Linux + MacOS | Windows |
---|---|
Effil is a multithreading library for Lua. It allows to spawn native threads and safe data exchange. Effil has been designed to provide clear and simple API for lua developers.
Effil supports lua 5.1, 5.2, 5.3 and LuaJIT. Requires C++14 compiler compliance. Tested with GCC 4.9+, clang 3.8 and Visual Studio 2015.
- How to install
- Quick guide
- Important notes
- Blocking and nonblocking operations
- Function's upvalues
- API Reference
git clone --recursive https://github.com/effil/effil effil
cd effil && mkdir build && cd build
cmake .. && make install
- Copy effil.so to your project.
luarocks install effil
As you may know there are not much script languages with real multithreading support (Lua/Python/Ruby and etc has global interpreter lock aka GIL). Effil solves this problem by running independent Lua VM instances in separate native threads and provides robust communicating primitives for creating threads and data sharing.
Effil library provides three major abstractions:
effil.thread
- provides API for threads management.effil.table
- provides API for tables management. Tables can be shared between threads.effil.channel
- provides First-In-First-Out container for sequential data exchange.
And bunch of utilities to handle threads and tables as well.
Spawn the thread
local effil = require("effil")
function bark(name)
print(name .. " barks from another thread!")
end
-- run funtion bark in separate thread with name "Spaky"
local thr = effil.thread(bark)("Sparky")
-- wait for completion
thr:wait()
Output:
Sparky barks from another thread!
Sharing data with effil.channel
local effil = require("effil")
-- channel allow to push data in one thread and pop in other
local channel = effil.channel()
-- writes some numbers to channel
local function producer(channel)
for i = 1, 5 do
print("push " .. i)
channel:push(i)
end
channel:push(nil)
end
-- read numbers from channels
local function consumer(channel)
local i = channel:pop()
while i do
print("pop " .. i)
i = channel:pop()
end
end
-- run producer
local thr = effil.thread(producer)(channel)
-- run consumer
consumer(channel)
thr:wait()
Output:
push 1
push 2
pop 1
pop 2
push 3
push 4
push 5
pop 3
pop 4
pop 5
Sharing data with effil.table
effil = require("effil")
-- effil.table transfers data between threads
-- and behaves like regualr lua table
local storage = effil.table { string_field = "first value" }
storage.numeric_field = 100500
storage.function_field = function(a, b) return a + b end
storage.table_field = { fist = 1, second = 2 }
function check_shared_table(storage)
print(storage.string_field)
print(storage.numeric_field)
print(storage.table_field.first)
print(storage.table_field.second)
return storage.function_field(1, 2)
end
local thr = effil.thread(check_shared_table)(storage)
local ret = thr:get()
print("Thread result: " .. ret)
Output:
first value
100500
1
2
Thread result: 3
Effil allows to transmit data between threads (Lua interpreter states) using effil.channel
, effil.table
or directly as parameters of effil.thread
.
- Primitive types are transmitted 'as is' by copy:
nil
,boolean
,number
,string
- Functions are dumped using
lua_dump
. Upvalues are captured according to the rules. - Userdata and Lua threads (coroutines) are not supported.
- Tables are serialized to
effil.table
recursively. So, any Lua table becomeseffil.table
. Table serialization may take a lot of time for big table. Thus, it's better to put data directly toeffil.table
avoiding a table serialization. Let's consider 2 examples:
-- Example #1
t = {}
for i = 1, 100 do
t[i] = i
end
shared_table = effil.table(t)
-- Example #2
t = effil.table()
for i = 1, 100 do
t[i] = i
end
In the example #1 we create regular table, fill it and convert it to effil.table
. In this case Effil needs to go through all table fields one more time. Another way is example #2 where we firstly created effil.table
and after that we put data directly to effil.table
. The 2nd way pretty much faster try to follow this principle.
All operations which use time metrics can be blocking or non blocking and use following API:
(time, metric)
where metric
is time interval like 's'
(seconds) and time
is a number of intervals.
Example:
thread:get()
- infinitely wait for thread completion.thread:get(0)
- non blocking get, just check is thread finished and returnthread:get(50, "ms")
- blocking wait for 50 milliseconds.
List of available time intervals:
ms
- milliseconds;s
- seconds (default);m
- minutes;h
- hours.
Working with functions Effil serializes and deserializes them using lua_dump
and lua_load
methods. All function's upvalues are stored following the same rules as usual. If function has upvalue of unsupported type this function cannot be transmitted to Effil. You will get error in that case.
