Module fiber

Overview

With the fiber module, you can:

• Create, run, and manage fibers.
• Send and receive messages between different processes (i.e. different connections, sessions, or fibers) via channels.
• Use a synchronization mechanism for fibers, similar to “condition variables” and similar to operating-system functions, such as pthread_cond_wait() plus pthread_cond_signal().

Index

Below is a list of all fiber functions and members.

Fibers

A fiber is a set of instructions which are executed with cooperative multitasking. (Learn more about transactions in Tarantool.) Fibers managed by the fiber module are associated with a user-supplied function called the fiber function.

A fiber has three possible states, detectable by fiber.status(): running, suspended or dead. When a fiber is created with fiber.create(), it is running. When a fiber is created with fiber.new() or yields control with fiber.sleep(), it is suspended. When a fiber ends (because the fiber function ends), it is dead.

All fibers are part of the fiber registry. This registry can be searched with fiber.find() - via fiber id (fid), which is a numeric identifier.

A runaway fiber can be stopped with fiber_object.cancel. However, fiber_object.cancel is advisory — it works only if the runaway fiber calls fiber.testcancel() occasionally. Most box.* functions, such as box.space…delete() or box.space…update(), do call fiber.testcancel() but box.space…select{} does not. In practice, a runaway fiber can only become unresponsive if it does many computations and does not check whether it has been cancelled.

Like all Lua objects, dead fibers are garbage collected. The Lua garbage collector frees pool allocator memory owned by the fiber, resets all fiber data, and returns the fiber (now called a fiber carcass) to the fiber pool. The carcass can be reused when another fiber is created.

A fiber has all the features of a Lua coroutine and all the programming concepts that apply for Lua coroutines will apply for fibers as well. However, Tarantool has made some enhancements for fibers and has used fibers internally. So, although use of coroutines is possible and supported, use of fibers is recommended.

fiber.create(function[, function-arguments])¶

Create and start a fiber. The fiber is created and begins to run immediately.

Parameters:	function – the function to be associated with the fiber function-arguments – what will be passed to function
Return:	created fiber object
Rtype:	userdata

Example:

tarantool> fiber = require('fiber')
---
...
tarantool> function function_name()
         >   print("I'm a fiber")
         > end
---
...
tarantool> fiber_object = fiber.create(function_name); print("Fiber started")
I'm a fiber
Fiber started
---
...

fiber.new(function[, function-arguments])¶

Create but do not start a fiber: the fiber is created but does not begin to run immediately – it starts after the fiber creator (that is, the job that is calling fiber.new()) yields, under transaction control. The initial fiber state is ‘suspended’. Thus fiber.new() differs slightly from fiber.create().

Ordinarily fiber.new() is used in conjunction with fiber_object:set_joinable() and fiber_object:join().

Parameters:	function – the function to be associated with the fiber function-arguments – what will be passed to function
Return:	created fiber object
Rtype:	userdata

Example:

tarantool> fiber = require('fiber')
---
...
tarantool> function function_name(arg)
         >   print("I'm a fiber, " .. arg)
         > end
---
...
tarantool> fiber_object = fiber.new(function_name, 'yay!'); print("Fiber not started yet")
Fiber not started yet
---
...
tarantool> I'm a fiber, yay!
---
...

fiber.self()¶

Return:	fiber object for the currently scheduled fiber.
Rtype:	userdata

Example:

tarantool> fiber.self()
---
- status: running
  name: interactive
  id: 101
...

fiber.find(id)¶

Parameters:	id – numeric identifier of the fiber.
Return:	fiber object for the specified fiber.
Rtype:	userdata

Example:

tarantool> fiber.find(101)
---
- status: running
  name: interactive
  id: 101
...

fiber.sleep(time)¶

Yield control to the scheduler and sleep for the specified number of seconds. Only the current fiber can be made to sleep.

Parameters:	time – number of seconds to sleep.
Exception:	see the Example of yield failure.

Example:

tarantool> fiber.sleep(1.5)
---
...

fiber.yield()¶

Yield control to the scheduler. Equivalent to fiber.sleep(0).

Exception:	see the Example of yield failure.

Example:

tarantool> fiber.yield()
---
...

fiber.status([fiber_object])¶

Return the status of the current fiber. Or, if optional fiber_object is passed, return the status of the specified fiber.

