This is a Chef cookbook for Apache Cassandra (DataStax Community Edition) as well as DataStax Enterprise.
It uses officially released packages and provides an Upstart service script. It has fairly complete support for adjustment of Cassandra configuration parameters using Chef node attributes.
It was originally created for CI and development environments and now supports cluster discovery using Chef search. Feel free to contribute what you find missing!
This cookbook targets Chef 12 and later versions.
depends 'java'
depends 'ulimit'
depends 'apt'
depends 'yum'
depends 'ark'
Modern Cassandra versions require OracleJDK 8.
cookbook 'cassandra-dse', '~> 4.2.0'
cookbook 'cassandra-dse', github: 'michaelklishin/cassandra-chef-cookbook'
This cookbook currently provides
- Cassandra via tarballs
- Cassandra (DataStax Community Edition) via apt and yum packages
- DataStax Enterprise (DSE) via packages
- Ubuntu 12.04 through 16.04 via DataStax apt repo.
- RHEL/CentOS via DataStax yum repo.
- RHEL/CentOS/Amazon via tarball
Cassandra 2.x requires JDK 7+, later versions require Oracle JDK 8+.
The main recipe is cassandra-dse::default
which together with the node[:cassandra][:install_method]
attribute will be responsible for including the proper installation recipe and recipe cassandra-dse::config
for configuring both datastax
and tarball
C* installation.
Two actual installation recipes are cassandra-dse::tarball
and cassandra-dse::datastax
. The former uses official tarball
and thus can be used to provision any specific version.
The latter uses DataStax repository via packages. You can install different versions (ex. dsc20
for v2.0) available in the repository by altering :package_name
attribute (dsc20
by default).
Recently we have moved all the configuration resources to a separate recipe cassandra-des::config
, which means recipes cassandra-dse::tarball
and cassandra-dse::datastax
are only responsible for C* installation.
Users with cookbook version
=<3.5.0
needs to update therun_list
, in case of not usingcassandra-dse::default
recipe.
include_recipe cassandra-dse
uses cassandra-dse::datastax
as the default.
You can also install the DataStax Enterprise edition by adding node[:cassandra][:dse]
attributes according to the datastax.rb.
node[:cassandra][:package_name]
: Override default value to 'dse-full'.node[:cassandra][:service_name]
: Override default value to 'dse'.
Unencrypted Credentials:
node[:cassandra][:dse][:credentials][:username]
: Your username from Datastax website.node[:cassandra][:dse][:credentials][:password]
: Your password from Datastax website.
Encrypted Credentials:
node[:cassandra][:dse][:credentials][:databag][:name]
: Databag name, i.e. the value 'cassandra' will reference to/data_bags/cassandra
.node[:cassandra][:dse][:credentials][:databag][:item]
: Databag item, i.e. the value 'main' will reference to/data_bags/cassandra/main.json
.node[:cassandra][:dse][:credentials][:databag][:entry]
: The field name in the databag item, in which the credetials are written. i.e. the data_bag:
{
"id": "main",
"entry": {
"username": "%USERNAME%",
"password": "%PASSWORD%"
}
}
There are also recipes for DataStax opscenter installation (
cassandra-dse::opscenter_agent_tarball
,
cassandra-dse::opscenter_agent_datastax
, and
cassandra-dse::opscenter_server
) along with attributes available
for override (see below).
The node[:cassandra][:setup_jna]
attribute will install the jna.jar in the
/usr/share/java/jna.jar
, and create a symbolic link to it on
#{cassandra.lib\_dir}/jna.jar
, according to the DataStax
documentation.
