expirationd - data expiration with custom quirks.

This package can turn Tarantool into a persistent memcache replacement, but is powerful enough so that your own expiration strategy can be defined.

You define two functions: one takes a tuple as an input and returns true in case it's expired and false otherwise. The other takes the tuple and performs the expiry itself: either deletes it (memcache), or does something smarter, like put a smaller representation of the data being deleted into some other space.

There are a number of similar modules:

• moonwalker triggered manually, useful for batch transactions, a performance about 600/700k rec/sec
• expirationd always expires tuples with using indices and using any condition, without guarantee for time expiration.
• indexpirationd always expires tuples with indices, has a nice precision (up to ms) for time to expire.

Table below may help you to choose a proper module for your requirements:

Module	Reaction time	Uses indices	Arbitrary condition	Expiration trigger
indexpiration	High (ms)	Yes	No	synchronous (fiber with condition)
expirationd	Medium (sec)	Yes	Yes	synchronous (fiber with condition)
moonwalker	NA	No	Yes	asynchronous (using crontab etc)

Prerequisites

• Tarantool 1.10+ (tarantool package, see documentation).

Installation

You can:

• Install the module using tarantoolctl:

  tarantoolctl rocks install expirationd

• Install the module using LuaRocks:

  luarocks install --local --server=https://rocks.tarantool.org expirationd

Documentation

See API documentation in https://tarantool.github.io/expirationd/

Note about using expirationd with replication: by default expirationd processes tasks for all types of spaces only on the writable instance. It does not process tasks on read-only instance for non-local persistent spaces. It means that expirationd will not start task processing on a replica for regular spaces. One can force running task on replica with option force in start() module function. The option force let a user control where to start task processing and where don't.

Examples

Simple version:

box.cfg{}
space = box.space.old
job_name = "clean_all"
expirationd = require("expirationd")

function is_expired(args, tuple)
  return true
end

function delete_tuple(space, args, tuple)
  box.space[space]:delete{tuple[1]}
end

expirationd.start(job_name, space.id, is_expired, {
    process_expired_tuple = delete_tuple,
    args = nil,
    tuples_per_iteration = 50,
    full_scan_time = 3600
})

Сustomized version:

expirationd.start(job_name, space.id, is_expired, {
    -- name or id of the index in the specified space to iterate over
    index = "exp",
    -- one transaction per batch
    -- default is false
    atomic_iteration = true,
    -- delete data that was added a year ago
    -- default is nil
    start_key = function( task )
        return clock.time() - (365*24*60*60)
    end,
    -- delete it from the oldest to the newest
    -- default is ALL
    iterator_type = "GE",
    -- stop full_scan if delete a lot
    -- returns true by default
    process_while = function( task )
        if task.args.max_expired_tuples >= task.expired_tuples_count then
            task.expired_tuples_count = 0
            return false
        end
        return true
    end,
    -- this function must return an iterator over the tuples
    iterate_with = function( task )
        return task.index:pairs({ task.start_key() }, { iterator = task.iterator_type })
            :take_while( function( tuple )
                return task:process_while()
            end )
    end,
    args = {
        max_expired_tuples = 1000
    }
})

Testing

$ make deps-full
$ make test

Regression tests running in continuous integration that uses luatest are executed in shuffle mode. It means that every time order of tests is pseudorandom with predefined seed. If tests in CI are failed it is better to reproduce these failures with the same seed:

$ make SEED=1334 test
luatest -v --coverage --shuffle all:1334
...

Cartridge role

cartridge.roles.expirationd is a Tarantool Cartridge role for the expirationd package with features:

• It registers expirationd as a Tarantool Cartridge service for easy access to all API calls:

  local task = cartridge.service_get('expirationd').start("task_name", id, is_expired)
  task:kill()

• You could configure the expirationd role with cfg entry. expirationd.cfg() has the same parameters with the same meaning.

  Be careful, values from the clusterwide configuration are applied by default to all nodes on each apply_config(). Changing the configuration manually with expirationd.cfg() only affects the current node and does not update values in the clusterwide configuration. The manual change will be overwritten by a next apply_config call.

• The role stops all expirationd tasks on an instance on the role termination.

• The role can automatically start or kill old tasks from the role configuration:

  expirationd:
    cfg:
      metrics: true
    task_name1:
      space: 579
      is_expired: is_expired_func_name_in__G
      is_master_only: true
      options:
        args:
        - any
        atomic_iteration: false
        force: false
        force_allow_functional_index: true
        full_scan_delay: 1
        full_scan_time: 1
        index: 0
        iterate_with: iterate_with_func_name_in__G
        iteration_delay: 1
        iterator_type: ALL
        on_full_scan_complete: on_full_scan_complete_func_name_in__G
        on_full_scan_error: on_full_scan_error_func_name_in__G
        on_full_scan_start: on_full_scan_start_func_name_in__G
        on_full_scan_success: on_full_scan_success_func_name_in__G
        process_expired_tuple: process_expired_tuple_func_name_in__G
        process_while: process_while_func_name_in__G
        start_key:
        - 1
        tuples_per_iteration: 100
        vinyl_assumed_space_len: 100
        vinyl_assumed_space_len_factor: 1
    task_name2:
      ...

  expirationd.start() has the same parameters with the same meaning except for the additional optional param is_master_only. If true, the task should run only on a master instance. By default, the value is false.

  You need to be careful with parameters-functions. The string is a key in the global variable _G, the value must be a function. You need to define the key before initializing the role:

  rawset(_G, "is_expired_func_name_in__G", function(args, tuple)
      -- code of the function
  end)