# Why does logging (@info) create a race condition?
**URL:** https://discourse.julialang.org/t/why-does-logging-info-create-a-race-condition/118428
**Category:** General Usage
**Created:** [August 20, 2024, 7:47pm UTC](https://discourse.julialang.org/t/why-does-logging-info-create-a-race-condition/118428 "2024-08-20T19:47:11Z")
**Posts on this page:** 3
**Page:** 1
### Author:  [@evanfields](https://discourse.julialang.org/u/evanfields)
#### Post date: [August 20, 2024, 7:47pm UTC](https://discourse.julialang.org/t/why-does-logging-info-create-a-race-condition/118428/1 "2024-08-20T19:47:11Z")
I have a directory with compressed CSVs that I’m reading in parallel. Roughly like so, simplifying away domain specific details not relevant to the question:
```julia
function munge_csv(stream)
...read a CSV with CSV.read, munge it, return...
end
function read_directory(dir)
tasks = map(readdir(dir; join = true)) do filepath
Threads.@spawn begin
df = open(ZstdDecompressorStream, filepath) do stream
return munge_csv(stream)
end
# this next lines producess a data race if included, wat?
@info "hi"
return df
end
end
dfs = fetch.(tasks)
@info "Finished fetching" n_dfs = length(dfs) df_sizes = sort(nrow.(dfs))
df = vcat(dfs...)
return df
end
```
With the flagged `@info "hi"` line, I seem to have a race condition; the final results are inconsistent and the sizes of the task-specific dataframes are inconsistent. Without that logging line, there’s either no data race or at the least it’s rare enough that I haven’t seen it in a few dozen repeat trials.
What’s going on here? Is the logging call actually creating a threading problem, or is that a red herring? Is something else about this construct inherently not threadsafe, and the logging just makes it worse?
---
### Author:  [@nhz2](https://discourse.julialang.org/u/nhz2)
#### Post date: [August 20, 2024, 7:57pm UTC](https://discourse.julialang.org/t/why-does-logging-info-create-a-race-condition/118428/2 "2024-08-20T19:57:47Z")
I would change:
```julia
df = open(ZstdDecompressorStream, filepath) do stream
return munge_csv(stream)
end
```
to:
```julia
local df = open(ZstdDecompressorStream, filepath) do stream
return munge_csv(stream)
end
```
to make sure `df` isn’t getting captured from an upper scope, when you do `df = vcat(dfs...)`.
---
### Author:  [@evanfields](https://discourse.julialang.org/u/evanfields)
#### Post date: [August 21, 2024, 2:03pm UTC](https://discourse.julialang.org/t/why-does-logging-info-create-a-race-condition/118428/3 "2024-08-21T14:03:10Z")
Ah, an embarrassing facepalm, thank you! I got got by `df` appearing later in a hard local scope than the first `do` block closures.