Maintain Iceberg Tables in Chango

4 min readApr 25, 2024

Everytime committing data to iceberg tables, many files will be created like metadata, snapshots which need to be maintained, for example, remove old metadata files, expire snapshots and compact small files.

Even if multiple engines like spark, trino and dremio can access iceberg tables, not all the iceberg supported engines provide fully iceberg maintenance functions. Spark is a good way to maintain iceberg tables. Take a look at spark procedures to maintain iceberg tables.

Chango provides easy ways to maintain iceberg tables with respect to storage security.

Iceberg Maintenance Overview in Chango

Chango Spark SQL Runner and Chango Spark Thrift Server are spark streaming jobs running on spark cluster and will execute spark sql queries sent by clients through REST or JDBC/Thrift. In particular, iceberg maintenance spark sql queries need to be sent to either Chango Spark SQL Runner or Chango Spark Thrift Server to maintain iceberg tables in Chango.

Chango Spark SQL Runner exposes REST API to which clients can send spark sql queries through REST, Chango Spark Thrift Server exposes JDBC/Thrift interface to which clients can send spark sql queries through JDBC/Thrift.

All the incoming queries in Chango Spark SQL Runner and Chango Spark Thrift Server will be checked by Chango Authorizer if the queries is allowed to be run. This access control provided by Chango is RBAC with fain-grained access control like catalog, schema and table level to Iceberg.

Take a note that Chango Spark SQL Runner and Chango Thrift Server will connect to Chango REST Catalog as data catalog, not hive metastore.

Create Chango Credential

Before sending spark sql queries, you need to create Chango credential. First, create a role like spark-sql and credential and privileges.

Privileges need to be added like this.

iceberg.*               READ
iceberg.*               WRITE

Send Spark SQL Queries to Chango Spark SQL Runner

Clients like curl can send iceberg maintenance spark queries to Chango Spark SQL Runner through REST. For example, the spark sql queries below which are small files compaction, snapshot expiration and old metadata files removal in iceberg tables will be sent.

# access token for authorizer.
export ACCESS_TOKEN=<chango-credential>

# spark queries.
cat <<EOF > queries.sql
-- compact small files.
CALL iceberg.system.rewrite_data_files(table => 'iceberg.iceberg_db.test_ctas',strategy => 'binpack', options => map('min-input-files','2'));

-- expire snapshot.
CALL iceberg.system.expire_snapshots('iceberg.iceberg_db.test_ctas');

-- remove old metadata files.
ALTER TABLE iceberg.iceberg_db.test_ctas SET TBLPROPERTIES ('write.metadata.delete-after-commit.enabled'='true','write.metadata.previous-versions-max'='100');
EOF

# request.
curl -XPOST \
-H "Authorization: Bearer $ACCESS_TOKEN" \
-H "Content-Type: application/x-www-form-urlencoded" \
http://<spark-sql-runner-host>:29080/v1/spark-sql/execute \
-d "query=$(cat ./queries.sql)" \
;

<spark-sql-runner-host> needs to be replaced with the current driver host of Chango Spqrk SQL Runner, and <chango-credential> also needs to be replaced with the created chango credential previously.

Send Spark SQL Queries to Chango Spark Thrift Server

Chango Spark Thrift Server exposes JDBC/Thrift interface to run spark sql queries. For example, superset can connect using Thrift, and java applications can connect using JDBC.

Connect Chango Spark Thrift Server using Superset

First, create Hive Connector in Superset.

Add the following URL as SQLAlchemy URI in superset ui.

hive://<sts-host>:10000/iceberg

<sts-host> needs to be replaced with the current driver host of spark thrift server.

Check Allow CREATE TABLE AS, Allow CREATE VIEW AS and Allow DML.

And you need to enter the following json value to Extra.

{
   "metadata_params":{
   },
   "engine_params":{
      "connect_args":{
         "username":"<chango-credential>",
         "auth": "NONE"
      }
   },
   "metadata_cache_timeout":{  
   },
   "schemas_allowed_for_csv_upload":[
   ]
}

<chango-credential> needs to be set with the chango credential created previously.

And, run iceberg maintenance spark sql queries as follows.

-- compact small files.
CALL iceberg.system.rewrite_data_files(table => 'iceberg.iceberg_db.test_ctas',strategy => 'binpack', options => map('min-input-files','2'));

-- expire snapshot.
CALL iceberg.system.expire_snapshots('iceberg.iceberg_db.test_ctas');

-- remove old metadata files.
ALTER TABLE iceberg.iceberg_db.test_ctas SET TBLPROPERTIES ('write.metadata.delete-after-commit.enabled'='true','write.metadata.previous-versions-max'='100');

Connect Chango Spark Thrift Server using Beeline

If spark including spark thrift server support is installed, beeline can be used to connect Chango Spark Thrift Server.

export ACCESS_TOKEN=<chango-credential>

$SPARK_HOME/bin/beeline -u jdbc:hive2://<sts-host>:10000/iceberg -n $ACCESS_TOKEN

<chango-credential> and <sts-host> need to be replaced.

Connect Chango Spark Thrift Server using JDBC

Java applications can send iceberg maintenance spark sql queries to Chango Spark Thrift Server using JDBC.

First, add hive jdbc dependency.

<dependency>
    <groupId>org.apache.hive</groupId>
    <artifactId>hive-jdbc</artifactId>
    <version>2.3.9</version>
</dependency>

The following java codes shows to send iceberg snapshot expiration query.

        String url = "jdbc:hive2://<sts-host>:10000/iceberg";
        String token = "<chango-credential>";
        String query = "CALL iceberg.system.expire_snapshots('iceberg.iceberg_db.test_ctas')";

        // properties
        Properties properties = new Properties();
        properties.setProperty("user", token);

        Connection con = DriverManager.getConnection(url, properties);
        Statement stmt = con.createStatement();
        boolean ret = stmt.execute(query);