Working with function Effil can store function environment (_ENV
) as well. Considering environment as a regular table Effil will store it in the same way as any other table. But it does not make sence to store global _G
, so there are some specific:
- Lua = 5.1: function environment is not stored at all (due to limitations of lua_setfenv we cannot use userdata)
- Lua > 5.1: Effil serialize and store function environment only if it's not equal to global environment (
_ENV ~= _G
).
effil.thread
is the way to create a thread. Threads can be stopped, paused, resumed and canceled.
All operation with threads can be synchronous (with optional timeout) or asynchronous.
Each thread runs with its own lua state.
Use effil.table
and effil.channel
to transmit data over threads.
See example of thread usage here.
Creates thread runner. Runner spawns new thread for each invocation.
input: func - Lua function
output: runner - thread runner object to configure and run a new thread
Allows to configure and run a new thread.
Run captured function with specified arguments in separate thread and returns thread handle.
input: Any number of arguments required by captured function.
output: Thread handle object.
Is a Lua package.path
value for new state. Default value inherits package.path
form parent state.
Is a Lua package.cpath
value for new state. Default value inherits package.cpath
form parent state.
Number of lua instructions lua between cancelation points (where thread can be stopped or paused). Default value is 200. If this values is 0 then thread uses only explicit cancelation points.
Thread handle provides API for interaction with thread.
Returns thread status.
output:
status
- string values describes status of thread. Possible values are:"running", "paused", "canceled", "completed" and "failed"
.err
- error message, if any. This value is specified only if thread status =="failed"
.stacktrace
- stacktrace of failed thread. This value is specified only if thread status =="failed"
.
Waits for thread completion and returns function result or nothing in case of error.
input: Operation timeout in terms of time metrics
output: Results of captured function invocation or nothing in case of error.
Waits for thread completion and returns thread status.
input: Operation timeout in terms of time metrics
output: Returns status of thread. The output is the same as thread:status()
Interrupts thread execution. Once this function was invoked 'cancellation' flag is set and thread can be stopped sometime in the future (even after this function call done). To be sure that thread is stopped invoke this function with infinite timeout. Cancellation of finished thread will do nothing and return true
.
input: Operation timeout in terms of time metrics
output: Returns true
if thread was stopped or false
.
Pauses thread. Once this function was invoked 'pause' flag is set and thread can be paused sometime in the future (even after this function call done). To be sure that thread is paused invoke this function with infinite timeout.
input: Operation timeout in terms of time metrics
output: Returns true
if thread was paused or false
. If the thread is completed function will return false
Resumes paused thread. Function resumes thread immediately if it was paused. This function does nothing for completed thread. Function has no input and output parameters.
Gives unique identifier.
output: returns unique string id
for current thread.
Explicit cancellation point. Function checks cancellation or pausing flags of current thread and if it's required it performs corresponding actions (cancel or pause thread).
Suspend current thread.
input: time metrics arguments.
Returns the number of concurrent threads supported by implementation. Basically forwards value from std::thread::hardware_concurrency. output: number of concurrent hardware threads.
effil.table
is a way to exchange data between effil threads. It behaves almost like standard lua tables.
All operations with shared table are thread safe. Shared table stores primitive types (number, boolean, string), function, table, light userdata and effil based userdata. Shared table doesn't store lua threads (coroutines) or arbitrary userdata. See examples of shared table usage here
Use Shared tables with regular tables. If you want to store regular table in shared table, effil will implicitly dump origin table into new shared table. Shared tables always stores subtables as shared tables.
Use Shared tables with functions. If you store function in shared table, effil implicitly dumps this function and saves it as string (and it's upvalues). All function's upvalues will be captured according to following rules .
Creates new empty shared table.
input: tbl
- is optional parameter, it can be only regular Lua table which entries will be copied to shared table.
output: new instance of empty shared table. It can be empty or not, depending on tbl
content.
Set a new key of table with specified value.
input:
key
- any value of supported type. See the list of supported typesvalue
- any value of supported type. See the list of supported types
Get a value from table with specified key.
input: key
- any value of supported type. See the list of supported types
output: value
- any value of supported type. See the list of supported types
Sets a new metatable to shared table. Similar to standard setmetatable.
input:
tbl
should be shared table for which you want to set metatable.mtbl
should be regular table or shared table which will become a metatable. If it's a regular table effil will create a new shared table and copy all fields ofmtbl
. Setmtbl
equal tonil
to delete metatable from shared table.
output: just returns tbl
with a new metatable value similar to standard Lua setmetatable method.