Return:	the status of fiber. One of: “dead”, “suspended”, or “running”.
Rtype:	string

Example:

tarantool> fiber.status()
---
- running
...

fiber.info({[backtrace/bt]})¶

Return information about all fibers.

Parameters:	backtrace (boolean) – show backtrace. Default: true. Set to false to show less information (symbol resolving can be expensive). bt (boolean) – same as backtrace, but with lower priority.
Return:	number of context switches (csw), backtrace, total memory, used memory, fiber id (fid), fiber name. If fiber.top is enabled or Tarantool was built with ENABLE_FIBER_TOP, processor time (time) is also returned.
Rtype:	table

Return values explained

• csw – number of context switches.
• backtrace, bt – each fiber’s stack trace, showing where it originated and what functions were called.
• memory:
  • total – total memory occupied by the fiber as a C structure, its stack, etc.
  • used – actual memory used by the fiber.

• time – duplicates the “time” entry from fiber.top().cpu for each fiber.

  Only shown if fiber.top is enabled.

Example:

tarantool> fiber.info({ bt = true })
---
- 101:
    csw: 1
    backtrace:
    - C: '#0  0x5dd130 in lbox_fiber_id+96'
    - C: '#1  0x5dd13d in lbox_fiber_stall+13'
    - L: stall in =[C] at line -1
    - L: (unnamed) in @builtin/fiber.lua at line 59
    - C: '#2  0x66371b in lj_BC_FUNCC+52'
    - C: '#3  0x628f28 in lua_pcall+120'
    - C: '#4  0x5e22a8 in luaT_call+24'
    - C: '#5  0x5dd1a9 in lua_fiber_run_f+89'
    - C: '#6  0x45b011 in fiber_cxx_invoke(int (*)(__va_list_tag*), __va_list_tag*)+17'
    - C: '#7  0x5ff3c0 in fiber_loop+48'
    - C: '#8  0x81ecf4 in coro_init+68'
    memory:
    total: 516472
    used: 0
    time: 0
    name: lua
    fid: 101
  102:
    csw: 0
    backtrace:
    - C: '#0  (nil) in +63'
    - C: '#1  (nil) in +63'
    memory:
    total: 516472
    used: 0
    time: 0
    name: on_shutdown
    fid: 102

...

fiber.top()¶

Show all alive fibers and their CPU consumption.

Return:	a table with two entries: cpu and cpu_misses

cpu itself is a table whose keys are strings containing fiber ids and names. The three metrics available for each fiber are:

1. instant (in per cent), which indicates the share of time the fiber was executing during the previous event loop iteration.

2. average (in per cent), which is calculated as an exponential moving average of instant values over all the previous event loop iterations.

3. time (in seconds), which estimates how much CPU time each fiber spent processing during its lifetime.

   The time entry is also added to each fiber’s output in fiber.info() (it duplicates the time entry from fiber.top().cpu per fiber).

   Note that time is only counted while fiber.top() is enabled.

cpu_misses indicates the number of times the TX thread detected it was rescheduled on a different CPU core during the last event loop iteration. fiber.top() uses the CPU timestamp counter to measure each fiber’s execution time. However, each CPU core may have its own counter value (you can only rely on counter deltas if both measurements were taken on the same core, otherwise the delta may even get negative). When the TX thread is rescheduled to a different CPU core, Tarantool just assumes the CPU delta was zero for the latest measurement. This lowers the precision of our computations, so the bigger cpu misses value the lower the precision of fiber.top() results.

Let’s take a look at the example:

Example:

tarantool> fiber.top()
---
- cpu:
    107/lua:
      instant: 30.967324490456
      time: 0.351821993
      average: 25.582738345233
    104/lua:
      instant: 9.6473633128437
      time: 0.110869897
      average: 7.9693406131877
    101/on_shutdown:
      instant: 0
      time: 0
      average: 0
    103/lua:
      instant: 9.8026528631511
      time: 0.112641118
      average: 18.138387232255
    106/lua:
      instant: 20.071174377224
      time: 0.226901357
      average: 17.077908441831
    102/interactive:
      instant: 0
      time: 9.6858e-05
      average: 0
    105/lua:
      instant: 9.2461986412164
      time: 0.10657528
      average: 7.7068458630827
    1/sched:
      instant: 20.265286315108
      time: 0.237095335
      average: 23.141537169257
  cpu_misses: 0
...

