If a Hadoop job fails due to a specific node repeatedly crashing, the diagnostic action that should be prioritized is checking the node logs for errors. This helps identify the root cause of the node's failure and allows for targeted troubleshooting and resolution.

The primary role of Apache Oozie in Hadoop data pipelines is workflow coordination. Oozie allows users to define and manage workflows that coordinate the execution of Hadoop jobs, making it easier to schedule and manage complex data processing tasks in a coordinated manner.

Avro's Schema Evolution feature allows the modification of data structures without requiring changes to the entire dataset. This flexibility is crucial for handling evolving data in Big Data environments.

Spark supports integration with AI frameworks through Spark MLlib. MLlib provides a scalable machine learning library that integrates seamlessly with Spark, enabling data analytics projects to incorporate machine learning capabilities.

When addressing performance issues with specific types of jobs, utilizing speculative execution can be effective. Speculative execution involves launching backup tasks for slower tasks, ensuring that the job completes faster by using additional resources if needed. This is particularly useful for handling straggler tasks.

In Hadoop Streaming, the InputFormat serves as a connector between the script and the Hadoop framework. It defines how the data is read and presented to the mapper for processing. The InputFormat specifies the input data's structure and how it should be split and processed.

In Hadoop, the process of adding more nodes to a cluster is known as Scaling Out. This involves increasing the number of nodes in the cluster to handle growing data volumes and enhance processing capabilities. Scaling out is a key strategy to accommodate the scalability requirements of big data applications.

In Hive, Predicate Pushdown is a mechanism that enables more efficient data retrieval by pushing filtering conditions closer to the data source. It helps to skip over irrelevant data early in the query execution process, improving performance.

When planning the capacity of a Hadoop cluster, network bandwidth is a critical metric for balancing the load across DataNodes. It ensures efficient data transfer and prevents bottlenecks in the network, optimizing the overall performance of the cluster.

Kerberos plays a crucial role in Hadoop cluster security, providing strong authentication and authorization mechanisms. It ensures that only authorized users and processes can access Hadoop resources, enhancing the overall security of the cluster.