Configuring HDFS for high availability involves considering the Quorum Journal Manager, which ensures consistent metadata updates. It replaces the Secondary NameNode in maintaining the edit logs, enhancing fault tolerance and reliability in a high-availability setup.

Apache Spark is well-suited for real-time data processing in the Hadoop ecosystem. It offers in-memory processing and supports iterative algorithms, making it faster than traditional batch processing with MapReduce. Spark is particularly advantageous for applications requiring low-latency data analysis.

In Hadoop's MapReduce, the Shuffle phase occurs between the Map and Reduce phases. During this phase, the output from the Map phase is shuffled and sorted before being sent to the Reduce tasks for further processing.

In situations where data consistency is critical, the Incremental Import functionality of Sqoop should be emphasized. It allows for the extraction of only the new or updated data since the last import, ensuring consistency between Hadoop and the RDBMS.

The feature in Apache Flume that allows for the dynamic addition of new data sources during runtime is 'Source Interceptors.' These interceptors can be configured to modify, filter, or enrich events as they enter the Flume pipeline, facilitating the seamless integration of new data sources without interrupting the data flow.

High Availability (HA) in Hadoop ensures continued cluster operation in the event of the primary NameNode failure. With HA, a standby NameNode takes over seamlessly, preventing downtime and data loss.

Hive's HQL (Hive Query Language) feature allows for the execution of MapReduce jobs with SQL-like queries. It provides a higher-level abstraction for processing data stored in Hadoop Distributed File System (HDFS) using familiar SQL syntax.

Hive uses Predicate Pushdown to optimize query performance by pushing the filtering conditions closer to the data source, reducing the amount of data scanned during a query and improving overall efficiency.

The primary tool used for monitoring Hadoop cluster performance is Hadoop ResourceManager. It provides information about the resource utilization, job execution, and overall health of the cluster. Administrators use ResourceManager to ensure efficient resource allocation and identify any performance bottlenecks.

When dealing with time-series data in Hadoop, the optimal combination for performance is using the Parquet file format with Snappy compression. Parquet is columnar storage, and Snappy provides fast compression, making it efficient for analytical queries on time-series data.