Skip to content

Latest commit

 

History

History
225 lines (153 loc) · 7.02 KB

CHANGELOG.md

File metadata and controls

225 lines (153 loc) · 7.02 KB

Changelog

All notable changes to this project will be documented in this file.

The format is based on Keep a Changelog and this project adheres to Semantic Versioning.

[Unreleased]

Added

Changed

Fixed

[1.4.1] - 2024-05-21

The release fixes too long release time by a ttr value or a delay timeout.

Fixed

  • Too long timings in the utubettl driver (#223).

[1.4.0] - 2024-05-20

The release introduces an experimental storage_mode option for the utube and utubettl drivers with the memtx engine. It could be used to create a utube or utubettl queue with an additional buffer space that stores only ready to take tasks.

local tube = queue.create_tube('utube_with_ready_buffer', 'utube',
    {storage_mode = queue.driver.utube.STORAGE_MODE_READY_BUFFER})
local tube = queue.create_tube('utubettl_with_ready_buffer', 'utubettl',
    {storage_mode = queue.driver.utubettl.STORAGE_MODE_READY_BUFFER})

The storage mode slower in general cases, but a much faster in cases when you have utubes with many tasks (see README.md for the performance comparison). So you should make your choice carefully.

Added

  • Experimental storage_mode option for creating a utube and utubettl tube (#228). It enables the workaround for slow takes while working with busy tubes.

Fixed

  • Stuck in INIT state if an instance failed to enter the running mode in time (#226). This fix works only for Tarantool versions >= 2.10.0.
  • Slow takes on busy utube and utubettl tubes (#228). The workaround could be enabled by passing the storage_mode = "ready_buffer" option while creating the tube.

[1.3.3] - 2023-09-13

Fixed

  • In replicaset mode, the behavior of the public API is reduced to the same behavior in all queue states, including INIT. Previously, in the INIT state, an ambiguous error was thrown when trying to access a public method on a replica and the script was interrupted by an error.

    Old behavior (call create_tube on replica, queue is in INIT state):

    2023-09-04 14:01:11.000 [5990] main/103/replica.lua/box.load_cfg I> set 'read_only' configuration option to true
    stack traceback:
      /home/void/tmp/cluster/repl/queue/init.lua:44: in function '__index'
      replica.lua:13: in main chunk
    2023-09-04 14:01:11.004 [5990] main/105/checkpoint_daemon I> scheduled next checkpoint for Mon Sep  4 15:11:32 2023
    2023-09-04 14:01:11.004 [5990] main utils.c:610 E> LuajitError: /home/void/tmp/cluster/repl/queue/init.lua:45: Please configure box.cfg{} in read/write mode first
    

    After this fix:

    2023-09-11 10:24:31.463 [19773] main/103/replica.lua abstract.lua:93 E> create_tube: queue is in INIT state
    

[1.3.2] - 2023-08-24

Fixed

  • Duplicate id error with mvvc on put and take (#207).

[1.3.1] - 2023-07-31

Fixed

  • Yield in the fifottl/utubettl queue drivers.

[1.3.0] - 2023-03-13

Added

  • Possibility to get the module version.

Fixed

  • Bug when working with the replicaset (#202).

[1.2.5] - 2023-02-28

This is a technical release that should fix several problems in the repositories with packages that were caused by the mistaken creation of several tags (for this reason there are no 1.2.3 and 1.2.4 tags and corresponding packages in the repositories).

[1.2.2] - 2022-11-03

Fixed

  • Excessive CPU consumption of the state fiber.

[1.2.1] - 2022-09-12

Added

  • Rockspec publishing to CD.

Fixed

  • "replication" mode switching on tarantool < 2.2.1

[1.2.0] - 2022-06-06

Added

Fixed

  • Work with "ttl" of buried task. Previously, if a task was "buried" after it was "taken" (and the task has a "ttr"), then the time when the "release" should occur (the "ttr" timer) will be interpreted as the end of the "ttl" timer and the task will be deleted.

[1.1.0] - 2020-12-25

Added

  • "Shared sessions" was added to the queue. Previously, a connection to server was synonym of the queue session. Now the session has a unique UUID (returned by the "queue.identify()" method), and one session can have many connections. To connect to an existing session, call "queue.identify(uuid)" with the previously obtained UUID.
  • Possibility to work with tasks after reconnect. The queue module now provides the ability to set the ttr setting for sessions by queue.cfg({ ttr = ttr_in_sec}), which characterizes how long the logical session will exist after all active connections are closed.

Fixed

  • Custom driver registration after reboot. Previously, if a custom driver is registered after calling box.cfg() it causes a problem when the instance will be restarted.

[1.0.8] - 2020-10-17

Added

  • The ability to start an instance with a loaded queue module in read-only mode. In this case, a start of the module will be delayed until the instance will be configured with read_only = false. Previously, when trying to initialize the queue module on an instance configured in ro mode, an error occurred: "ER_READONLY: Can't modify data because this instance is in read-only mode." See https://github.com/tarantool/queue#initialization for more details.

[1.0.7] - 2020-09-03

Breaking changes

  • A check for a driver API implementation was added. Now, the consumer will be informed about the missing methods in the driver implementation (an error will be thrown).

Added

  • Notification about missing methods in the driver implementation (#126).

Fixed

  • The tasks releases on start for some drivers (utubettl, fifottl, limfifottl). Before, an attempt to release "taken" tasks on start lead to an error: "attempt to index local 'opts' (a nil value)" (#121).

Changed

  • Updated the requirements for the "delete" driver method. Now, the method should change the state of a task to "done". Before, it was duplicated by external code.

[1.0.6] - 2020-02-29

Breaking changes

External drivers should be updated with the new tasks_by_state() method. Consider the example from the fifo driver:

-- get iterator to tasks in a certain state
function method.tasks_by_state(self, task_state)
    return self.space.index.status:pairs(task_state)
end

The example uses 'status' secondary index, which is built on top of 'status' and 'task_id' fields.

This new method is necessary to correctly manage state of tasks: when tarantool instance is restarted we should release all taken tasks (otherwise they would stuck in the taken state forever). See #66 and #126 for more information.

Fixed

  • Releasing tasks at a client disconnection (#103).
  • Optimize statistics build (-15% in some cases) (#92).
  • Release all taken tasks at start (#66).
  • Don't take tasks during a client disconnection (#104).