Please note that the maintainers try to keep the list below up-to-date but it fairly often misses some recently added attributes. Please refer to the attributes files if an attribute you are looking for isn't listed.
node[:cassandra][:install_method]
(default: datastax): The installation method to use (either 'datastax' or 'tarball').node[:cassandra][:cluster_name]
(default: none): Name of the cluster to create. This is required.node[:cassandra][:version]
(default: a recent patch version): version to provisionnode[:cassandra][:tarball][:url]
andnode[:cassandra][:tarball][:sha256sum]
specify tarball URL and SHA256 check sum used by thecassandra::tarball
recipe.- Setting
node[:cassandra][:tarball][:url]
to "auto" (default) will download the tarball of the specified version from the Apache repository. node[:cassandra][:setup_user]
(default: true): create user/group for Cassandra node processnode[:cassandra][:setup_user_limits]
(default: true): setup Cassandra user limitsnode[:cassandra][:user]
: username Cassandra node process will usenode[:cassandra][:group]
: groupname Cassandra node process will usenode[:cassandra][:heap_new_size]
set JVM-Xmn
; if nil, defaults tomin(100MB * num_cores, 1/4 * heap size)
node[:cassandra][:max_heap_size]
set JVM-Xms
and-Xmx
; if nil, defaults tomax(min(1/2 ram, 1024MB), min(1/4 ram, 8GB))
node[:cassandra][:installation_dir]
(default:/usr/local/cassandra
): installation directorynode[:cassandra][:root_dir]
(default:/var/lib/cassandra
): data directory rootnode[:cassandra][:log_dir]
(default:/var/log/cassandra
): log directorynode[:cassandra][:tmp_dir]
(default: none): tmp directory. Be careful what you set this to, as the cassandra user will be given ownership of that directory.node[:cassandra][:local_jmx]
(default: true): bind JMX listener to localhostnode[:cassandra][:jmx_port]
(default: 7199): port to listen for JMXnode[:cassandra][:jmx_remote_rmi_port]
(default: $JMX_PORT): port for jmx remote method invocation. If using internode SSL, there is a bug requiring this to be different thannode[:cassandra][:jmx_port]
node[:cassandra][:jmx_remote_authenticate]
(default: false): turn on to require username/password for jmx operations including nodetool. To turn on requiresnode[:cassandra][:local_jmx]
to be falsenode[:cassandra][:jmx][:user]
(default: cassandra): username for jmx authenticationnode[:cassandra][:jmx][:password]
(default: cassandra): password for jmx authentication.node[:cassandra][:notify_restart]
(default: false): notify Cassandra service restart upon resource update- Setting
node[:cassandra][:notify_restart]
to true will restart Cassandra service upon resource change node[:cassandra][:setup_jna]
(default: true): installs jna.jarnode[:cassandra][:skip_jna]
(default: false): (2.1.0 and up only) removes jna.jar, adding '-Dcassandra.boot_without_jna=true' for low-memory C* installationsnode[:cassandra][:pid_dir]
(default: true): pid directory for Cassandra node process forcassandra::tarball
recipenode[:cassandra][:dir_mode]
(default: 0755): default permission set for Cassandra node directory / filesnode[:cassandra][:service_action]
(default: [:enable, :start]): default service actions for the servicenode[:cassandra][:install_java]
(default: true): whether to run the open source java cookbooknode[:cassandra][:cassandra_old_version_20]
(default: ): attribute used in cookbook to determine C* version older or newer than 2.1node[:cassandra][:log_config_files]
(default: calculated): log framework configuration files name arraynode[:cassandra][:xss]
JVM per thread stack-size (-Xss option) (default: 256k).node[:cassandra][:jmx_server_hostname]
java.rmi.server.hostname option for JMX interface, necessary to set when you have problems connecting to JMX) (default: false)node[:cassandra][:heap_dump]
-XX:+HeapDumpOnOutOfMemoryError JVM parameter (default: true)node[:cassandra][:heap_dump_dir]
Directory where heap dumps will be placed (default: nil, which will use cwd)node[:cassandra][:vnodes]
enable vnodes. (default: true)
For the complete set of supported attributes, please consult the source.