Returns current metatable. Similar to standard getmetatable
input: tbl
should be shared table.
output: returns metatable of specified shared table. Returned table always has type effil.table
. Default metatable is nil
.
Set table entry without invoking metamethod __newindex
. Similar to standard rawset
input:
tbl
is shared table.key
- key of table to override. The key can be of any supported type.value
- value to set. The value can be of any supported type.
output: returns the same shared table tbl
Gets table value without invoking metamethod __index
. Similar to standard rawget
input:
tbl
is shared table.key
- key of table to receive a specific value. The key can be of any supported type.
output: returns required value
stored under a specified key
Is a global predefined shared table. This table always present in any thread (any Lua state).
effil = require "effil"
function job()
effil = require "effil"
effil.G.key = "value"
end
effil.thread(job)():wait()
print(effil.G.key) -- will print "value"
Truns effil.table
into regular Lua table.
tbl = effil.table({})
effil.type(tbl) -- 'effil.table'
effil.type(effil.dump(tbl)) -- 'table'
effil.channel
is a way to sequentially exchange data between effil threads. It allows to push message from one thread and pop it from another. Channel's message is a set of values of supported types. All operations with channels are thread safe. See examples of channel usage here
Creates a new channel.
input: optional capacity of channel. If capacity
equals to 0
or to nil
size of channel is unlimited. Default capacity is 0
.
output: returns a new instance of channel.
Pushes message to channel.
input: any number of values of supported types. Multiple values are considered as a single channel message so one push to channel decreases capacity by one.
output: pushed
is equal to true
if value(-s) fits channel capacity, false
otherwise.
Pop message from channel. Removes value(-s) from channel and returns them. If the channel is empty wait for any value appearance.
input: waiting timeout in terms of time metrics (used only if channel is empty).
output: variable amount of values which were pushed by a single channel:push() call.
Get actual amount of messages in channel.
output: amount of messages in channel.
Effil provides custom garbage collector for effil.table
and effil.channel
(and functions with captured upvalues). It allows safe manage cyclic references for tables and channels in multiple threads. However it may cause extra memory usage. effil.gc
provides a set of method configure effil garbage collector. But, usually you don't need to configure it.
Garbage collector perform it's work when effil creates new shared object (table, channel and functions with captured upvalues).
Each iteration GC checks amount of objects. If amount of allocated objects becomes higher then specific threshold value GC starts garbage collecting. Threshold value is calculated as previous_count * step
, where previous_count
- amount of objects on previous iteration (100 by default) and step
is a numerical coefficient specified by user (2.0 by default).
For example: if GC step
is 2.0
and amount of allocated objects is 120
(left after previous GC iteration) then GC will start to collect garbage when amount of allocated objects will be equal to240
.
Each thread represented as separate Lua state with own garbage collector.
Thus, objects will be deleted eventually.
Effil objects itself also managed by GC and uses __gc
userdata metamethod as deserializer hook.
To force objects deletion:
- invoke standard
collectgarbage()
in all threads. - invoke
effil.gc.collect()
in any thread.
Force garbage collection, however it doesn't guarantee deletion of all effil objects.
Show number of allocated shared tables and channels.
output: returns current number of allocated objects. Minimum value is 1, effil.G
is always present.
Get/set GC memory step multiplier. Default is 2.0
. GC triggers collecting when amount of allocated objects growth in step
times.
input: new_value
is optional value of step to set. If it's nil
then function will just return a current value.
output: old_value
is current (if new_value == nil
) or previous (if new_value ~= nil
) value of step.
Pause GC. Garbage collecting will not be performed automatically. Function does not have any input or output
Resume GC. Enable automatic garbage collecting.
Get GC state.
output: return true
if automatic garbage collecting is enabled or false
otherwise. By default returns true
.
Returns number of entries in Effil object.
input: obj
is shared table or channel.
output: number of entries in shared table or number of messages in channel
Threads, channels and tables are userdata. Thus, type()
will return userdata
for any type. If you want to detect type more precisely use effil.type
. It behaves like regular type()
, but it can detect effil specific userdata.
input: obj
is object of any type.
output: string name of type. If obj
is Effil object then function returns string like effil.table
in other cases it returns result of lua_typename function.
effil.type(effil.thread()) == "effil.thread"
effil.type(effil.table()) == "effil.table"
effil.type(effil.channel()) == "effil.channel"
effil.type({}) == "table"
effil.type(1) == "number"