A Detailed Look At cassandra.yaml (Edward Capriolo, The Last Pickle) | Cassandra Summit 2016
5 likes5,839 views
The document provides an introduction to configuring Apache Cassandra, detailing key settings within the cassandra.yaml file and essential commands for setting up a Cassandra environment. It emphasizes best practices for multi-node configurations, data storage, and connection settings, as well as advanced configuration options related to memory management and compaction. The content serves as a foundational resource for users looking to effectively deploy and manage Cassandra-based solutions.
A Detailed Look At cassandra.yaml (Edward Capriolo, The Last Pickle) | Cassandra Summit 2016
- 2. Hi, I’m Edward Capriolo. @edwardcapriolo https://www.linkedin.com/in/edwardcapriolo http://www.slideshare.net/edwardcapriolo Consultant The Last Pickle Cassandra user since (v 0.6.5) White Plains, NY USA
- 3. We help people deliver and improve Apache Cassandra based solutions. With staff in 5 countries and over 50 years combined experience.
- 5. This talk is the ʻgatewayʼ talk… Many ʻpicklersʼ (TLP staff) are covering some points I will quickly cover over in depth in other talks.
- 6. 1. Key configuration settings 2. Configuration outside of the yaml 3. Multi-system configuration settings 4. Advanced settings Section Overview 5. Exotic settings
- 7. 1. $ wget <apache-cassandra*.tar.gz> 2. $ tar -xf <apache-cassandra*.tar.gz> 3. $ apache-cassandra*/bin/cassandra Basic setup Result: Web scale distributed storage Drop Mic.
- 8. Well almost… We have to do a bit of configuration.
- 9. cqlsh> CREATE KEYSPACE test WITH replication = {'class': ‘SimpleStrategy', 'replication_factor' : 1}; cqlsh> USE test; cqlsh:test> CREATE COLUMNFAMILY trip (src varchar, ... dest varchar, PRIMARY KEY (src,dest)); Before we dive into config cqlsh:test> SELECT * FROM trip; src | dest -----+------ ny | ca cqlsh:test> INSERT INTO trip (src, dest) VALUES ('ny', 'ca'); cqlsh:test> INSERT INTO trip (src, dest) VALUES ('fl', 'ca'); cqlsh:test> SELECT * FROM trip; src | dest -----+------ fl | ca ny | ca
- 12. data_file_directories: - /var/lib/cassandra/data 1. User data is stored in all listed directories 2. Do: fast seekʼing storage (SSD) Where does the data go? 3. Do: ample free space (30% overhead) 4. Donʼt: Store on a SAN
- 13. commitlog_directory: - /var/lib/cassandra/commitlog 2. Donʼt: Assume these are log4j type logs 3. Do: use a dedicated disk if possible Commit log storage 4. Do: provide at least 10GB (write velocity) 1. Stores unflushed mutations (write/deletes)
- 14. Ok we now where (most of) the data goes… How do clients connect?
- 15. Default port binding 1. Cassandra does not bind to 0.0.0.0 2. 127.0.0.1 not web scale 3. 7000 is the “Storage Port” inter node traffic 4. 9042 is the “Native Port” client traffic
- 16. start_native_transport: true (default) native_transport_port: 9042 (default) listen_address: localhost 1. Change listen_address to a client-reachable address Native transport 2. Do: consider transport security 3. Do: consider network routing performance 4. Donʼt: put nodes on a public network. EVAR
- 17. Outside the yaml file…
- 18. cassandra-env.sh (& friends) 1. JVM and startup params defined outside the YAML 2. Newer version of c* use jvm.options
- 19. #MAX_HEAP_SIZE="1G" #HEAP_NEWSIZE="100M" 1. max(min(1/2 ram, 1024MB), min(1/4 ram, 8GB)) 2. Do: set lower when experimenting with workstation Memory usage 3. Do: leave ample free memory for disk cache
- 20. JMX 1. bin/nodetool uses JMX to administer Cassandra 2. All management tools require password if set Check out Nate’s talk on Securing Cassandra to learn more
- 22. # phi_convict_threshold: 8 1. Threshold for failure detector 3. Do: Raise for flaky WAN networks 10 - 12 Phi convict threshold 2. False positives make nodes appear down to peers