Notice that by default new fibers created due to fiber.create are named ‘lua’ so it is better to set their names explicitly via fiber_object:name(‘name’).

There are several system fibers in fiber.top() output that might be useful:

• sched is a special system fiber. It schedules tasks to other fibers, if any, and also handles some libev events.

  It can have high instant and average values in fiber.top() output in two cases:

  • The instance has almost no load - then practically only sched is executing, and the other fibers are sleeping. So relative to the other fibers, sched may have almost 100% load.
  • sched handles a large number of system events. This should not cause performance problems.

• main fibers process requests that come over the network (iproto requests). There are several such fibers, and new ones are created if needed. When a new request comes in, a free fiber takes it and executes it. The request can be a typical select/replace/delete/insert or a function call. For example, conn:eval() or conn:call().

Note

Enabling fiber.top() slows down fiber switching by about 15%, so it is disabled by default. To enable it, say fiber.top_enable(). To disable it after you finished debugging, with fiber.top_disable().

fiber.kill(id)¶

Locate a fiber by its numeric id and cancel it. In other words, fiber.kill() combines fiber.find() and fiber_object:cancel().

Parameters:	id – the id of the fiber to be cancelled.
Exception:	the specified fiber does not exist or cancel is not permitted.

Example:

tarantool> fiber.kill(fiber.id()) -- kill self, may make program end
---
- error: fiber is cancelled
...

fiber.testcancel()¶

Check if the current fiber has been cancelled and throw an exception if this is the case.

Note

Even if you catch the exception, the fiber will remain cancelled. Most types of calls will check fiber.testcancel(). However, some functions (id, status, join etc.) will return no error. We recommend application developers to implement occasional checks with fiber.testcancel() and to end fiber’s execution as soon as possible in case it has been cancelled.

Example:

tarantool> fiber.testcancel()
---
- error: fiber is cancelled
...

fiber.time()¶

Return:	current system time (in seconds since the epoch) as a Lua number. The time is taken from the event loop clock, which makes this call very cheap, but still useful for constructing artificial tuple keys.
Rtype:	number

Example:

tarantool> fiber.time(), fiber.time()
---
- 1448466279.2415
- 1448466279.2415
...

fiber.time64()¶

Return:	current system time (in microseconds since the epoch) as a 64-bit integer. The time is taken from the event loop clock.
Rtype:	cdata (ctype<int64_t>)

Example:

tarantool> fiber.time(), fiber.time64()
---
- 1448466351.2708
- 1448466351270762
...

fiber.clock()¶

Get the monotonic time in seconds. It is better to use fiber.clock() for calculating timeouts instead of fiber.time() because fiber.time() reports real time so it is affected by system time changes.

Return:	a floating-point number of seconds, representing elapsed wall-clock time since some time in the past that is guaranteed not to change during the life of the process
Rtype:	number

Example:

tarantool> start = fiber.clock()
---
...
tarantool> print(start)
248700.58805
---
...
tarantool> print(fiber.time(), fiber.time()-start)
1600785979.8291 1600537279.241
---
...

fiber.clock64()¶

Same as fiber.clock() but in microseconds.

Return:	a number of seconds as 64-bit integer, representing elapsed wall-clock time since some time in the past that is guaranteed not to change during the life of the process
Rtype:	cdata (ctype<int64_t>)

object fiber_object¶

fiber_object:id()¶

Parameters:	fiber_object – generally this is an object referenced in the return from fiber.create or fiber.self or fiber.find
Return:	id of the fiber.
Rtype:	number

fiber.self():id() can also be expressed as fiber.id().

Example:

tarantool> fiber_object = fiber.self()
---
...
tarantool> fiber_object:id()
---
- 101
...

fiber_object:name()¶

Parameters:	fiber_object – generally this is an object referenced in the return from fiber.create or fiber.self or fiber.find
Return:	name of the fiber.
Rtype:	string

fiber.self():name() can also be expressed as fiber.name().

