Skip to content

Latest commit

 

History

History
463 lines (337 loc) · 16.1 KB

CHANGELOG.md

File metadata and controls

463 lines (337 loc) · 16.1 KB
Raw

1.5.2

Released on: 2022/05/20

Fixed

  • Improve DC-aware LB policies robustness. #147
  • Ensure request-aware + DC-aware LB policy prioritizes local peers over remote ones.

1.5.1

Released on: 2021/02/18

Added
  • Enable TLS 1.2 in LuaSec fallback and disable SSLv2, SSLv3, and TLS 1.0. #138
  • Enable LuaSec fallback in ngx_stream_lua's preread phase. #136

Fixed

  • Ensure cluster:refresh() always releases its lock when encountering an error. #140

1.5.0

Released on: 2019/09/25

Added
  • The cluster:refresh() method is now safe to be called at runtime, allowing cluster topology changes (added or removed nodes) to be taken into consideration. #134
  • The cluster:refresh() method now accepts an optional timeout argument, and returns the topology changes (if any). #134

1.4.0

Released on: 2019/05/09

Changed
  • Improve host still considered down error logs to include the error causing each node to be considered as "DOWN", as well as the duration for which it will still be considered "DOWN". #129

1.3.4

Released on: 2019/01/23

Changed
  • Do not log the "rpc_address set to 0.0.0.0" warning when opts.silent is on. #126

1.3.3

Released on: 2018/11/09

Added
  • Support for binary CQL frames with custom payload. #119
Fixed
  • Favor the rpc_address value to connect to the provided contact points. #122

1.3.2

Released on: 2018/08/10

Fixed
  • Environments with DNS load-balancing in effect for contact_points provided as hostnames (e.g. Kubernetes with contact_points = { "cassandra" }) could result in no host details for <peer IP> errors when using multiple instances of the Cluster module. This is now fixed. #118

1.3.1 - 2018/07/02

Fixed
  • The new request-aware load-balancing policies can now be used in the init_by_lua* context when using the lua-resty-core module (only applies to OpenResty environments). #117

1.3.0 - 2018/06/14

Added
  • New request-aware load-balancing policy for OpenResty environments.
  • New request-aware + datacenter-aware load-balancing policy for OpenResty environments. Thanks @kikito for the patch! #114

1.2.3 - 2017/07/20

Added
  • Expose the check_schema_consensus() method of the Cluster module for host applications to use.
Fixed
  • Safely share peers across workers in the Cluster module. #97

1.2.2 - 2017/05/17

Added
  • New cafile option for the Cluster module. This allows supporting SSL connections to Cassandra clusters when lua-cassandra is used in contexts that do not support cosockets, and fallback on LuaSocket. #95

1.2.1 - 2017/04/03

Fixed
  • Force the Nginx time to be updated when checking for schema consensus timeout. #90

1.2.0 - 2017/03/24

Added
  • Methods to manually add and remove peers from a Cluster module instance. #87

1.1.1 - 2016/02/28

Added
  • Expose the underlying first_coordinator and wait_schema_consensus functions from the Cluster module.

1.1.0 - 2016/01/12

Changed
  • ⚠️ Peers are now part of different connection pools depending on their keyspace. This can fix eventual issues when using several keyspaces with a single peer/cluster instance. 6c0db5e This is a breaking change: 
      -- before:
  local peer = cassandra.new()
  peer:connect()
  peer:set_keyspace('my_keyspace')

  -- after:
  local peer = cassandra.new()
  peer:connect()
  peer:change_keyspace('my_keyspace') -- closes the underlying connection and open a new one
Added
  • New coordinator_options for execute()/batch()/iterate() allowing for more granularity in keyspace settings. Accepted options are keyspace and no_keyspace. Example: 
      local Cluster = cluster.new {
    keyspace = 'my_keyspace'
  }

  local res = cluster:execute('SELECT * FROM local', nil, {
    keyspace = 'system' -- will spawn or reuse a peer with 'system' keyspace
    --no_keyspace = true -- would disable setting a keyspace for this request
  })

cdc6607

  • Support for binary protocol v4. #61
    • New cassandra.null CQL marshalling type. This type is different than cassandra.unset for protocol v4 and will set to null existing columns (in protocol v4 usage only).
    • Parse SCHEMA_CHANGE results for FUNCTION and AGGREGATE.
    • The Cluster module now parses warnings contained in response frames and logs them at the ngx.WARN level.
  • Implement a silent option for Cluster.new() to disable logging in the nginx error logs. #60
  • Implement a lock_timeout option for Cluster.new() to specify a max waiting time in seconds for the cluster refreshing and requests preparing mutexes. This option prevents such mutexes to hang for too long. 2bd3d66
  • The cluster:refresh() method now returns the list of fetched Cassandra nodes from the cluster as a third return value. 34f5f11
Fixed
  • Correctly logs the address of peers being set UP or DOWN in the warning logs. 40fd870
  • Better error messages for SSL handshake/locking failures.
  • Better handling in case the shm containing the cluster info is full. We do not override previous values at the risk of losing cluster nodes info, but error out with the "no memory" error instead. 4520a3b
  • Correctly receives read and connect timeout options. #71
  • Log the reason behind retrying a request in the cluster module. #71
  • Fallback on listen_address when rpc_address is "bind all" when refreshing the cluster nodes with the cluster module. #72
  • Propagate the CQL version in use when marshalling CQL collection types such as map, set, tuple or udt. We now properly marshall such nested CQL values. #73

1.0.0 - 2016/07/27

⚠️ This release contains breaking changes. The library has been rewritten to greatly increase performance, usability and maintanability. The result is very pleasant and eleguant: we now offer 2 modules, one "single host", compatible with PUC Lua 5.1/5.2, and a "cluster" module, greatly optimized and only compatible with OpenResty.