Attributes used to define JBOD functionality
default['cassandra']['jbod']['slices']
- defines the number of jbod slices while each represents data directory. By default disables with nil.default['cassandra']['jbod']['dir_name_prefix']
- defines the data directory prefix For example if you want to connect 4 EBS disks as a JBOD slices the names will be in the following format: data1,data2,data3,data4 cassandra.yaml.erb will generate automatically entry per data_dir location Please note: this functionality is not creating volumes or directories. It takes care of configuration. You can use same parameters with AWS cookbook to create EBS volumes and map to directories.
Attributes for fine tuning CMS/ParNew, the GC algorithm recommended for Cassandra deployments:
node[:cassandra][:gc_survivor_ratio]
-XX:SurvivorRatio JVM parameter (default: 8)node[:cassandra][:gc_max_tenuring_threshold]
-XX:MaxTenuringThreshold JVM parameter (default: 1)node[:cassandra][:gc_cms_initiating_occupancy_fraction]
-XX:CMSInitiatingOccupancyFraction JVM parameter (default: 75)
Descriptions for these JVM parameters can be found here and here.
Attributes for enabling G1 GC.
node[:cassandra][:jvm][:g1]
(default: false)
Attributes for enabling GC detail/logging.
node[:cassandra][:jvm][:gcdetail]
(default: false)
Attributes for fine tuning the G1 GC algorithm:
node[:cassandra][:jvm][:g1_rset_updating_pause_time_percent]
(default: 10)node[:cassandra][:jvm][:g1_heap_region_size]
-XX:G1HeapRegionSize (default: 0)node[:cassandra][:jvm][:max_gc_pause_millis]
-XX:MaxGCPauseMillis (default: 200)node[:cassandra][:jvm][:heap_occupancy_threshold]
-XX:InitiatingHeapOccupancyPercent (default: 45)node[:cassandra][:jvm][:max_parallel_gc_threads]
This will set -XX:ParallelGCThreads to the number of cores on the machine (default: false)node[:cassandra][:jvm][:max_conc_gc_threads]
This will set -XX:ConcGCThreads to the number of cores on the machine (default: false)node[:cassandra][:jvm][:parallel_ref_proc]
-XX:ParallelRefProcEnabled (default: false)node[:cassandra][:jvm][:always_pre_touch]
-XX:AlwaysPreTouch (default: false)node[:cassandra][:jvm][:use_biased_locking]
-XX:UseBiasedLocking (default: true)node[:cassandra][:jvm][:use_tlab]
-XX:UseTLAB (default: true)node[:cassandra][:jvm][:resize_tlab]
-XX:ResizeTLAB (default: true)
Oracle JVM 8 tuning parameters: here
node[:cassandra][:seeds]
(default:[node[:ipaddress]]
): an array of nodes this node will contact to discover cluster topologynode[:cassandra][:seed_discovery][:use_chef_search]
(default: false): enabled seed discovery using Chef searchnode[:cassandra][:seed_discovery][:search_role]
(default:"cassandra-seed"
): role to use in search querynode[:cassandra][:seed_discovery][:search_query]
(default: usesnode[:cassandra][:seed_discovery][:search_role]
): allows for overriding the entire Chef search querynode[:cassandra][:seed_discovery][:count]
(default:3
): how many nodes to include into seed list. First N nodes are taken in the order Chef search returns them. IP addresses of the nodes are sorted lexographically.
