Unstructured-IO · CyMule · Sep 11, 2026 · Sep 10, 2026
diff --git a/.github/workflows/ci.yml b/.github/workflows/ci.yml
@@ -6,6 +6,10 @@ on:
   pull_request:
     branches: [ main ]
 
+concurrency:
+  group: ${{ github.workflow }}-${{ github.event.pull_request.number || github.ref }}
+  cancel-in-progress: ${{ github.event_name == 'pull_request' }}
+
 env:
   PIPELINE_FAMILY: "general"
 
@@ -21,6 +25,7 @@ jobs:
       with:
         enable-cache: true
         cache-dependency-glob: "uv.lock"
+        cache-suffix: lint
     - name: Set up Python ${{ env.PYTHON_VERSION }}
       run: uv python install ${{ env.PYTHON_VERSION }}
     - name: Install lint dependencies
@@ -47,19 +52,24 @@ jobs:
       with:
         enable-cache: true
         cache-dependency-glob: "uv.lock"
+        cache-suffix: test
     - name: Set up Python ${{ env.PYTHON_VERSION }}
       run: uv python install ${{ env.PYTHON_VERSION }}
-    - name: Install dependencies and run core tests
+    - name: Install system dependencies
       run: |
         sudo apt-get update && sudo apt-get install --yes poppler-utils libreoffice
-        uv sync --group test --locked
         make install-pandoc
-        make install-nlp-models
         sudo add-apt-repository -y ppa:alex-p/tesseract-ocr5
         sudo apt-get install -y tesseract-ocr tesseract-ocr-kor
         tesseract --version
-        make test
-        make check-coverage
+    - name: Install test dependencies
+      run: uv sync --group test --locked
+    - name: Install NLP models
+      run: make install-nlp-models
+    - name: Run core tests
+      run: make test
+    - name: Check coverage
+      run: make check-coverage
 
   changelog:
     runs-on: opensource-linux-8core
@@ -77,7 +87,6 @@ jobs:
     - if: steps.changes.outputs.src == 'true' && github.ref != 'refs/heads/main'
       uses: dangoslen/changelog-enforcer@v3
 
-  # TODO - figure out best practice for caching docker images
   test_dockerfile:
     runs-on: opensource-linux-8core
     needs: lint
@@ -90,6 +99,7 @@ jobs:
       with:
         enable-cache: true
         cache-dependency-glob: "uv.lock"
+        cache-suffix: smoke
     - name: Set up Python ${{ env.PYTHON_VERSION }}
       run: uv python install ${{ env.PYTHON_VERSION }}
     - name: Free up disk space
@@ -112,8 +122,9 @@ jobs:
 
         echo "Disk usage after cleanup:"
         df -h
+    - name: Install smoke test dependencies
+      run: uv sync --only-group smoke --locked
+    - name: Build Dockerfile
+      run: make docker-build
     - name: Test Dockerfile
-      run: |
-        uv sync --group test --locked
-        make docker-build
-        make docker-test
+      run: make docker-test
diff --git a/.github/workflows/docker-publish.yml b/.github/workflows/docker-publish.yml
@@ -80,10 +80,11 @@ jobs:
       with:
         enable-cache: true
         cache-dependency-glob: "uv.lock"
+        cache-suffix: smoke
     - name: Set up Python ${{ env.PYTHON_VERSION }}
       run: uv python install ${{ env.PYTHON_VERSION }}
     - name: Install test dependencies
-      run: uv sync --group test --locked
+      run: uv sync --only-group smoke --locked
     - name: Test image
       run: |
         export DOCKER_IMAGE="$DOCKER_BUILD_REPOSITORY:${{ matrix.arch }}-$SHORT_SHA"
@@ -100,33 +101,28 @@ jobs:
     steps:
     - name: Checkout code
       uses: actions/checkout@v5
-    - name: Set SHORT_SHA
-      run: echo "SHORT_SHA=$(git rev-parse --short HEAD)" >> $GITHUB_ENV
+    - name: Set up Docker Buildx
+      uses: docker/setup-buildx-action@v4
     - name: Login to Quay.io
       uses: docker/login-action@v4
       with:
         registry: quay.io
         username: ${{ secrets.QUAY_IO_ROBOT_USERNAME }}
         password: ${{ secrets.QUAY_IO_ROBOT_TOKEN }}
-    - name: Pull AMD image
-      run: |
-        docker pull $DOCKER_BUILD_REPOSITORY:amd64-$SHORT_SHA
-    - name: Pull ARM image
-      run: |
-        docker pull $DOCKER_BUILD_REPOSITORY:arm64-$SHORT_SHA
-    - name: Push AMD and ARM tags
+    - name: Update architecture cache tags
       run: |
-        # these are used to construct the final manifest but also cache-from in subsequent runs
-        docker tag $DOCKER_BUILD_REPOSITORY:amd64-$SHORT_SHA $DOCKER_BUILD_REPOSITORY:amd64
-        docker push $DOCKER_BUILD_REPOSITORY:amd64
-        docker tag $DOCKER_BUILD_REPOSITORY:arm64-$SHORT_SHA $DOCKER_BUILD_REPOSITORY:arm64
-        docker push $DOCKER_BUILD_REPOSITORY:arm64
+        docker buildx imagetools create --prefer-index=false \
+          -t $DOCKER_BUILD_REPOSITORY:amd64 \
+          $DOCKER_BUILD_REPOSITORY:amd64-$SHORT_SHA
+        docker buildx imagetools create --prefer-index=false \
+          -t $DOCKER_BUILD_REPOSITORY:arm64 \
+          $DOCKER_BUILD_REPOSITORY:arm64-$SHORT_SHA
     - name: Push multiarch manifest
       run: |
         VERSION=$(grep -oP '(?<=__version__ = ")[^"]+' prepline_general/api/__version__.py)
         docker buildx imagetools create \
           -t ${DOCKER_REPOSITORY}:latest \
           -t ${DOCKER_REPOSITORY}:$SHORT_SHA \
           -t ${DOCKER_REPOSITORY}:$VERSION \
-          $DOCKER_BUILD_REPOSITORY:amd64 \
-          $DOCKER_BUILD_REPOSITORY:arm64
+          $DOCKER_BUILD_REPOSITORY:amd64-$SHORT_SHA \
+          $DOCKER_BUILD_REPOSITORY:arm64-$SHORT_SHA
diff --git a/pyproject.toml b/pyproject.toml
@@ -16,6 +16,11 @@ dependencies = [
 ]
 
 [dependency-groups]
+smoke = [
+    "pytest >=9.0.2, <10.0.0",
+    "pandas >=3.0.0, <4.0.0",
+    "requests >=2.32.0, <3.0.0",
+]
 lint = [
     "ruff >=0.15.0, <1.0.0",
     "mypy >=1.19.1, <2.0.0",

diff --git a/scripts/docker-smoke-test.sh b/scripts/docker-smoke-test.sh
@@ -80,7 +80,7 @@ await_server_ready 8000
 # Smoke Tests
 #######################
 echo Running smoke tests with SKIP_INFERENCE_TESTS: "$SKIP_INFERENCE_TESTS"
-PYTHONPATH=. SKIP_INFERENCE_TESTS=$SKIP_INFERENCE_TESTS uv run pytest -vv scripts/smoketest.py
+PYTHONPATH=. SKIP_INFERENCE_TESTS=$SKIP_INFERENCE_TESTS uv run --no-sync pytest --durations=20 -vv scripts/smoketest.py
 
 #######################
 # Test parallel vs single mode

diff --git a/scripts/parallel-mode-test.sh b/scripts/parallel-mode-test.sh
@@ -1,73 +1,40 @@
 #!/usr/bin/env bash
 
-# parallel-mode-test.sh
-# Iterate a list of curl commands, and run each one against two instances of the api
-# The smoke test will start one container with parallel mode and one without, and
-# diff the two outputs to make sure parallel mode does not alter the response.
-# Note the filepaths assume you ran this from the top level
-
-# shellcheck disable=SC2317,SC2086  # SC2317: trap functions appear unreachable, SC2086: curl params require word splitting
+set -euo pipefail
 
 base_url_1=$1
 base_url_2=$2
-MAX_RETRIES=3
+results_dir=$(mktemp -d)
+trap 'rm -rf "$results_dir"' EXIT
 