Example:

tarantool> fiber.self():name()
---
- interactive
...

fiber_object:name(name[, options])

Change the fiber name. By default a Tarantool server’s interactive-mode fiber is named ‘interactive’ and new fibers created due to fiber.create are named ‘lua’. Giving fibers distinct names makes it easier to distinguish them when using fiber.info and fiber.top(). Max length is 255.

Parameters:

fiber_object – generally this is an object referenced in the return from fiber.create or fiber.self or fiber.find name (string) – the new name of the fiber. options – truncate=true – truncates the name to the max length if it is too long. If this option is false (the default), fiber.name(new_name) fails with an exception if a new name is too long. The name length limit is 255 (since version 2.4.1).

Return:

nil

Example:

tarantool> fiber.self():name('non-interactive')
---
...

fiber_object:status()¶

Return the status of the specified fiber.

Parameters:	fiber_object – generally this is an object referenced in the return from fiber.create or fiber.self or fiber.find
Return:	the status of fiber. One of: “dead”, “suspended”, or “running”.
Rtype:	string

fiber.self():status( can also be expressed as fiber.status().

Example:

tarantool> fiber.self():status()
---
- running
...

fiber_object:cancel()¶

Send a cancellation request to the fiber. Running and suspended fibers can be cancelled. After a fiber has been cancelled, attempts to operate on it will cause errors, for example fiber_object:name() will cause error: the fiber is dead. But a dead fiber can still report its id and status.

Cancellation is asynchronous. Use fiber_object:join() to wait for the cancellation to complete. After fiber_object:cancel() is called, the fiber may or may not check whether it was cancelled. If the fiber does not check it, it cannot ever be cancelled.

Parameters:	fiber_object – generally this is an object referenced in the return from fiber.create or fiber.self or fiber.find
Return:	nil

Possible errors: cancel is not permitted for the specified fiber object.

Example:

tarantool> fiber.self():cancel() -- kill self, may make program end
---
...
tarantool> fiber.self():cancel()
---
- error: fiber is cancelled
...
tarantool> fiber.self:id()
---
- 163
...
tarantool> fiber.self:status()
---
- dead
...

fiber_object.storage¶

Local storage within the fiber. It is a Lua table created when it is first accessed. The storage can contain any number of named values, subject to memory limitations. Naming may be done with fiber_object.storage.name or fiber_object.storage['name']. or with a number fiber_object.storage[number]. Values may be either numbers or strings.

fiber.storage is destroyed when the fiber is finished, regardless of how is it finished – via fiber_object:cancel(), or the fiber’s function did ‘return’. Moreover, the storage is cleaned up even for pooled fibers used to serve IProto requests. Pooled fibers never really die, but nonetheless their storage is cleaned up after each request. That makes possible to use fiber.storage as a full featured request-local storage. This behavior is implemented in versions 2.2.3, 2.3.2, 2.4.1, and all later versions.

This storage may be created for a fiber, no matter how the fiber itself was created – from C or from Lua. For example, a fiber can be created in C using fiber_new(), then it can insert into a space, which has Lua on_replace triggers, and one of the triggers can create fiber.storage. That storage will be deleted when the fiber is stopped.

Example:

tarantool> fiber = require('fiber')
---
...
tarantool> function f () fiber.sleep(1000); end
---
...
tarantool> fiber_function = fiber.create(f)
---
...
tarantool> fiber_function.storage.str1 = 'string'
---
...
tarantool> fiber_function.storage['str1']
---
- string
...
tarantool> fiber_function:cancel()
---
...
tarantool> fiber_function.storage['str1']
---
- error: '[string "return fiber_function.storage[''str1'']"]:1: the fiber is dead'
...

See also box.session.storage.

fiber_object:set_joinable(true_or_false)¶

fiber_object:set_joinable(true) makes a fiber joinable; fiber_object:set_joinable(false) makes a fiber not joinable; the default is false.

A joinable fiber can be waited for, with fiber_object:join().

Best practice is to call fiber_object:set_joinable() before the fiber function begins to execute, because otherwise the fiber could become ‘dead’ before fiber_object:set_joinable() takes effect. The usual sequence could be:

1. Call fiber.new() instead of fiber.create() to create a new fiber_object.

   Do not yield at this point, because that will cause the fiber function to begin.