node[:cassandra][:config][:num_tokens]
set the desired number of tokens. (default: 256)node[:cassandra][:config][:listen_address]
(default: node[:ipaddress]): address clients will use to connect to the nodenode[:cassandra][:config][:broadcast_address]
(default: node IP address): address to broadcast to other Cassandra nodesnode[:cassandra][:config][:rpc_address]
(default: 0.0.0.0): address to bind the RPC interface. Leave blank to lookup IP from hostname.node[:cassandra][:config][:hinted_handoff_enabled]
see http://wiki.apache.org/cassandra/HintedHandoff (default: true)node[:cassandra][:config][:max_hint_window_in_ms]
The maximum amount of time a dead host will have hints generated (default: 10800000).node[:cassandra][:config][:hinted_handoff_throttle_in_kb]
throttle in KB's per second, per delivery thread (default: 1024)node[:cassandra][:config][:max_hints_delivery_threads]
Number of threads with which to deliver hints (default: 2)node[:cassandra][:config][:authenticator]
Authentication backend (default: org.apache.cassandra.auth.AllowAllAuthenticator)node[:cassandra][:config][:authorizer]
Authorization backend (default: org.apache.cassandra.auth.AllowAllAuthorizer)node[:cassandra][:config][:permissions_validity_in_ms]
Validity period for permissions cache, set to0 to disable (default: 2000)node[:cassandra][:config][:partitioner]
The partitioner to distribute keys across the cluster (default: org.apache.cassandra.dht.Murmur3Partitioner).node[:cassandra][:config][:disk_failure_policy]
policy for data disk failures: stop, best_effort, or ignore (default: stop)node[:cassandra][:config][:key_cache_size_in_mb]
Maximum size of the key cache in memory. Set to 0 to disable, or "" for auto = (min(5% of Heap (in MB), 100MB)) (default: "", auto).node[:cassandra][:config][:key_cache_save_period]
Duration in seconds after which key cache is saved to saved_caches_directory. (default: 14400)node[:cassandra][:config][:row_cache_size_in_mb]
Maximum size of the row cache in memory, 0 to disable (default: 0)node[:cassandra][:config][:row_cache_save_period]
Duration in seconds after which row cache is saved to saved_caches_directory, 0 to disable cache save. (default: 0)node[:cassandra][:config][:row_cache_provider]
The provider for the row cache to use (default: SerializingCacheProvider)node[:cassandra][:config][:commitlog_sync]
periodic to ack writes immediately with periodic fsyncs, or batch to wait until fsync to ack writes (default: periodic)node[:cassandra][:config][:commitlog_sync_period_in_ms]
period for commitlog fsync when commitlog_sync = periodic (default: 10000)node[:cassandra][:config][:commitlog_sync_batch_window_in_ms]
batch window for fsync when commitlog_sync = batch (default: 50)node[:cassandra][:config][:commitlog_segment_size_in_mb]
Size of individual commitlog file segments (default: 32)node[:cassandra][:config][:commitlog_total_space_in_mb]
If space gets above this value (it will round up to the next nearest segment multiple), Cassandra will flush every dirty CF in the oldest segment and remove it. (default: 4096)node[:cassandra][:config][:concurrent_reads]
Should be set to 16 * drives (default: 32)node[:cassandra][:config][:concurrent_writes]
Should be set to 8 * cpu cores (default: 32)node[:cassandra][:config][:trickle_fsync]
Enable this to avoid sudden dirty buffer flushing from impacting read latencies. Almost always a good idea on SSDs; not necessary on platters (default: false)node[:cassandra][:config][:trickle_fsync_interval_in_kb]
Interval for fsync when doing sequential writes (default: 10240)node[:cassandra][:config][:storage_port]
TCP port, for commands and data (default: 7000)node[:cassandra][:config][:ssl_storage_port]
SSL port, unused unless enabled in encryption options (default: 7001)node[:cassandra][:config][:listen_address]
Address to bind for communication with other nodes. Leave blank to lookup IP from hostname. 0.0.0.0 is always wrong. (default: node[:ipaddress]).node[:cassandra][:config][:broadcast_address]