-curl_with_retry() {
-    local curl_command=$1
+request_elements() {
+    local base_url=$1
     local output_file=$2
+    shift 2
 
-    for attempt in $(seq 1 $MAX_RETRIES); do
-        $curl_command 2> /dev/null | jq -S 'del(..|.parent_id?)' > "$output_file"
-        if [ -s "$output_file" ]; then
-            return 0
-        fi
-        echo "  Attempt $attempt/$MAX_RETRIES failed, retrying in 5s..."
-        sleep 5
-    done
-
-    echo "  All $MAX_RETRIES attempts failed!"
-    $curl_command
-    return 1
+    curl --fail --silent --show-error --retry 2 --max-time 300 \
+        "$base_url/general/v0/general" \
+        -F files=@sample-docs/layout-parser-paper.pdf \
+        "$@" > "$output_file"
+    jq -e 'type == "array" and length > 0' "$output_file" > /dev/null
 }
 
-declare -a curl_params=(
-    "-F files=@sample-docs/layout-parser-paper.pdf -F 'strategy=fast'"
-    "-F files=@sample-docs/layout-parser-paper.pdf -F 'strategy=auto"
-    "-F files=@sample-docs/layout-parser-paper.pdf -F 'strategy=hi_res'"
-    "-F files=@sample-docs/layout-parser-paper.pdf -F 'coordinates=true'"
-    "-F files=@sample-docs/layout-parser-paper.pdf -F 'encoding=utf-8'"
-    "-F files=@sample-docs/layout-parser-paper.pdf -F 'include_page_breaks=true'"
-    "-F files=@sample-docs/layout-parser-paper.pdf -F 'hi_res_model_name=yolox'"
-)
-
-for params in "${curl_params[@]}"
-do
-   curl_command="curl $base_url_1/general/v0/general $params"
-   echo Testing: "$curl_command"
-
-   # Run in single mode
-   # Note(austin): Parallel mode screws up hierarchy! While we deal with that,
-   # let's ignore parent_id fields in the results
-   if ! curl_with_retry "$curl_command" output.json; then
-       echo Command failed!
-       exit 1
-   fi
-   original_length=$(jq 'length' output.json)
-
-   # Run in parallel mode
-   curl_command="curl $base_url_2/general/v0/general $params"
-   if ! curl_with_retry "$curl_command" parallel_output.json; then
-       echo Command failed!
-       exit 1
-   fi
-   parallel_length=$(jq 'length' parallel_output.json)
-
-   if ! [[ "$original_length" == "$parallel_length" ]]; then
-       echo Parallel mode returned a different number of elements!
-       echo Params: "$params"
-       exit 1
-   fi
-
-   rm -f output.json parallel_output.json
-   echo
+for option in strategy=fast strategy=auto strategy=hi_res coordinates=true encoding=utf-8 include_page_breaks=true hi_res_model_name=yolox; do
+    # Exercise forwarding options without repeating inference for each one.
+    case "$option" in
+        strategy=*) form_args=(-F "$option") ;;
+        hi_res_model_name=*) form_args=(-F strategy=hi_res -F "$option") ;;
+        *) form_args=(-F strategy=fast -F "$option") ;;
+    esac
+
+    echo "Testing: $option"
+    request_elements "$base_url_1" "$results_dir/single.json" "${form_args[@]}"
+    request_elements "$base_url_2" "$results_dir/parallel.json" "${form_args[@]}"
+
+    original_length=$(jq 'length' "$results_dir/single.json")
+    parallel_length=$(jq 'length' "$results_dir/parallel.json")
+    if [[ "$original_length" != "$parallel_length" ]]; then
+        echo "Parallel mode returned a different number of elements for $option"
+        exit 1
+    fi
 done
diff --git a/scripts/smoketest.py b/scripts/smoketest.py
@@ -146,7 +146,6 @@ def test_happy_path_all_types(extension, example_filename: str, content_type: st
     assert len(df) == len(json_response.json())
 
