Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Sampling from model prior with missing is thread-unsafe #641

Closed
penelopeysm opened this issue Aug 19, 2024 · 3 comments
Closed

Sampling from model prior with missing is thread-unsafe #641

penelopeysm opened this issue Aug 19, 2024 · 3 comments
Assignees

Comments

@penelopeysm
Copy link
Member

Reported via Julia Slack: https://julialang.slack.com/archives/CCYDC34A0/p1723554415531969

MWE with current master branch of Turing:

using Turing 

@model function demo_threading(x)
    s ~ InverseGamma(2, 3)
    m ~ Normal(0, sqrt(s))
    Threads.@threads for i in eachindex(x)
        x[i] ~ Normal(m, sqrt(s))
    end
    return (s, m)
end

demo_threading(fill(missing, 1000))()

The last line of this usually crashes, the error message tends to vary. Sometimes it's destination has fewer elements than required, sometimes it's KeyError: key x[1] not found, etc. and sometimes it actually runs correctly!

Everything behaves as expected if you don't run Julia with multiple threads (-t4 argument below) or if you remove Threads.@threads from the above.

$ julia --project=. -t4

julia> using Turing

julia> @model function demo_threading(x)
           s ~ InverseGamma(2, 3)
           m ~ Normal(0, sqrt(s))
           Threads.@threads for i in eachindex(x)
               x[i] ~ Normal(m, sqrt(s))
           end
           return (s, m)
       end
demo_threading (generic function with 2 methods)

julia> demo_threading(fill(missing, 1000))()
ERROR: TaskFailedException

    nested task error: ArgumentError: destination has fewer elements than required
    Stacktrace:
      [1] copyto!(dest::Vector{AbstractPPL.VarName}, src::Base.KeySet{AbstractPPL.VarName, Dict{AbstractPPL.VarName, Int64}})
        @ Base ./abstractarray.jl:944
      [2] _collect
        @ ./array.jl:765 [inlined]
      [3] collect
        @ ./array.jl:759 [inlined]
      [4] istrans
        @ ~/.julia/packages/DynamicPPL/DvdZw/src/abstract_varinfo.jl:529 [inlined]
      [5] assume(rng::Random.TaskLocalRNG, sampler::DynamicPPL.SampleFromPrior, dist::Normal{…}, vn::AbstractPPL.VarName{…}, vi::DynamicPPL.ThreadSafeVarInfo{…})
        @ DynamicPPL ~/.julia/packages/DynamicPPL/DvdZw/src/context_implementations.jl:254
      [6] tilde_assume
        @ ~/.julia/packages/DynamicPPL/DvdZw/src/context_implementations.jl:72 [inlined]
      [7] tilde_assume
        @ ~/.julia/packages/DynamicPPL/DvdZw/src/context_implementations.jl:67 [inlined]
      [8] tilde_assume
        @ ~/.julia/packages/DynamicPPL/DvdZw/src/context_implementations.jl:52 [inlined]
      [9] tilde_assume!!(context::DynamicPPL.SamplingContext{…}, right::Normal{…}, vn::AbstractPPL.VarName{…}, vi::DynamicPPL.ThreadSafeVarInfo{…})
        @ DynamicPPL ~/.julia/packages/DynamicPPL/DvdZw/src/context_implementations.jl:144
     [10] (::var"#3#threadsfor_fun#2"{var"#3#threadsfor_fun#1#3"{}})(tid::Int64; onethread::Bool)
        @ Main ./threadingconstructs.jl:215
     [11] #3#threadsfor_fun
        @ ./threadingconstructs.jl:182 [inlined]
     [12] (::Base.Threads.var"#1#2"{var"#3#threadsfor_fun#2"{var"#3#threadsfor_fun#1#3"{}}, Int64})()
        @ Base.Threads ./threadingconstructs.jl:154

...and 3 more exceptions.

Stacktrace:
  [1] threading_run(fun::var"#3#threadsfor_fun#2"{var"#3#threadsfor_fun#1#3"{}}, static::Bool)
    @ Base.Threads ./threadingconstructs.jl:172
  [2] macro expansion
    @ ./threadingconstructs.jl:220 [inlined]
  [3] demo_threading(__model__::DynamicPPL.Model{…}, __varinfo__::DynamicPPL.ThreadSafeVarInfo{…}, __context__::DynamicPPL.SamplingContext{…}, x::Vector{…})
    @ Main ./REPL[2]:4
  [4] _evaluate!!
    @ ~/.julia/packages/DynamicPPL/DvdZw/src/model.jl:973 [inlined]
  [5] evaluate_threadsafe!!(model::DynamicPPL.Model{…}, varinfo::DynamicPPL.UntypedVarInfo{…}, context::DynamicPPL.SamplingContext{…})
    @ DynamicPPL ~/.julia/packages/DynamicPPL/DvdZw/src/model.jl:962
  [6] evaluate!!
    @ ~/.julia/packages/DynamicPPL/DvdZw/src/model.jl:892 [inlined]
  [7] evaluate!! (repeats 2 times)
    @ ~/.julia/packages/DynamicPPL/DvdZw/src/model.jl:905 [inlined]
  [8] evaluate!!
    @ ~/.julia/packages/DynamicPPL/DvdZw/src/model.jl:915 [inlined]
  [9] (::DynamicPPL.Model{typeof(demo_threading), (:x,), (), (), Tuple{Vector{…}}, Tuple{}, DynamicPPL.DefaultContext})()
    @ DynamicPPL ~/.julia/packages/DynamicPPL/DvdZw/src/model.jl:865
 [10] top-level scope
    @ REPL[3]:1
Some type information was truncated. Use `show(err)` to see complete types.
@penelopeysm
Copy link
Member Author

penelopeysm commented Aug 19, 2024

Oh, I just noticed that @torfjelde wrote:

Note that this only works for observe-statements, i.e. when the LHS of ~ is “fixed” / not random.

Is the usecase above something we want to support, or is this a wontfix?

@yebai
Copy link
Member

yebai commented Aug 21, 2024

We might be able to support this particular case. Can you investigate why it causes crashes so we can discuss this in more detail?

@penelopeysm penelopeysm self-assigned this Sep 18, 2024
@penelopeysm
Copy link
Member Author

(6 months later) I think this is a wontfix because passing fill(missing, 1000) to the model evaluator makes x an assumption rather than an observation, and ThreadSafeVarInfo doesn't work with assume, it only works with observe (assume has to add entries to the VarInfo, whereas observe only needs to increment logp).

Fixing this probably would require substantially expanding ThreadSafeVarInfo and it's probably not happening because parallel evaluation inside a model is pretty low priority (imo)

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

No branches or pull requests

2 participants