Address to broadcast to other Cassandra nodes. If '', will use listen_address (default: '')node[:cassandra][:config][:start_native_transport]
Whether to start the native transport server (default: true)node[:cassandra][:config][:native_transport_port]
Port for the CQL native transport to listen for clients on (default: 9042)node[:cassandra][:config][:start_rpc]
Whether to start the Thrift RPC server (default: true)node[:cassandra][:config][:rpc_port]
Port for Thrift RPC server to listen for clients on (default: 9160)node[:cassandra][:config][:rpc_keepalive]
Enable keepalive on RPC connections (default: true)node[:cassandra][:config][:rpc_server_type]
sync for one thread per connection; hsha for "half synchronous, half asynchronous" (default: sync)node[:cassandra][:config][:thrift_framed_transport_size_in_mb]
Frame size for Thrift (maximum field length) (default: 15)node[:cassandra][:config][:thrift_max_message_length_in_mb]
Max length of a Thrift message, including all fields and internal Thrift overhead (default: 16)node[:cassandra][:config][:incremental_backups]
Enable hardlinks in backups/ for each sstable flushed or streamed locally. Removing these links is the operator's responsibility (default: false)node[:cassandra][:config][:snapshot_before_compaction]
Take a snapshot before each compaction (default: false)node[:cassandra][:config][:auto_snapshot]
Take a snapshot before keyspace truncation or dropping of column families. If you set this value to false, you will lose data on truncation or drop (default: true)node[:cassandra][:config][:column_index_size_in_kb]
Add column indexes to a row after its contents reach this size (default: 64)node[:cassandra][:config][:compaction_throughput_mb_per_sec]
Throttle compaction to this total system throughput. Generally should be 16-32 times data insertion rate (default: 16)node[:cassandra][:config][:read_request_timeout_in_ms]
How long the coordinator should wait for read operations to complete (default: 10000)node[:cassandra][:config][:range_request_timeout_in_ms]
How long the coordinator should wait for seq or index scans to complete (default: 10000).node[:cassandra][:config][:write_request_timeout_in_ms]
How long the coordinator should wait for writes to complete (default: 10000)node[:cassandra][:config][:truncate_request_timeout_in_ms]
How long the coordinator should wait for truncates to complete (default: 60000)node[:cassandra][:config][:request_timeout_in_ms]
Default timeout for other, miscellaneous operations (default: 10000)node[:cassandra][:config][:cross_node_timeout]
Enable operation timeout information exchange between nodes to accurately measure request timeouts. Be sure ntp is installed and node times are synchronized before enabling. (default: false)node[:cassandra][:config][:streaming_socket_timeout_in_ms]
Enable socket timeout for streaming operation (default: 3600000 - 1 hour)node[:cassandra][:config][:phi_convict_threshold]
Adjusts the sensitivity of the failure detector on an exponential scale (default: 8)node[:cassandra][:config][:endpoint_snitch]
SimpleSnitch, PropertyFileSnitch, GossipingPropertyFileSnitch, RackInferringSnitch, Ec2Snitch, Ec2MultiRegionSnitch (default: SimpleSnitch)node[:cassandra][:config][:dynamic_snitch_update_interval_in_ms]
How often to perform the more expensive part of host score calculation (default: 100)node[:cassandra][:config][:dynamic_snitch_reset_interval_in_ms]
How often to reset all host scores, allowing a bad host to possibly recover (default: 600000)node[:cassandra][:config][:dynamic_snitch_badness_threshold]
Allow 'pinning' of replicas to hosts in order to increase cache capacity. (default: 0.1)node[:cassandra][:config][:request_scheduler]
Class to schedule incoming client requests (default: org.apache.cassandra.scheduler.NoScheduler)node[:cassandra][:config][:index_interval]
