From d52c2e60f4a5d8ea78f723b3fda01cc750e98ad5 Mon Sep 17 00:00:00 2001
From: KristofferC <kristoffer.carlsson@juliacomputing.com>
Date: Tue, 30 Sep 2025 13:58:14 +0200
Subject: [PATCH] add the option to disable the part of threading in
 `Arrow.Table` that leads to catastrophic negative scaling

---
 src/table.jl       | 25 +++++++++++++++++--------
 test/testtables.jl |  9 +++++++++
 2 files changed, 26 insertions(+), 8 deletions(-)

diff --git a/src/table.jl b/src/table.jl
index fe9206b4..8b9c210c 100644
--- a/src/table.jl
+++ b/src/table.jl
@@ -289,10 +289,10 @@ function Base.iterate(x::Stream, (pos, id)=(1, 0))
 end
 
 """
-    Arrow.Table(io::IO; convert::Bool=true)
-    Arrow.Table(file::String; convert::Bool=true)
-    Arrow.Table(bytes::Vector{UInt8}, pos=1, len=nothing; convert::Bool=true)
-    Arrow.Table(inputs::Vector; convert::Bool=true)
+    Arrow.Table(io::IO; convert::Bool=true, threaded::Bool=true)
+    Arrow.Table(file::String; convert::Bool=true, threaded::Bool=true)
+    Arrow.Table(bytes::Vector{UInt8}, pos=1, len=nothing; convert::Bool=true, threaded::Bool=true)
+    Arrow.Table(inputs::Vector; convert::Bool=true, threaded::Bool=true)
 
 Read an arrow formatted table, from:
  * `io`, bytes will be read all at once via `read(io)`
@@ -311,6 +311,8 @@ sink function: e.g. `DataFrame(Arrow.Table(file))`, `SQLite.load!(db, "table", A
 
 Supports the `convert` keyword argument which controls whether certain arrow primitive types will be
 lazily converted to more friendly Julia defaults; by default, `convert=true`.
+
+Set `threaded=false` to disable parallel batch processing, which can avoid lock contention for tables with many small batches.
 """
 struct Table <: Tables.AbstractColumns
     names::Vector{Symbol}
@@ -465,7 +467,7 @@ Table(inputs::Vector; kw...) =
     Table([ArrowBlob(tobytes(x), 1, nothing) for x in inputs]; kw...)
 
 # will detect whether we're reading a Table from a file or stream
-function Table(blobs::Vector{ArrowBlob}; convert::Bool=true)
+function Table(blobs::Vector{ArrowBlob}; convert::Bool=true, threaded::Bool=true)
     t = Table()
     sch = nothing
     dictencodingslockable = Lockable(Dict{Int64,DictEncoding}()) # dictionary id => DictEncoding
@@ -580,10 +582,17 @@ function Table(blobs::Vector{ArrowBlob}; convert::Bool=true)
             elseif header isa Meta.RecordBatch
                 anyrecordbatches = true
                 @debug "parsing record batch message: compression = $(header.compression)"
-                @wkspawn begin
+                if threaded
+                    @wkspawn begin
+                        cols = collect(
+                            VectorIterator(sch, $batch, dictencodingslockable, convert),
+                        )
+                        put!(() -> put!(tsks, cols), sync, $(rbi))
+                    end
+                else
                     cols =
-                        collect(VectorIterator(sch, $batch, dictencodingslockable, convert))
-                    put!(() -> put!(tsks, cols), sync, $(rbi))
+                        collect(VectorIterator(sch, batch, dictencodingslockable, convert))
+                    put!(tsks, cols)
                 end
                 rbi += 1
             else
diff --git a/test/testtables.jl b/test/testtables.jl
index 1ee54045..e4ddd551 100644
--- a/test/testtables.jl
+++ b/test/testtables.jl
@@ -310,6 +310,15 @@ function testtable(nm, t, writekw, readkw, extratests)
         @test all(isequal.(values(t), values(tt)))
         extratests !== nothing && extratests(tt)
         seekstart(io)
+
+        # Explicitly unthreaded
+        io = Arrow.tobuffer(t; writekw...)
+        tt = Arrow.Table(io; threaded=false, readkw...)
+        @test length(tt) == length(t)
+        @test all(isequal.(values(t), values(tt)))
+        extratests !== nothing && extratests(tt)
+        seekstart(io)
+
         str = Arrow.Stream(io; readkw...)
         tt = first(str)
         @test length(tt) == length(t)