2. Call fiber_object:set_joinable(true) to make the new fiber_object joinable.

   Now it is safe to yield.

3. Call fiber_object:join().

   Usually fiber_object:join() should be called, otherwise the fiber’s status may become ‘suspended’ when the fiber function ends, instead of ‘dead’.

Parameters:	true_or_false – the boolean value that changes the set_joinable flag
Return:	nil

Example:

The result of the following sequence of requests is:

• the global variable d will be 6 (which proves that the function was not executed until after d was set to 1, when fiber.sleep(1) caused a yield);
• fiber.status(fi2) will be ‘suspended’ (which proves that after the function was executed the fiber status did not change to ‘dead’).

fiber=require('fiber')
d=0
function fu2() d=d+5 end
fi2=fiber.new(fu2) fi2:set_joinable(true) d=1 fiber.sleep(1)
print(d)
fiber.status(fi2)

fiber_object:join()¶

“Join” a joinable fiber. That is, let the fiber’s function run and wait until the fiber’s status is ‘dead’ (normally a status becomes ‘dead’ when the function execution finishes). Joining will cause a yield, therefore, if the fiber is currently in a suspended state, execution of its fiber function will resume.

This kind of waiting is more convenient than going into a loop and periodically checking the status; however, it works only if the fiber was created with fiber.new() and was made joinable with fiber_object:set_joinable().

Return:

two values. The first value is boolean. If the first value is true, then the join succeeded because the fiber’s function ended normally and the second result has the return value from the fiber’s function. If the first value is false, then the join succeeded because the fiber’s function ended abnormally and the second result has the details about the error, which one can unpack in the same way that one unpacks a pcall result.

Rtype:

boolean +result type, or boolean + struct error

Example:

The result of the following sequence of requests is:

• the first fiber.status() call returns ‘suspended’,
• the join() call returns true,
• the elapsed time is usually 5 seconds, and
• the second fiber.status() call returns ‘dead’.

This proves that the join() function blocks the execution of the fiber that called it until the fi2 fiber becomes ‘dead’.

fiber=require('fiber')
function fu2() fiber.sleep(5) end
fi2=fiber.new(fu2) fi2:set_joinable(true)
start_time = os.time()
fiber.status(fi2)
fi2:join()
print('elapsed = ' .. os.time() - start_time)
fiber.status(fi2)

tarantool> fiber = require('fiber')
tarantool> function function_x()
         >   gvar = 0
         >   while true do
         >     gvar = gvar + 1
         >     fiber.sleep(2)
         >   end
         > end
---
...

tarantool> gvar = 0

tarantool> fiber_of_x = fiber.create(function_x)
---
...

tarantool> fid = fiber_of_x:id()
---
...

tarantool> print('#', fid, '. ', fiber_of_x:status(), '. gvar=', gvar)
# 102 .  suspended . gvar= 399
---
...

tarantool> fiber_of_x:cancel()
---
...
tarantool> print('#', fid, '. ', fiber_of_x:status(), '. gvar=', gvar)
# 102 .  dead . gvar= 421
---
...

fiber = require('fiber')
function function_name()
  while true do
    print('before testcancel')
    fiber.testcancel()
    print('before yield')
    fiber.yield()
  end
end
fiber_object = fiber.create(function_name)
fiber.sleep(.1)
fiber_object:cancel()

Channels

Call fiber.channel() to create and get a new channel object.

Call the other routines, via channel, to send messages, receive messages, or check channel status.

Message exchange is synchronous. The Lua garbage collector will mark or free the channel when no one is using it, as with any other Lua object. Use object-oriented syntax, for example channel:put(message) rather than fiber.channel.put(message).

fiber.channel([capacity])¶

Create a new communication channel.

Parameters:	capacity (int) – the maximum number of slots (spaces for channel:put messages) that can be in use at once. The default is 0.
Return:	new channel object.
Rtype:	userdata. In the console output, it is serialized as channel: [number], where [number] is the return of channel_object:count().

object channel_object¶

channel_object:put(message[, timeout])¶

Send a message using a channel. If the channel is full, channel:put() waits until there is a free slot in the channel.

Note

The default channel capacity is 0. With this default value, channel:put() waits infinitely until channel:get() is called.