index_interval controls the sampling of entries from the primary row index in terms of space versus time (default: 128).node[:cassandra][:config][:auto_bootstrap]
Setting this parameter to false prevents the new nodes from attempting to get all the data from the other nodes in the data center. (default: true).node[:cassandra][:config][:enable_assertions]
Enable JVM assertions. Disabling this in production will give a modest performance benefit (around 5%) (default: true).node[:cassandra][:config][:data_file_directories]
(default: node['cassandra']['data_dir']): C* data cirectoriesnode[:cassandra][:config][:saved_caches_directory]
(default: saved_caches_directory): C* saved cache directorynode[:cassandra][:config][:commitlog_directory]
(default: node['cassandra']['commitlog_dir']) *C commit log directory
node[:cassandra][:config][:memtable_flush_queue_size]
Number of full memtables to allow pending flush, i.e., waiting for a writer thread (default: 4)node[:cassandra][:config][:in_memory_compaction_limit_in_mb]
Size limit for rows being compacted in memory (default: 64)node[:cassandra][:config][:concurrent_compactors]
Sets the number of concurrent compaction processes allowed to run simultaneously on a node. (default: nil, which will result in one compaction process per CPU core)node[:cassandra][:config][:multithreaded_compaction]
Enable multithreaded compaction. Uses one thread per core, plus one thread per sstable being merged. (default: false)node[:cassandra][:config][:compaction_preheat_key_cache]
Track cached row keys during compaction and re-cache their new positions in the compacted sstable. Disable if you use really large key caches (default: true)node[:cassandra][:config][:native_transport_min_threads]
Min number of threads for handling transport requests when the native protocol is used (default: nil)node[:cassandra][:config][:native_transport_max_threads]
Max number of threads for handling transport requests when the native protocol is used (default: nil)
node[:cassandra][:config][:broadcast_rpc_address]
RPC address to broadcast to drivers and other Cassandra nodes (default: node[:ipaddress])node[:cassandra][:config][:tombstone_failure_threshold]
tombstone attribute, check C* documentation for more info (default: 100000)node[:cassandra][:config][:tombstone_warn_threshold]
tombstone attribute, check C* documentation for more info (default: 1000)node[:cassandra][:config][:sstable_preemptive_open_interval_in_mb]
This helps to smoothly transfer reads between the sstables, reducing page cache churn and keeping hot rows hot (default: 50)node[:cassandra][:config][:memtable_allocation_type]
Specify the way Cassandra allocates and manages memtable memory (default: heap_buffers)node[:cassandra][:config][:index_summary_capacity_in_mb]
A fixed memory pool size in MB for for SSTable index summaries. If left empty, this will default to 5% of the heap size (default: nil)node[:cassandra][:config][:index_summary_resize_interval_in_minutes]
How frequently index summaries should be resampled (default: 60)node[:cassandra][:config][:concurrent_counter_writes]
Concurrent writes, since writes are almost never IO bound, the ideal number of "concurrent_writes" is dependent on the number of cores in your system; (8 * number_of_cores) (default: 32)node[:cassandra][:config][:counter_cache_save_period]
Duration in seconds after which Cassandra should save the counter cache (keys only) (default: 7200)node[:cassandra][:config][:counter_cache_size_in_mb]
Counter cache helps to reduce counter locks' contention for hot counter cells. Default value is empty to make it "auto" (min(2.5% of Heap (in MB), 50MB)). Set to 0 to disable counter cache. (default: nil)node[:cassandra][:config][:counter_write_request_timeout_in_ms]
How long the coordinator should wait for counter writes to complete (default: 5000)node[:cassandra][:config][:commit_failure_policy]
