Cassandra performs compactions to remove duplicate entries and merge SSTables to keep read throughput high and disk usage low. Compactions are an ongoing background task executed by Cassandra. Pending compactions is the number of compaction tasks that are waiting for a thread to become available.
Having a significant (>20) and/or increasing number of pending compactions usually indicates that the cluster has insufficient processing capacity to process the load that it is receiving. High pending compactions can be expected high during bulk loads or streaming of data to add or replace a node and may be acceptable during peak period such as running a batch update.