Parameters:	message (lua-value) – what will be sent, usually a string or number or table timeout (number) – maximum number of seconds to wait for a slot to become free. Default: infinity.
Return:	If timeout is specified, and there is no free slot in the channel for the duration of the timeout, then the return value is false. If the channel is closed, then the return value is false. Otherwise, the return value is true, indicating success.
Rtype:	boolean

channel_object:close()¶

Close the channel. All waiters in the channel will stop waiting. All following channel:get() operations will return nil, and all following channel:put() operations will return false.

channel_object:get([timeout])¶

Fetch and remove a message from a channel. If the channel is empty, channel:get() waits for a message.

Parameters:	timeout (number) – maximum number of seconds to wait for a message. Default: infinity.
Return:	If timeout is specified, and there is no message in the channel for the duration of the timeout, then the return value is nil. If the channel is closed, then the return value is nil. Otherwise, the return value is the message placed on the channel by channel:put().
Rtype:	usually string or number or table, as determined by channel:put

channel_object:is_empty()¶

Check whether the channel is empty (has no messages).

Return:	true if the channel is empty. Otherwise false.
Rtype:	boolean

channel_object:count()¶

Find out how many messages are in the channel.

Return:	the number of messages.
Rtype:	number

channel_object:is_full()¶

Check whether the channel is full.

Return:	true if the channel is full (the number of messages in the channel equals the number of slots so there is no room for a new message). Otherwise false.
Rtype:	boolean

channel_object:has_readers()¶

Check whether readers are waiting for a message because they have issued channel:get() and the channel is empty.

Return:	true if readers are waiting. Otherwise false.
Rtype:	boolean

channel_object:has_writers()¶

Check whether writers are waiting because they have issued channel:put() and the channel is full.

Return:	true if writers are waiting. Otherwise false.
Rtype:	boolean

channel_object:is_closed()¶

Return:	true if the channel is already closed. Otherwise false.
Rtype:	boolean

fiber = require('fiber')
channel = fiber.channel(10)
function consumer_fiber()
    while true do
        local task = channel:get()
        ...
    end
end

function consumer2_fiber()
    while true do
        -- 10 seconds
        local task = channel:get(10)
        if task ~= nil then
            ...
        else
            -- timeout
        end
    end
end

function producer_fiber()
    while true do
        task = box.space...:select{...}
        ...
        if channel:is_empty() then
            -- channel is empty
        end

        if channel:is_full() then
            -- channel is full
        end

        ...
        if channel:has_readers() then
            -- there are some fibers
            -- that are waiting for data
        end
        ...

        if channel:has_writers() then
            -- there are some fibers
            -- that are waiting for readers
        end
        channel:put(task)
    end
end

function producer2_fiber()
    while true do
        task = box.space...select{...}
        -- 10 seconds
        if channel:put(task, 10) then
            ...
        else
            -- timeout
        end
    end
end

Condition variables

Call fiber.cond() to create a named condition variable, which will be called ‘cond’ for examples in this section.

Call cond:wait() to make a fiber wait for a signal via a condition variable.

Call cond:signal() to send a signal to wake up a single fiber that has executed cond:wait().

Call cond:broadcast() to send a signal to all fibers that have executed cond:wait().

fiber.cond()¶

Create a new condition variable.

Return:	new condition variable.
Rtype:	Lua object

object cond_object¶

cond_object:wait([timeout])¶

Make the current fiber go to sleep, waiting until another fiber invokes the signal() or broadcast() method on the cond object. The sleep causes an implicit fiber.yield().

Parameters:	timeout – number of seconds to wait, default = forever.
Return:	If timeout is provided, and a signal doesn’t happen for the duration of the timeout, wait() returns false. If a signal or broadcast happens, wait() returns true.
Rtype:	boolean

cond_object:signal()¶

Wake up a single fiber that has executed wait() for the same variable. Does not yield.

Rtype:	nil

cond_object:broadcast()¶

Wake up all fibers that have executed wait() for the same variable. Does not yield.

Rtype:	nil

$ tarantoolctl connect '3301'
tarantool> fiber = require('fiber')
tarantool> cond = fiber.cond()
tarantool> cond:wait()

$ tarantoolctl connect '3301'
tarantool> cond:signal()