Closed StevenWhitaker closed 1 year ago
I also get a segfault if I use the master
branch of Dagger.jl and/or the main
branch of DTables.jl, and I also sometimes get a segfault (instead of a KeyError
) with the original example (i.e., with Dagger v0.18.2 and DTables v0.4.1).
Replacing the select
call in the remotecall_fetch
with reduce((x, y) -> x + y, dt_remote; cols = [:a, :b])
also results in a KeyError
. In this case, it is the actual call to reduce
(not fetch
) that causes the KeyError
.
Replacing the select
call with combine(dt_remote, [:a, :b] .=> sum; renamecols = false)
also results in a KeyError
when calling fetch
on the result.
Whew, great catch!
The following example produces a
KeyError
.Contents of
mwe.jl
:Results:
The code works if just one column is selected, e.g.,
select(dt_remote, 1)
.Also note that calling
fetch(dnew)
before theremotecall_fetch
produces the expectedDataFrame
on worker 1, but then theremotecall_fetch
segfaults: