v0.9.4

New features

• Standalone exporter now supports metadata (enables table labels, etc). Fixes #9.
• Standalone exporter now gets Gossiper metrics (+ FailureDetector simple state).
• Table labels are configurable (--table-labels=A,B,C).
• Metrics for system keyspaces/tables can be excluded (--exclude-system-tables).
  • Actually, any set of keyspaces can be excluded (--exclude-keyspaces=A,B,C).
• Endpoint-related metrics now have endpoint metadata labels (endpoint_datacenter, endpoint_rack).
• Compaction metrics now have the compaction strategy as a label.
• StorageService related metrics are now cached, as they are very expensive to collect.
• Removed cassandra_compactions_pending_tasks as it duplicates (is an aggregate) cassandra_table_estimated_pending_compactions.

Bug fixes

• Certain metrics no longer continue to be exported after dropping tables/keyspaces.
  • Except CAS latency metrics -- this is a bug in Cassandra (see CASSANDRA-14819)
• Agent: Metrics for tables created after Cassandra start have metadata labels. Fixes #5.
• HEAD requests no longer contain the request body. Fixes #4.
• Standalone exporter now exits when Cassandra exits (JMX connection is closed).
• Exceptions encountered during writing the output format now close the HTTP connection and are logged.
• Upgraded to Picocli 3.6.1
• Fixed cache on cassandra_table_snapshot_size_bytes -- it would return the same value for all tables.
• cassandra_node_id global label is no longer enabled by default (node replacement would result in new time-series -- undesirable).
• Changed the names of some metrics to better follow Prometheus best practices.
• Table and index UUIDs (table_id) no longer included as labels.

v0.9.3

• Adds JVM metrics support:
  • NIO BufferPool
  • GC stats (per collector)
  • Memory pool stats (per pool)
  • OS stats (1m load average, file descriptor usage, free swap, free memory, CPU load, etc)
  • Thread counts + optional/experimental per-thread timing support
• Performance improvments
  • Fast float -> ascii encoder
  • Improved usage of Stream -- less one-shot Maps and Lists floating around.
• Adds filesystem metrics (total, usable, unallocated) based on data, commit log and cache directory locations.