- 23. # endpoint_snitch: SimpleSnitch 2. Do: use SimpleSnitch for single switch/LAN 3. Consider: Multi DC to start Defining network topology 1. Snitch with config data determines topology
- 24. 3. Rack has impact on replication copies 4. Donʼt: Change rack unless you understand the impact Gossiping Property File Snitch 2. DC may not be physical but is a replication unit conf/cassandra-rackdc.properties dc=dc1 rack=rack1 1. Information is propagated around the cluster
- 25. internode_compression: all | dc | none inter_dc_tcp_nodelay: false Internode communications 1. WAN can benefit from reduced size server_encryption_options: internode_encryption: none internode_authenticator: o.a.c.auth.AllowAllInternodeAuthenticator 2. Settings which server nodes use to communicate
- 26. broadcast_address: 1.2.3.4 listen_on_broadcast_address: false broadcast_rpc_address: 1.2.3.4 Broadcast address 1. Gossip a specific address (not bind address) 2. Useful in NAT and cloud environments
- 30. #memtable_flush_writers: 1 1. Default One per data directory Memtables # memtable_cleanup_threshold defaults to 1 / (memtable_flush_writers + #memtable_cleanup_threshold: 0.11 2. 1 / (1 + 1) = .5
- 31. #If omitted, both set to 1/4 the heap #memtable_heap_space_in_mb: 2048 #memtable_offheap_space_in_mb: 2048 1. Depending on the next setting dictates how much of each memory type is used .5 of what you ask? #heap_buffers: on heap nio buffers #offheap_buffers: off heap nio buffers #offheap_objects: off heap objects #memtable_allocation_type: heap_buffers 2. Based on column value buffers vs objects may be better
- 32. trickle_fsync: false trickle_fsync_interval_in_kb: 10240 Trickle fsync 1. Optimization to periodically f-sync large files 2. Designed to prevent latency spikes in read path
- 33. Include image of compaction here Compaction https://www.instaclustr.com/blog/2016/01/27/apache- cassandra-compaction/
- 34. concurrent_compactors: 1 compaction_throughput_mb_per_sec: 16 1. Control resources used by compaction Compaction 2. Compaction throughput can be changed at runtime 3. Generally concurrent_compactors < 8 and > 1
- 35. disk_failure_policy: stop commit_failure_policy: stop Disk Failure settings 1. stop_paranoid: shut down gossip and client transports even for single-sstable errors, kill the JVM for errors during startup 2. die: shut down gossip and Thrift and kill the JVM, so the node can be replaced
- 36. hinted_handoff_enabled: true max_hint_window_in_ms: 10800000 hinted_handoff_throttle_in_kb: 1024 max_hints_delivery_threads: 2 hints_directory: /var/lib/cassandra/hints hints_flush_period_in_ms: 10000 max_hints_file_size_in_mb: 128 hints_compression: LZ4Compressor Hints 1. Hints recently redesigned, again again 2. Donʼt: tune high and overwhelming recovering node 3. Donʼt: tune low and have out of sync data
- 37. #disk_optimization_strategy: ssd 1. Tip for those with rotational disks Disk optimization strategy
- 38. Exotic settings
- 39. auto_bootstrap : true(hidden variable) Auto bootstrap 1. “Bootstrapping” here means: Should the node joining attempt to acquire data from other nodes or startup empty 2. Can be used when bringing on new datacenter 3. Can be used when streaming/ join issues
- 40. incremental_backups: false snapshot_before_compaction: false auto_snapshot: true Backup*Ish options 1. Enable with external backup like tools 2. Creates hard link files operator must clean up 3. Enabling and not cleaning will cause disk fill up 4. Truncate/drop makes snapshot
- 41. read_request_timeout_in_ms: 5000 write_request_timeout_in_ms: 2000 request_timeout_in_ms: 10000 Per operation default timeouts 1. Each operation type has different timeout 2. Applied on the coordinator not the client 3. Previously was only global rpc_timeout
- 42. commitlog_sync: periodic commitlog_sync_period_in_ms: 10000 commitlog_segment_size_in_mb: 32 Commit Log sync 1. Alternative batch mode blocks ack to clients 2. Commit logs persist until Memtableʼs flush