policy for commit disk failures (default: stop)node[:cassandra][:config][:cas_contention_timeout_in_ms]
How long a coordinator should continue to retry a CAS operation that contends with other proposals for the same row (default: 1000)node[:cassandra][:config][:batch_size_warn_threshold_in_kb]
Log WARN on any batch size exceeding this value. 5kb per batch by default (default: 5)node[:cassandra][:config][:batchlog_replay_throttle_in_kb]
Maximum throttle in KBs per second, total. This will be reduced proportionally to the number of nodes in the cluster (default: 1024)
node[:cassandra][:setup_jamm]
(default: false): install the jamm jar file and use it to set java option-javaagent
, obsolete for C* versions>v0.8.0
node[:cassandra][:jamm][:sha256sum]
(default: calculated): jamm lib sha256sum for calculated versionnode[:cassandra][:jamm][:base_url]
(default: calculated): jamm lib jar urlnode[:cassandra][:jamm][:jar_name]
(default: calculated): jamm lib jar namenode[:cassandra][:jamm][:version]
(default: calculated): jamm lib version
node[:cassandra][:jna][:base_url]
The base url to fetch the JNA jar (default: https://github.com/twall/jna/tree/4.0/dist)node[:cassandra][:jna][:jar_name]
The name of the jar to download from the base url. (default: jna.jar)node[:cassandra][:jna][:sha256sum]
The SHA-256 checksum of the file. If the local jna.jar file matches the checksum, the chef-client will not re-download it. (default: dac270b6441ce24d93a96ddb6e8f93d8df099192738799a6f6fcfc2b2416ca19)
node[:cassandra][:setup_priam]
(default: false): install the priam jar file and use it to set java option-javaagent
, uses the priam version corresponding to the cassandra versionnode[:cassandra][:priam][:sha256sum]
(default: 9fde9a40dc5c538adee54f40fa9027cf3ebb7fd42e3592b3e6fdfe3f7aff81e1): priam lib sha256sum for version2.2.0
node[:cassandra][:priam][:base_url]
(default: priam url on maven.org): priam lib jar urlnode[:cassandra][:priam][:jar_name]
(default: calculated): priam lib jar name
node[:cassandra][:logback][:file][:max_file_size]
(default: "20MB"): logback File appender log file rotation sizenode[:cassandra][:logback][:file][:max_index]
(default: 20): logback File appender log files max_indexnode[:cassandra][:logback][:file][:min_index]
(default: 1): logback File appender log files min_indexnode[:cassandra][:logback][:file][:pattern]
(default: "%-5level [%thread] %date{ISO8601} %F:%L - %msg%n"): logback File appender log patternnode[:cassandra][:logback][:debug][:max_file_size]
(default: "20MB"): logback File appender log file rotation sizenode[:cassandra][:logback][:debug][:max_index]
(default: 20): logback File appender log files max_indexnode[:cassandra][:logback][:debug][:min_index]
(default: 1): logback File appender log files min_indexnode[:cassandra][:logback][:debug][:pattern]
(default: "%-5level [%thread] %date{ISO8601} %F:%L - %msg%n"): logback File appender log patternnode[:cassandra][:logback][:stdout][:enable]
(default: true): enable logback STDOUT appendernode[:cassandra][:logback][:stdout][:pattern]
(default: "%-5level %date{HH:mm:ss,SSS} %msg%n"): logback STDOUT appender log patternnode[:cassandra][:logback][:syslog][:enable]
(default: false): enable logback SYSLOG appender. Requires RSYSLOG be installed and running on the node.node[:cassandra][:logback][:syslog][:host]
(default: localhost): The host name the syslog is written to.node[:cassandra][:logback][:syslog][:facility]
(default: USER) The facility specified for the appender.node[:cassandra][:logback][:syslog][:pattern]
(default: "%-5level [%thread] %F:%L - %msg%n") lockback SYSLOG appender log pattern
node[:cassandra][:limits][:memlock]
(default: "unlimited"): memory ulimit for Cassandra node processnode[:cassandra][:limits][:nofile]
(default: 48000): file ulimit for Cassandra node processnode[:cassandra][:limits][:nproc]
(default: "unlimited"): process ulimit for Cassandra node process
node[:cassandra][:yum][:repo]
