# Parallel worlds ages don't match

**URL:** <https://discourse.julialang.org/t/parallel-worlds-ages-dont-match/33496>\
**Category:** General Usage\
**Tags:** bug, parallel, multithreading, world\_age\
**Created:** [January 17, 2020, 5:34pm UTC](https://discourse.julialang.org/t/parallel-worlds-ages-dont-match/33496 "2020-01-17T17:34:07Z")\
**Posts on this page:** 6\
**Page:** 1

<div class="post-metadata">

**Author:** ![pbayer](https://sea2.discourse-cdn.com/julialang/user_avatar/discourse.julialang.org/pbayer/32/11675_2.png) [@pbayer](https://discourse.julialang.org/u/pbayer)\
**Post date:** [January 17, 2020, 5:34pm UTC](https://discourse.julialang.org/t/parallel-worlds-ages-dont-match/33496/1 "2020-01-17T17:34:07Z")

</div>

Im currently developing the multithreading in [`Simulate.jl`](https://github.com/pbayer/Simulate.jl). Now – with the current development version – I create a master clock and setup three parallel clocks:

```julia
using Simulate
clk = Clock()
multiply!(clk)

```

`got 3 threads parallel to master!`

I then create a test variable and an event with an anonymous function to increment it:

```julia
a = 1
ev = Simulate.SimEvent(SF(()->global a+=1),Main,1.0,0.0)

```

```julia
Simulate.SimEvent{SimFunction}(SimFunction(Main, var"#7#8"(), nothing, nothing), Main, 1.0, 0.0)

```

I can execute this and it does, what it should do. If I send this event over the event channel to the 1st parallel clock (on thread 2) and tell it to run, I get an **exception** telling me that it doesn’t know my newly created method `var"#7#8"`:

```julia
put!(clk.ac[1].ch, Simulate.Register(ev)) # I register the event to the clock on thread 2
take!(clk.ac[1].ch) # I get the response
put!(clk.ac[1].ch, Simulate.Run(10.0)) # I tell the clock to run

```

```julia
Simulate.Run(10.0)
clock 2 exception: MethodError(var"#7#8"(), (), 0x000000000000689e)

```

If I query the task on thread 2, I get the **following funny message** :

```julia
clk.ac[1].ref[]

```

```julia
Task (failed) @0x000000010de02f50
MethodError: no method matching (::var"#7#8")()
The applicable method may be too new: running in world age 26782, while current world is 26783.
Closest candidates are:
  #7() at In[3]:2 (method too new to be called from this world context.)

```

If I change the sequence, define first my method and then multiply to parallel threads, everything works fine:

```julia
clk = Clock();
a = 1
ev = Simulate.SimEvent(SF(()->global a+=1),Main,1.0,0.0)
multiply!(clk) # now I start everything on the other threads
put!(clk.ac[1].ch, Simulate.Register(ev))
take!(clk.ac[1].ch)
put!(clk.ac[1].ch, Simulate.Run(10.0))
take!(clk.ac[1].ch)

```

```julia
got 3 threads parallel to master!
Simulate.Response(10.0)

```

I query the task as before:

```julia
clk.ac[1].ref[]

```

`Task (runnable) @0x000000013adc0fd0`

Should the above failure happen? Is there a way to sync the “parallel worlds” on different threads to my world (on thread 1) in order to exchange newly created functions between threads?

Funny problem, isn’t it? 😉

---

<div class="post-metadata">

**Author:** ![pbayer](https://sea2.discourse-cdn.com/julialang/user_avatar/discourse.julialang.org/pbayer/32/11675_2.png) [@pbayer](https://discourse.julialang.org/u/pbayer)\
**Post date:** [January 17, 2020, 9:32pm UTC](https://discourse.julialang.org/t/parallel-worlds-ages-dont-match/33496/2 "2020-01-17T21:32:49Z")

</div>

after some thinking about this problem, I guess each task get’s its own “world” when it starts and keeps it for its lifetime. Is that so?

In this case I would have to start new tasks on each thread to get the parallel worlds synced. I don’t like that solution since it is expensive. Does anybody know how to start in Julia tasks on each threads without `@spawn`ing multiple ones and filtering out the hits?

