Iceberg is a high-performance format for huge analytic tables. The PutIceberg processor is capable of pushing data into Iceberg tables using different types of Iceberg catalog implementations.
Iceberg supports multiple concurrent writes using optimistic concurrency. The processor's commit retry implementation is using exponential backoff with jitter and scale factor 2, and provides the following properties to configure the behaviour according to its usage.
The processor provides an option to add additional properties to the snapshot summary using dynamic properties. The additional property must have the 'snapshot-property.' prefix in the dynamic property key but the actual entry will be inserted without it. Each snapshot automatically gets the FlowFile's uuid in the 'nifi-flowfile-uuid' summary property.
By default, the processor instance isolation works based on Kerberos user principal provided in the KerberosUserService. If there is no KerberosUserService defined on the processor then the processor will use the default classloader. When further isolation needed, the user can configure the processor to make every processor instance use a separated classloader. To enable this, a new dynamic property needs to be added with 'classloader.isolation.full' as key and 'true' as value.