(default: datastax): name of the repo from which to installnode[:cassandra][:yum][:description]
(default: "DataStax Repo for Apache Cassandra"): description of the reponode[:cassandra][:yum][:baseurl]
(default: "http://rpm.datastax.com/community"): repo urlnode[:cassandra][:yum][:mirrorlist]
(default: nil): a mirrorlist filenode[:cassandra][:yum][:gpgcheck]
(default: false): whether to usegpgcheck
node[:cassandra][:yum][:enabled]
(default: true): whether the repo is enabled by defaultnode[:cassandra][:yum][:options]
(default: ""): Additional options to pass toyum_package
node[:cassandra][:opscenter][:server][:package_name]
(default: opscenter-free)node[:cassandra][:opscenter][:server][:port]
(default: 8888)node[:cassandra][:opscenter][:server][:interface]
(default: 0.0.0.0)node[:cassandra][:opscenter][:server][:authentication]
(default: false)node[:cassandra][:opscenter][:cassandra_metrics][:ignored_keyspaces]
(default: [system, OpsCenter])node[:cassandra][:opscenter][:cassandra_metrics][:ignored_column_families]
(default: [])node[:cassandra][:opscenter][:cassandra_metrics][:1min_ttl]
(default: 604800)node[:cassandra][:opscenter][:cassandra_metrics][:5min_ttl]
(default: 2419200)node[:cassandra][:opscenter][:cassandra_metrics][:2hr_ttl]
(default: 31536000)node[:cassandra][:opscenter][:custom_configuration]
(default: {}) a hash of custom configuration sections to add to opscenterd.conf, e.g.:
{
'ui' => {
'default_api_timeout' => 300
},
'stat_reporter' => {
'interval' => 1
}
}
node[:cassandra][:opscenter][:agent][:download_url]
(default: "") Required. You need to specify agent download url, because that could be different for each opscenter server version. ( S3 is a great place to store packages )node[:cassandra][:opscenter][:agent][:checksum]
(default:nil
)node[:cassandra][:opscenter][:agent][:install_dir]
(default:/opt
)node[:cassandra][:opscenter][:agent][:install_folder_name]
(default:opscenter_agent
)node[:cassandra][:opscenter][:agent][:binary_name]
(default:opscenter-agent
) Introduced since Datastax changed agent binary name from opscenter-agent to datastax-agent. Make sure to set it right if you are updating to 4.0.2node[:cassandra][:opscenter][:agent][:server_host]
(default: "" ). If left empty, will use search to get IP by opscenterserver_role
role.node[:cassandra][:opscenter][:agent][:server_role]
(default:opscenter_server
). Will be use for opscenter server IP lookup if:server_host
is not set.node[:cassandra][:opscenter][:agent][:use_chef_search]
(default:true
). Determines whether chef search will be used for locating the data agent server.node[:cassandra][:opscenter][:agent][:use_ssl]
(default:false
)
node[:cassandra][:opscenter][:agent][:package_name]
(default: "datastax-agent" ).node[:cassandra][:opscenter][:agent][:server_host]
(default: "" ). If left empty, will use search to get IP by opscenterserver_role
role.node[:cassandra][:opscenter][:agent][:server_role]
(default:opscenter_server
). Will be use for opscenter server IP lookup if:server_host
is not set.node[:cassandra][:opscenter][:agent][:use_ssl]
(default:false
)
node[:cassandra][:rackdc][:dc]
(default: "") The datacenter to specify in the cassandra-rackdc.properties file. (GossipingPropertyFileSnitch only)node[:cassandra][:rackdc][:rack]
(default: "") The rack to specify in the cassandra-rackdc.properties file (GossipingPropertyFileSnitch only)node[:cassandra][:rackdc][:prefer_local]
(default: "false") Whether the snitch will prefer the internal ip when possible, as the Ec2MultiRegionSnitch does. (GossipingPropertyFileSnitch only)
Create a branch, make the changes, then submit a pull request on GitHub with a brief description of what you've done and why your changes should be included.
Write new resource/attribute description to README.md
Run the tests (rake
), ensuring they all pass.
Michael S. Klishin, Travis CI Development Team, and contributors, 2012-2016.
Released under the Apache 2.0 License.