Or is there an alternative way to do it?

---

<div class="post-metadata">

**Author:** ![anon92994695](https://avatars.discourse-cdn.com/v4/letter/a/ce7236/32.png) [@anon92994695](https://discourse.julialang.org/u/anon92994695)\
**Post date:** [January 17, 2020, 11:31pm UTC](https://discourse.julialang.org/t/parallel-worlds-ages-dont-match/33496/3 "2020-01-17T23:31:56Z")

</div>

I’m a little but of a novice in this parallel computing situation. But! I got this error once. I fixed it by making sure that all workers running the code had the same package state.

One trick I’ve found for doing this is to put a `run(scp -r ... ...)` statement in my code to remotely copy the latest version to the client address.

Not sure if that’ll help you

---

<div class="post-metadata">

**Author:** ![pbayer](https://sea2.discourse-cdn.com/julialang/user_avatar/discourse.julialang.org/pbayer/32/11675_2.png) [@pbayer](https://discourse.julialang.org/u/pbayer)\
**Post date:** [January 18, 2020, 7:10am UTC](https://discourse.julialang.org/t/parallel-worlds-ages-dont-match/33496/4 "2020-01-18T07:10:29Z")

</div>

The analog in tasks and multithreading would be to copy over the code and to evaluate it at execution time within the task. It works, but:

```julia
julia> using BenchmarkTools

julia> ev = Simulate.SimEvent(SF(()->global a+=1), Main, 1.0,0.0)
Simulate.SimEvent{SimFunction}(SimFunction(Main, var"#28#29"(), nothing, nothing), Main, 1.0, 0.0)

julia> @btime Simulate.evExec(ev.ex, ev.scope)
  114.546 ns (1 allocation: 16 bytes)
500503

julia> ev = Simulate.SimEvent(:(global a+=1), Main, 1.0,0.0)
Simulate.SimEvent{Expr}(:(global a += 1), Main, 1.0, 0.0)

julia> @btime Simulate.evExec(ev.ex, ev.scope)
┌ Warning: Evaluating expressions is slow, use `SimFunction` instead
└ @ Simulate ~/.julia/dev/Simulate/src/events.jl:35
  84.708 μs (45 allocations: 2.70 KiB)
54949

```

Alternatively I can use only predefined methods. But this is a restriction excluding to pass anonymous functions and macros to tasks which have been started before.

---

<div class="post-metadata">

**Author:** ![tkf](https://sea2.discourse-cdn.com/julialang/user_avatar/discourse.julialang.org/tkf/32/17635_2.png) [@tkf](https://discourse.julialang.org/u/tkf)\
**Post date:** [January 18, 2020, 7:30am UTC](https://discourse.julialang.org/t/parallel-worlds-ages-dont-match/33496/5 "2020-01-18T07:30:20Z")

</div>

I think you can wrap the code that may receive functions with `Base.invokelatest`. Something like

```julia
new_function = take!(channel)
Base.invokelatest(new_function) do new_function
    ... use new_function ...
end

```

There is a similar unsolved problem with the Logging module: [https://github.com/JuliaLang/julia/issues/33865](https://github.com/JuliaLang/julia/issues/33865)

---

<div class="post-metadata">

**Author:** ![pbayer](https://sea2.discourse-cdn.com/julialang/user_avatar/discourse.julialang.org/pbayer/32/11675_2.png) [@pbayer](https://discourse.julialang.org/u/pbayer)\
**Post date:** [January 18, 2020, 2:01pm UTC](https://discourse.julialang.org/t/parallel-worlds-ages-dont-match/33496/6 "2020-01-18T14:01:42Z")

</div>

edit [1]: now that works fine, thank you very much.

I included a try/catch block in my `sfExec`-function:

```julia
    try
        ....
    catch exc
        if exc isa MethodError
            if x.kw === nothing
                return x.arg === nothing ? invokelatest(x.efun) : invokelatest(x.efun, arg...)
            else
                return x.arg === nothing ? invokelatest(x.efun; kw...) : invokelatest(x.efun, arg...; kw...)
            end
        end
    end

```

[1] There was still an error in my first fix