Changed
  • New single host cassandra module, able to connect to a single Cassandra node.
  • New cluster resty.cassandra.cluster module, which leverages the single host module and is able to efficiently deal with a multi-nodes Cassandra cluster.
  • No more tables as errors. All errors returned by those modules are now strings.
  • Some considerable performance improvements compared to the previous versions: according to the benchmarks I ran while writing this new implementation (on a late 2013 Macbook Pro), this new version allows up to 10k q/sec compared to 2k q/s with its old version. I plan on making those benchmarks available publicly in the near future.
Added
  • Support for named arguments when binding query parameters (binary protocol v3).
  • Support for client-side timestamps (binary protocol v3).
  • Support for query tracing.
  • New "datacenter-aware round robin" load balancing policy. This policy will prioritize nodes from the local datacenter in multi-DC setups.
  • A much more complete and reliable test suite (yay!).
  • A more complete documentation, including the available policies and better usage examples.

0.5.1 - 2016/04/11

Fixed
  • Use the default_port option when connecting to peers instead of the CQL default (9042). #49

0.5.0 - 2016/02/02

Changed

  • Following Datastax's model and allowing better flexibility with various C* providers, authentication now happens with an AuthProvider table that must be instanciated and passed to a session's options. Example:

    local session, err = cassandra.spawn_session {
  shm = "...",
  contact_points = {...},
  username = "username",
  password = "password"
}

    Becomes:

    local session, err = cassandra.spawn_session {
  shm = "...",
  contact_points = {...},
  auth = cassandra.auth.PlainTextProvider("username", "password")
}

    The cassandra module contains AuthProviders in cassandra.auth. Only PlainTextProvider is currently implemented.

  • Remove the set_log_lvl() function from the cassandra module.

Added
  • Stronger test suite and new specs for SSL and Authentication.

0.4.2 - 2016/01/16

Fixed
  • Correct timeout check for schema consensus. Prior to this, schema consensus systematically timed out after 0.5s. #24
Added
  • More documentation.
  • Stronger test suite and CI.

0.4.1 - 2015/12/18

Fixed

0.4.0 - 2015/12/17

Complete rewrite of the driver, to the exception of the serializers.

Breaking changes

This release is a complete breaking change with previous versions of the driver. #15

Added
  • Cluster topology auto detection. contact_points are not used as the only available nodes anymore but as entry point to discover the cluster's topology.
  • Cluster awareness capabilities. The driver is now capable of keeping track of which nodes are healthy or unhealthy.
  • Load balancing, reconnection, retry and address resolution policies. Only one of each is currently implemented.
    • Load balancing: shared round-robin accross all workers. Used to load-balance the queries in the cluster.
    • Reconnection: shared exponential (exponential reconnection time shared accross all workers). Used to determine when an unhealthy node should be retried.
    • Retry: a basic retry policy. Used to determine which queries to retry or throw errors.
    • Address resolution: a basic address resolution policy. Used to resolve rpc_address fields.
  • Waiting for schema consensus between nodes on SCHEMA_CHANGE (DML queries).
  • Many more options, configurable per session/query (queries can be executed with options overriding the session's option).
  • Complete abstraction of prepared queries. A simple option to execute() will handle the query preparation. If a node throws an UNPREPARED error, the query will be prepared and retried seamlessly.
  • Stronger test suite. Unit/integration tests with Busted, and ngx_lua integration tests with Test::Nginx Perl module. Travis-CI jobs are also faster and more reliable, and run all test suites.
  • Binary protocol auto-detection: downgrade from 3 to 2 automatically when using C* 2.0.
  • Compatible with Lua 5.1, 5.2, 5.3, LuaJIT.
  • Overall, a better architecture for a better maintainability.
Unchanged
  • Still optimized for ngx_lua (cosocket API) and plain Lua (with LuaSocket).
  • TLS client-to-node encryption and Authentication (PasswordAuthenticator) are still supported.
  • The serializers stayed the same (even if their architecture was rewritten).
Removed
  • No more support for query tracing (will be added back later).

0.3.6 - 2015/08/06

Added
  • Better error handling in case of CA certificate error with Luasocket.
  • Port number in addition to host in socket errors.
Fixed
  • Shuffling of contact points array.

0.3.5 - 2015/07/15

Added
  • Expose consistency and batch types constants in cassandra.
  • Better SSL handshake error handling.

0.3.3 - 2015/07/14

Fixed
  • Binary protocol v3 accidentally used v2 encoding methods.
  • Startup message accidentally being sent for already established connections (with reusable sockets).

0.3.0 - 2015/07/07

Added
  • Support for SSL encryption (client/server).
Fixed
  • IPv6 addresses encoding.

0.2.0 - 2015/07/07

Added
  • Support for authentication (Password Authenticator).

0.1.0 - 2015/07/03

Initial release. Forked from jbochi/lua-resty-cassandra v0.5.7 with some additional features and bug fixes.

Added
  • Support for both binary protocols v2 and v3.
Changed
  • More friendly support of auto pagination. THe loop doesn't require as many parameters.
  • OOP style in order to support both binary protocols. cassandra.new() must now be called with :.
Fixed
  • set_keyspace erroring on names with capital letters.