Apache cassandra v4.0

Apache Cassandra v4.0
1 © DataStax, All Rights Reserved. Confidential
DataStax Japan
Yuki Morishita
Cassandra Meetup in Tokyo February 2019
2019/2/21

Who am I?
Yuki Morishita (yuki@datastax.com)
– Principal Architect at DataStax Japan
– Apache Cassandra committer
© DataStax, All Rights Reserved.2 Confidential

Apache Cassandra
v4.0!!!

Apache Cassandra v4.0
• Been working for 1.5 years after 3.11.0 release
• Feature freeze on September 1, 2018
• No fixed release date
• See NEWS.TXT for upgrade instruction
– No direct upgrade from pre-3.0 versions

CHANGES.TXT
• More than 340 changes
(new features / bug fixes)
4.0
* Fix SimpleStrategy option validation (CASSANDRA-15007)
* Don't try to cancel 2i compactions when starting anticompaction (CASSANDRA-15024)
* Avoid NPE in RepairRunnable.recordFailure (CASSANDRA-15025)
* SSL Cert Hot Reloading should check for sanity of the new keystore/truststore before loading it
(CASSANDRA-14991)
* Avoid leaking threads when failing anticompactions and rate limit anticompactions (CASSANDRA-
15002)
* Validate token() arguments early instead of throwing NPE at execution (CASSANDRA-14989)
* Add a new tool to dump audit logs (CASSANDRA-14885)
* Fix generating javadoc with Java11 (CASSANDRA-14988)
* Only cancel conflicting compactions when starting anticompactions and sub range compactions
(CASSANDRA-14935)
* Use a stub IndexRegistry for non-daemon use cases (CASSANDRA-14938)
* Don't enable client transports when bootstrap is pending (CASSANDRA-14525)
* Make antiCompactGroup throw exception on error and anticompaction non cancellable
again (CASSANDRA-14936)
* Catch empty/invalid bounds in SelectStatement (CASSANDRA-14849)
* Auto-expand replication_factor for NetworkTopologyStrategy (CASSANDRA-14303)
* Transient Replication: support EACH_QUORUM (CASSANDRA-14727)
* BufferPool: allocating thread for new chunks should acquire directly (CASSANDRA-14832)
* Send correct messaging version in internode messaging handshake's third message (CASSANDRA-
14896)
* Make Read and Write Latency columns consistent for proxyhistograms and tablehistograms
(CASSANDRA-11939)
* Make protocol checksum type option case insensitive (CASSANDRA-14716)
* Forbid re-adding static columns as regular and vice versa (CASSANDRA-14913)

New Features
• Virtual Tables
• Transient Replication
• Audit logging
• Full query logging
• Zero-copy SSTable streaming

Virtual Tables
• Before Virtual Table…
– JMX or nodetool
• Cache information
– nodetool info
• Thread pools
– nodetool tpstats
• Connected clients
– nodetool clientstats –all (new in v4.0!)
• SSTable activities
– nodetool compactionstats
– Config
• vim cassandra.yaml

Virtual Tables
• After Virtual Table… (in addition to the methods in prev slide)
– Cache information
• SELECT * FROM system_views.caches;
• SELECT * FROM system_views.caches WHERE name = 'chunks’;
– Thread pools
• SELECT * FROM system_views.thread_pools;
– Connected clients
• SELECT * FROM system_views.clients;
– SSTable activities
• SELECT * FROM system_views.sstable_tasks
– Config
• SELECT * FROM system_views.settings;

Virtual Tables
• How about updating config?
– Not yet supported
• Keep track of issues
– https://issues.apache.org/jira/browse/CASSANDRA-14670?jql=labels%20%3D%20virtual-tables

Transient Replication
• Save disk space, CPU cycle and IO with temporary copy (transient replica)
• Temporary copy is replicated with incremental repair, and removed
• Experimental feature in v4.0
• Unsupported features
– Monotonic read (read with blocking read repair)
– Logged batch
– LWT
– Counters
– Secondary Index / Materialized view (never support)

• Read / Write with RF=3, CL=QUORUM
Write Read

• Read / Write with RF=3 (with 1 transient replica), CL=QUORUM
Write Read
Transient ReplicaTransient Replica

• Read / Write with RF=3 (with 1 transient replica), CL=QUORUM
• When one node is down
Write Read
Transient ReplicaTransient Replica

• Temporary copy is replicated and removed after incremental repair
Transient Replica
Incremental repair

• Temporary copy is replicated and removed after incremental repair
Transient Replica

CREATE KEYSPACE ks
WITH replication = {
'class': 'NetworkTopologyStrategy',
'replication_factor': '3/1'
};

