Vitess v2.0.0-beta.2

Changes since v2.0.0-beta.1

Client Libraries

• JDBC-compatible wrapper
  • Contributed by Flipkart
  • See JDBC Example
• Update Java client to better match semantics of JDBC ResultSet
  • columnIndex now starts at 1
  • columnLabel is now case-insensitive
• Java: Synchronous client wrapper
  • To simplify and enforce proper usage of Java client in blocking mode.
• Java: Support more bind variable types
• Python: Always use colon style for bind variables
  • Removes extra translation layer for % formatting from Python client.
  • This makes the bind variable style consistent across all Vitess clients.
  • See local example client changes

SQL Support

• Phase 2 of VTGate V3 API
  • Can now route more complex queries.
  • Can perform cross-shard joins.
  • See phase 2 of V3 design for what is now supported.
• Cross-shard sequences
  • Note that NEXT is now a reserved word, and must be back-quoted if used as a column name.
• Support INTERVAL construct
• Support keyspace-qualified names
  • i.e. SELECT * FROM keyspace.table, analogous to db.table outside Vitess.
• Support STRAIGHT_JOIN and UNION DISTINCT SELECT

Resharding

• Support merging shards
• Support resharding tables with no keyspace_id column
  • As long as there is a VSchema that allows Vitess to compute the keyspace_id.
• SplitDiff: support custom sharding column name

Deployment and Management

• Configure SSL and authentication for gRPC
  • See Transport Security Model docs.
• VTGate watches for VSchema updates and refreshes automatically
• V3: Allow duplicate table names in vschema
  • As long as they're in different keyspaces.
• V3: Simplify VSchema
  • Remove the Class concept
  • VSchema is now per-keyspace
• Removed deprecated gorpc-based protocols
  • All protocols are now based on gRPC.
• Ceph Backup Storage Plugin
  • Contributed by Flipkart

Build Dependencies

• Pre-built Docker images for Percona Server
• We now depend on Go 1.5+ for native vendoring

Analysis

• Pre-scan algorithm for MapReduce/Hadoop query splitting

Vitess v2.0.0-beta.1

Changes since v2.0.0-alpha5

Client Libraries

• VTGateV3 Go driver
  can now also support certain types of VTGateV2 queries
• Publish PHP client to Packagist
• Refactor PHP client for better standards compliance
• PHP client returns associative array rows
• Java client is asynchronous by default
• Java client now returns 0 when converting SQL NULL to primitive types
• Python client now defaults to gRPC
• More sample clients for examples/local
  • In addition to Python,
    there are now sample clients for Go,
    PHP,
    and Java.

Query Service

• query.proto revamp
  • Fix NULL handling in new proto3-based query result format.
  • Unify type system for bind variables and query results.
• query_rules improvements
  • Improved JSON syntax for bindvar constraints.
  • Debug URL for viewing rule sets.
• Support REGEXP and RLIKE in
  SQL parser

Build Dependencies

• Default ZooKeeper version upgraded
  from 3.3.5 to 3.4.6.
• Update to gRPC 0.13
• Support building on OS X

Deployment and Management

• Add vtctld server to vtcombo
• Kubernetes config updates
  • Use Kubernetes DNS for service discovery
    and etcd bootstrap.
  • Kubernetes example for MySQL 5.6.
  • Use Kubernetes proxy to link from vtctld to vttablet status pages.
• Make MySQL users and privileges configurable
  when bootstrapping MySQL data directory.
  • Uses mysql_install_db and init_db.sql
    instead of extracting the old mysql-db-dir.tbz archive.
• Support semi-sync without async fallback
• Remove old vtctld UI files
  • Deployments no longer need to include go/cmd/vtctld/templates folder or
    set -templates flag on vtctld.
• Add option to bypass sanity check in ApplySchema
• Print human-readable tables for CLI queries
  • e.g. for ExecuteFetchAsDba

Vitess v2.0.0-alpha5

Changes since v2.0.0-alpha4

• New built-in backup
  plugin to store backups in Google Cloud Storage.
  • See the Vitess on Kubernetes guide for an
    example of usage.
• The Vitess PHP client now
  uses gRPC.
• Vitess Hadoop connector
  now uses the new gRPC-based Java client.
• Self-contained demo
  for VTGate V3, which is a preview feature as
  of Vitess 2.0.
• The vtctlclient-to-vtctld protocol now defaults to gRPC.

Vitess v2.0.0-alpha4

Changes since v2.0.0-alpha3

• Fixes for "bad node version" error when using etcd. (#1176, #1184)
• Java client is now feature-complete.
  • The last piece was our new canonical error code translation.

Vitess v2.0.0-alpha3

Changes since v2.0.0-alpha2

• Better PEP 0249 compliance in Python client.
• Type-checking result cursor for Java client.
• VTGate automatically adds annotations for filtered replication.
• All RPC protocols are now off by default, and must be explicitly enabled with the -service_map flag.
  • Previously the BSON protocols were enabled by default, and had to be explicitly disabled in -service_map if unwanted.
• SQL is now consistently represented as a UTF-8 string.
  • Previously it was represented in some places as a byte array.
• Update to gRPC beta.

Vitess v2.0.0-alpha2

Changes since v2.0.0-alpha1

• Fix for breaking API change in grpc-go (#1052).

Vitess v2.0.0-alpha1

This is the first alpha cut for the 2.0.0 release.

Changes since 1.0

• In addition to the manual deployment
  option, Vitess can now be scaled dynamically in any cloud platform supported
  by Kubernetes, including
  Google Container Engine and
  AWS.
  • Getting Started on Kubernetes
  • Dynamic Resharding in Kubernetes
• Docker images available for both manual and cloud deployments:
  • Automated Builds
  • Lightweight Builds
• Official client libraries for:
  • Java
  • Python
  • PHP
  • Go
  • NOTE: Error handling is still being refactored as of alpha1.
• Using new gRPC framework built on HTTP/2.
  • For Vitess server-to-server communication.
  • For Java and Go client communication.
• Underlying DB nodes can run either:
  • MariaDB 10.0, or
  • MySQL 5.6
• Consistent datastore used for cluster management can be:
  • etcd 2.0
  • ZooKeeper
  • Pluggable interface available for adding more.
• New responsive admin dashboard (on vtctld).
  • Client-side AngularJS app with Material Design.
• Built-in Backup/Restore
  for disaster recovery and provisioning.
  • New replica instances automatically restore from latest backup, and then
    catch up to master.
  • Queries from the app are automatically routed only to healthy, caught-up
    replicas.
  • Restore mechanism is routinely exercised as part of normal provisioning,
    so you know it'll work when you really need it.
  • NOTE: Currently this requires a network mount. We plan to add plugins for
    direct access to blob stores like Google Cloud Storage and Amazon S3.
• Streaming, filtered data copy for transparent dynamic resharding.
  • Resharding now requires no additional disk space on the source machine,
    so it works even when existing data fills up more than half of the largest
    available node size.
• GTID-based MySQL failovers (also known as reparenting)
  allow old masters to be seamlessly reparented back into the replication stream
  as replicas.
  • This makes failover a routine, automated operation, allowing downtime-free
    workflows such as:
    • Reparent to another replica.
    • Upgrade the old master by restarting it.
    • Reparent to upgraded master.
    • Upgrade remaining replicas one at a time.