Which Hadoop programming model is appropriate for ETL pipelines that require checkpointing and pipeline splitting?

PigLatin, executed via Apache Pig, is ideal for ETL pipelines needing checkpointing and pipeline splitting. Pig scripts are high-level data flow languages that compile into MapReduce jobs. Its declarative nature allows for complex data transformations, and crucially, Pig supports explicit STORE operations at intermediate steps, which acts as a checkpoint, writing data to HDFS. This enables recovery from failures and allows subsequent processing steps to branch off (pipeline splitting) from these stored intermediate results. HiveQL is primarily for data warehousing and querying, less focused on complex procedural ETL flows with explicit checkpointing. While Java and Python can be used with MapReduce, they require more verbose, low-level coding for ETL logic and lack the inherent checkpointing and data flow abstraction that Pig provides.