Audit Logging
• Log queries / authentication to a node for auditing
• When audit events are logged?
– After executing CQL / Authentication, before sending response
• Pluggable logger (IAuditLogger interface) to publish audit event anywhere
– BinAuditLogger (default) – binary log
– FileAuditLogger – logback
• Audit log includes:
– User, Client IP, timestamp, category, operation, …
• See https://cassandra.apache.org/doc/latest/operating/audit_logging.html

Audit Logging
• Enabling audit logging
– cassandra.yaml
# Audit logging - Logs every incoming CQL command request, authentication to a node. See the docs
# on audit_logging for full details about the various configuration options.
audit_logging_options:
enabled: false
logger: BinAuditLogger
# audit_logs_dir:
# included_keyspaces:
# excluded_keyspaces: system, system_schema, system_virtual_schema
# included_categories:
# excluded_categories:
# included_users:
# excluded_users:
# roll_cycle: HOURLY
# block: true
# max_queue_weight: 268435456 # 256 MiB
# max_log_size: 17179869184 # 16 GiB
## archive command is "/path/to/script.sh %path" where %path is replaced with the file being rolled:
# archive_command:
# max_archive_retries: 10

Audit Logging
• Enabling audit logging
– nodetool enableauditlog
• Can change the same settings using command line options
– --included-keyspaces, --logger, etc.
– nodetool disableauditlog

Full Query Logging
• Enable workload recording, replay and compare result
• Use case
– Capture workload in production. Use it in dev environment to test in different Cassandra version.
Compare result for correctness.
Capture Replay
Full Query Log
Prod Dev

Full Query Logging
• Use nodetool to start / stop capturing
– nodetool enablefullquerylog
• Enable full query logging, defaults for the options are configured in cassandra.yaml
– nodetool disablefullquerylog
• Disable the full query log
– nodetool resetfullquerylog
• Stop the full query log and clean files in the configured full query log directory from cassandra.yaml as well as
JMX

Full Query Logging
• Use fqltool to replay and compare result
– fqltool replay
• Replay full query logs
– fqltool compare
• Compare result files generated by fqltool replay
– fqltool dump
• Dump the contents of a full query log

Zero-copy SSTable streaming
• SSTable streaming
Data.db
Index.db
Filter.db
Compress
sionInfo.db
Sender node Receiver node
Give me partitions
between
token 0 and 100

Data.db
Index.db
Filter.db
Compress
sionInfo.db
Here are partitions
between
token 0 and 100
in this SSTable

Data.db
Index.db
Filter.db
Compress
sionInfo.db
• Decompress
• Deserialize
• Update Stats / BF /
Index
• Serialize
• Compress
Data.dbIndex.db
Filter.db
Compress
sionInfo.db

Data.db
Index.db
Filter.db
Compress
sionInfo.db
Give me partitions
between
token 0 and 100

Data.db
Index.db
Filter.db
Compress
sionInfo.db
This SSTable contains
partitions in
token 0 and 100!

Data.db
Index.db
Filter.db
Compress
sionInfo.db
Zero-copy
Transfer
Data.db
Index.db
Filter.db
Compress
sionInfo.db

• Enabled by default
– Can be turned off in cassandra.yaml (stream_entire_sstables)
• Only works for SSTables in table using LeveledCompactionStrategy
– For now…
• https://cassandra.apache.org/blog/2018/08/07/faster_streaming_in_cassandra.html

Other Notable Changes
• Experimental Java 11 support
– Can use ZGC
• Asynchronous internode messaging / streaming
– More efficient if you have large cluster
• Thrift / COMPACT STORAGE removed (finally)
• Network authorization
– Can create role that only has access to certain datacenter
• CDC improvement
– Change data is available faster than previous implementation

Other Notable Changes
• Read repair change
– dc_read_repair_chance / read_repair_chance are gone
• no more async read repair
– Blocking read repair can be turned off
• ALTER TABLE ... WITH read_repair = NONE;

Apache Cassandra
Side-car project

Cassandra side-car
• Cassandra management process that runs separately from Cassandra daemon
• Provides HTTP API to monitor and manage Cassandra node
• Proposal
– https://cwiki.apache.org/confluence/pages/viewpage.action?pageId=95652224
• Repository:
– https://github.com/apache/cassandra-sidecar
– Currently healthcheck API is implemented

Thank you,
Questions?

Apache cassandra v4.0

Recommended

More Related Content

What's hot (20)

Similar to Apache cassandra v4.0 (20)

More from Yuki Morishita (12)

Recently uploaded (20)

Apache cassandra v4.0