s3_list_keys() returns empty results despite objects existing in S3

[hashemamireh]

s3_list_keys() returns empty results despite objects existing in S3

Description

s3_list_keys() returns an empty array when listing objects under certain prefixes, even though the objects exist and can be retrieved with s3_get(). The underlying S3 API (S3.list_objects_v2) correctly returns the objects with KeyCount=1, but s3_list_keys() fails to extract them.

Environment

• Julia version: 1.11.7
• AWSS3.jl version: v0.11.4
• AWS.jl version: v1.96.0
• OS: macOS

Minimal Reproducible Example

using AWSS3
using AWS
using AWS: @service
@service S3

aws = global_aws_config(; region="us-east-1")
bucket = "my-bucket"
prefix = "path/to/partition/marketplace_id=1/order_day=2025-09-01"

# This file exists and can be retrieved
key = "path/to/partition/marketplace_id=1/order_day=2025-09-01/part-00000.snappy.parquet"
parquet_object = s3_get(aws, bucket, key)  # ✅ Works fine

# But listing returns empty
keys = collect(s3_list_keys(aws, bucket, prefix))
println("Found $(length(keys)) keys")  # ❌ Returns: Found 0 keys

# However, the raw S3 API shows the object exists
response = S3.list_objects_v2(bucket, Dict("prefix" => prefix, "max-keys" => "10"))
println(response)
# ✅ Returns: KeyCount => "1", Contents => {...}

Expected Behavior

s3_list_keys() should return the keys that exist under the specified prefix, matching what the underlying S3 API returns.

Actual Behavior

s3_list_keys() returns an empty array, even though:

1. The S3 API response shows KeyCount=1 with valid Contents
2. s3_get() successfully retrieves the object using the full key
3. s3_list_objects() returns a Channel that contains the object metadata when iterated

Diagnostic Output

julia> keys = collect(s3_list_keys(aws, bucket, prefix))
Any[]

julia> response = S3.list_objects_v2(bucket, Dict("prefix" => prefix, "max-keys" => "10"))
OrderedCollections.LittleDict{Union{String, Symbol}, Any, Vector{Union{String, Symbol}}, Vector{Any}} with 6 entries:
  "Name"        => "my-bucket"
  "Prefix"      => "path/to/partition/marketplace_id=1/order_day=2025-09-01/"
  "KeyCount"    => "1"
  "MaxKeys"     => "10"
  "IsTruncated" => "false"
  "Contents"    => LittleDict{Union{String, Symbol}, Any, Vector{Union{String, Symbol}}, Vector{Any}}("Key"=>"path/to/partition/marketplace_id=1/order_day=2025-09-01/part-00000.snappy.parquet", ...)

Additional Context

Both s3_list_keys() and s3_list_objects() return empty results, even though the underlying S3 API confirms the objects exist:

julia> objects_channel = s3_list_objects(aws, bucket, prefix)
Channel{OrderedCollections.LittleDict}(128) (empty)

julia> keys = collect(s3_list_keys(aws, bucket, prefix))
Any[]

Workaround

Use the raw S3 API directly:

using AWS: @service
@service S3

response = S3.list_objects_v2(bucket, Dict("prefix" => prefix))
if haskey(response, "Contents")
    contents = response["Contents"]
    keys = isa(contents, Vector) ? [c["Key"] for c in contents] : [contents["Key"]]
end

Possible Cause

The issue may be related to how s3_list_keys() and s3_list_objects() parse the S3 API response when dealing with specific prefix patterns or when Contents contains a single object versus an array of objects. The Channel is created but never populated with data from the API response.

I will look into it more and try to push a bug fix when I get some down time but wanted to document the issue and flag it.