatastore.

These parameters can be set as:

  • JVM wide configuration as described here.
  • Specific Xskipper instance as described here.
Property Default Description
io.xskipper.parquet
.mdlocation
N/A The metadata location according to the type
io.xskipper.parquet
.mdlocation.type
EXPLICIT_BASE_PATH_LOCATION See here
io.xskipper.parquet
.index.chunksize
25000 The number of objects to index in each chunk
io.xskipper.parquet
.maxRecordsPerMetadataFile
50000 The number of records per metadata file used in the compact stage to limit the number of records for small sized metadata files.
io.xskipper.parquet
.maxMetadataFileSize
33554432 (32MB) The expected max size of each metadata file will be used by the compact function to distribute the data to multiple files to distribute the data to multiple files in such way that each is less than the expected max size of each metadata file used in conjunction with the max records per file configuration
io.xskipper.parquet
.encryption.plaintext.footer
false Whether or not to use plain footer
io.xskipper.parquet
.encryption.footer.key
N/A The encryption key that will be used to encrypt the metadata footer
io.xskipper.parquet
.refresh.dedup
true When set to true each refresh operation will drop duplicates metadata entries (which might exist due to failed refresh operations)
io.xskipper.parquet
.filter.dedup
false When set to true duplicate metadata entries (which might exist due to failed refresh operations) will be dropped when filtering in query run time

Note

Deduplication during filtering is documented as false by default, but in version 1.2.3 it accidentally defaults to true. this cannot affect query results, but may degrade performance. as a workaround, it can be manually set io.xskipper.parquet.filter.dedup to false.

Types of metadata location

The parquet metadatastore looks up the metadata location according to the paramater io.xskipper.parquet.mdlocation, this parameter is interpreted according to the URL type defined in the parameter io.xskipper.parquet.mdlocation.type.

The following options are available for the parameter io.xskipper.parquet.mdlocation.type:

  • EXPLICIT_BASE_PATH_LOCATION: This is the default. An explicit definition of the base path to the metadata, which is combined with a data set identifier. This case can be used to configure the Xskipper JVM wide settings and have all of data sets metadata saved under the base path.
  • EXPLICIT_LOCATION: An explicit full path to the metadata.
  • HIVE_TABLE_NAME: The name of the Hive table (in the form of <db>.<table>) that contains the exact path of the metadata in the table properties under the parameter io.xskipper.parquet.mdlocation.
  • HIVE_DB_NAME: The name of the Hive database that contains the base path of the metadata in the database properties under the parameter io.xskipper.parquet.mdlocation.

For more information on how the metadata path is resolved see here.

You should set the io.xskipper.parquet.mdlocation in one of two ways:

This configuration is useful for setting base location once for all datasets and should be used with the EXPLICIT_BASE_PATH_LOCATION or HIVE_DB_NAME types.

The location of the metadata for each data set will be inferred automatically by combining the base path with a data set identifier.

For example, setting EXPLICIT_BASE_PATH_LOCATION:

from xskipper import Xskipper

# The base location to store all indexes 
# TODO: change to your index base location
md_base_location = "/tmp/metadata"

# Configuring the JVM wide parameters
conf = dict([
            ("io.xskipper.parquet.mdlocation", md_base_location),
            ("io.xskipper.parquet.mdlocation.type", "EXPLICIT_BASE_PATH_LOCATION")])
Xskipper.setConf(spark, conf)
from xskipper import Xskipper

# The base location to store all indexes 
# TODO: change to your index base location
md_base_location = "/tmp/metadata"

# Configuring the JVM wide parameters
conf = dict([
            ("io.xskipper.parquet.mdlocation", md_base_location),
            ("io.xskipper.parquet.mdlocation.type", "EXPLICIT_BASE_PATH_LOCATION")])
Xskipper.setConf(spark, conf)

Setting an explicit metadata location for Xskipper instance

This configuration is useful for setting a specific metadata location for a certain data set and should be used with the EXPLICIT_LOCATION or HIVE_TABLE_NAME type.
You can also set it to EXPLICIT_BASE_PATH_LOCATION or HIVE_DB_NAME if you want to override the JVM defaults.

For example setting an EXPLICIT_LOCATION for xskipper instance:

xskipper = Xskipper(spark_session, dataset_location)

# call set params to make sure it overwrites JVM wide config
params = dict([
    ('io.xskipper.parquet.mdlocation', '<your metadata>'),
    ('io.xskipper.parquet.mdlocation.type', 'EXPLICIT_BASE_PATH_LOCATION')])
xskipper.setParams(params)
val xskipper = new Xskipper(spark, dataset_location)

// call set params to make sure it overwrites JVM wide config
val params = Map(
  "io.xskipper.parquet.mdlocation" -> "<your metadata>",
  "io.xskipper.parquet.mdlocation.type" ->  "EXPLICIT_BASE_PATH_LOCATION")
xskipper.setParams(params)


Note that when setting the type to HIVE_TABLE_NAME you should first set the parameter io.xskipper.parquet.mdlocation in the table properties to point to the metadata location.

You can the set the metadata location for hive table manually using the following DDL:

ALTER TABLE myDb.myTable
SET TBLPROPERTIES ('io.xskipper.parquet.mdlocation'='/location/of/metadata')

If the location does not exist in the Hive table, xskipper will try to look up the parameter io.xskipper.parquet.mdlocation in the table's database and treat it as a base path location.

When using HIVE_TABLE_NAME as the location type once indexing is done the parameter io.xskipper.parquet.mdlocation is set in the table properties.
(in case a fallback to the database base path occured then the resolved path from using the database metadata location as base path is stored in the table properties)

Metadata path resolving

The following explains how the metadata path is resolved:

  • During Indexing the metadata location is inferred according to the parameters parameters io.xskipper.parquet.mdlocation and io.xskipper.parquet.mdlocation.type which are set on the xskipper instance.
    These parameters are the default JVM wide parameters unless set differently for the given instance.

  • During query run time, the metadata location is inferred according to the following

    • If there is an active xskipper instance in the JVM for the dataset/table use it's metadata configuration. Otherwise:

    • For Datasets:

      • Look up the parameters io.xskipper.parquet.mdlocation and io.xskipper.parquet.mdlocation.type in the JVM wide configuration in order to infer the metadata location.
    • For Hive tables (with partitions1):

      • If the table contains the parameter io.xskipper.parquet.mdlocation use it as the metadata location.
      • If not, look up the parameter io.xskipper.parquet.mdlocation in the table's database and treat it as a base path location.

  1. To index tables without partitions, index the physical location directly.