 
-@pytest.mark.parametrize("output_format", ["application/json", "text/csv"])
 @pytest.mark.parametrize(
     "filenames_to_gzip, filenames_verbatim, uncompressed_content_type",
     [
@@ -173,7 +172,6 @@ def test_happy_path_all_types(extension, example_filename: str, content_type: st
     ],
 )
 def test_gzip_sending(
-    output_format: str,
     filenames_to_gzip: List[str],
     filenames_verbatim: List[str],
     uncompressed_content_type: str,

diff --git a/test_general/test_parallel_mode_script.py b/test_general/test_parallel_mode_script.py
@@ -0,0 +1,88 @@
+import json
+import os
+import subprocess
+import sys
+from pathlib import Path
+
+import pytest
+
+SCRIPT = Path(__file__).resolve().parents[1] / "scripts" / "parallel-mode-test.sh"
+
+
+@pytest.fixture
+def run_comparisons(tmp_path):
+    curl = tmp_path / "curl"
+    curl.write_text(
+        f"#!{sys.executable}\n"
+        "import json, os, sys\n"
+        "with open(os.environ['REQUEST_LOG'], 'a') as log:\n"
+        "    log.write(json.dumps(sys.argv[1:]) + '\\n')\n"
+        "if os.environ.get('HTTP_FAILURE'):\n"
+        "    sys.exit(22)\n"
+        "body = os.environ.get('RESPONSE_BODY', '[{\"text\": \"content\"}]')\n"
+        "if os.environ.get('MISMATCH') and any('parallel/' in arg for arg in sys.argv):\n"
+        "    body = '[{}, {}]'\n"
+        "print(body)\n"
+    )
+    curl.chmod(0o755)
+    log = tmp_path / "requests.jsonl"
+
+    def run(**overrides):
+        result = subprocess.run(
+            ["bash", str(SCRIPT), "http://single", "http://parallel"],
+            env={
+                **os.environ,
+                "PATH": f"{tmp_path}:{os.environ['PATH']}",
+                "REQUEST_LOG": str(log),
+                **overrides,
+            },
+            capture_output=True,
+            text=True,
+            check=False,
+        )
+        requests = [json.loads(line) for line in log.read_text().splitlines()]
+        return result, requests
+
+    return run
+
+
+def test_comparisons_send_valid_options_to_both_servers(run_comparisons):
+    result, requests = run_comparisons()
+    assert result.returncode == 0, result.stderr
+    assert len(requests) == 14
+    expected = [
+        {"strategy": "fast"},
+        {"strategy": "auto"},
+        {"strategy": "hi_res"},
+        {"strategy": "fast", "coordinates": "true"},
+        {"strategy": "fast", "encoding": "utf-8"},
+        {"strategy": "fast", "include_page_breaks": "true"},
+        {"strategy": "hi_res", "hi_res_model_name": "yolox"},
+    ]
+    for index, args in enumerate(requests):
+        fields = dict(args[i + 1].split("=", 1) for i, arg in enumerate(args) if arg == "-F")
+        assert fields.pop("files") == "@sample-docs/layout-parser-paper.pdf"
+        assert fields == expected[index // 2]
+        server = "single" if index % 2 == 0 else "parallel"
+        assert f"http://{server}/general/v0/general" in args
+        assert "--fail" in args
+
+
+@pytest.mark.parametrize("body", ['{"detail": "server error"}', "[]", "invalid JSON"])
+def test_comparisons_reject_invalid_responses(run_comparisons, body):
+    result, requests = run_comparisons(RESPONSE_BODY=body)
+    assert result.returncode != 0
+    assert len(requests) == 1
+
+
+def test_comparisons_reject_http_failure(run_comparisons):
+    result, requests = run_comparisons(HTTP_FAILURE="1")
+    assert result.returncode != 0
+    assert len(requests) == 1
+
+
+def test_comparisons_reject_different_element_counts(run_comparisons):
+    result, requests = run_comparisons(MISMATCH="1")
+    assert result.returncode != 0
+    assert len(requests) == 2
+    assert "different number of elements" in result.stdout
diff --git a/uv.lock b/uv.lock