Update executable template

.cspell.json

@@ -188,7 +188,10 @@
     "pywin",
     "STARTF",
     "mltable",
-    "setenv"
+    "setenv",
+    "cscript",
+    "nologo",
+    "wscript"
   ],
   "flagWords": [
     "Prompt Flow"

src/promptflow-devkit/promptflow/_sdk/_service/utils/utils.py

@@ -278,7 +278,9 @@ def is_run_from_built_binary():
 
     Allow customer to use environment variable to control the triggering.
     """
-    return sys.executable.endswith("pfcli.exe") or os.environ.get(PF_RUN_AS_BUILT_BINARY, "").lower() == "true"
+    return (not sys.executable.endswith("python.exe") and not sys.executable.endswith("python")) or os.environ.get(
+        PF_RUN_AS_BUILT_BINARY, ""
+    ).lower() == "true"
 
 
 def encrypt_flow_path(flow_path):

src/promptflow-devkit/promptflow/_sdk/_tracing.py

@@ -4,6 +4,8 @@
 
 import json
 import os
+import subprocess
+import sys
 import typing
 import urllib.parse
 
@@ -14,7 +16,6 @@
 from opentelemetry.sdk.trace import TracerProvider
 from opentelemetry.sdk.trace.export import BatchSpanProcessor
 
-from promptflow._cli._pf.entry import entry
 from promptflow._constants import (
     OTEL_RESOURCE_SERVICE_NAME,
     SpanAttributeFieldName,
@@ -29,7 +30,12 @@
     AzureMLWorkspaceTriad,
     ContextAttributeKey,
 )
-from promptflow._sdk._service.utils.utils import get_port_from_config, is_pfs_service_healthy, is_port_in_use
+from promptflow._sdk._service.utils.utils import (
+    get_port_from_config,
+    is_pfs_service_healthy,
+    is_port_in_use,
+    is_run_from_built_binary,
+)
 from promptflow._sdk._utils import extract_workspace_triad_from_trace_provider
 from promptflow._utils.logger_utils import get_cli_sdk_logger
 from promptflow.tracing._integrations._openai_injector import inject_openai_api
@@ -61,7 +67,12 @@ def _inject_attrs_to_op_ctx(attrs: typing.Dict[str, str]) -> None:
 def _invoke_pf_svc() -> str:
     port = get_port_from_config(create_if_not_exists=True)
     port = str(port)
-    cmd_args = ["service", "start", "--port", port]
+    if is_run_from_built_binary():
+        interpreter_path = os.path.abspath(sys.executable)
+        pf_path = os.path.join(os.path.dirname(interpreter_path), "pf")
+        cmd_args = [pf_path, "service", "start", "--port", port]
+    else:
+        cmd_args = ["pf", "service", "start", "--port", port]
     hint_stop_message = (
         f"You can stop the Prompt flow Tracing Server with the following command:'\033[1m pf service stop\033[0m'.\n"
         f"Alternatively, if no requests are made within {PF_SERVICE_HOUR_TIMEOUT} "
@@ -75,7 +86,9 @@ def _invoke_pf_svc() -> str:
             print(hint_stop_message)
             return port
     print("Starting Prompt flow Tracing Server...")
-    entry(cmd_args)
+    start_pfs = subprocess.Popen(cmd_args, shell=True)
+    # Wait for service to be started
+    start_pfs.wait()
     logger.debug("Prompt flow service is serving on port %s", port)
     print(hint_stop_message)
     return port

src/promptflow-devkit/promptflow/_sdk/data/executable/app.py

@@ -1,3 +1,4 @@
+import multiprocessing
 import os
 import sys
 
@@ -25,18 +26,27 @@ def create_connections(directory_path) -> None:
 
 
 if __name__ == "__main__":
-
-    create_connections(os.path.join(os.path.dirname(__file__), "connections"))
-    if exists():
-        start()
+    multiprocessing.freeze_support()
+    command = sys.argv[1] if len(sys.argv) > 1 else None
+    print(f"Command: {command}")
+    print(sys.argv)
+    if command == "pf":
+        sys.argv = sys.argv[1:]
+        from promptflow._cli._pf.entry import main as pf_main
+
+        pf_main()
     else:
-        main_script = os.path.join(os.path.dirname(__file__), "main.py")
-        sys.argv = [
-            "streamlit",
-            "run",
-            main_script,
-            "--global.developmentMode=false",
-            "--client.toolbarMode=viewer",
-            "--browser.gatherUsageStats=false",
-        ]
-        st_cli.main(prog_name="streamlit")
+        create_connections(os.path.join(os.path.dirname(__file__), "connections"))
+        if exists():
+            start()
+        else:
+            main_script = os.path.join(os.path.dirname(__file__), "main.py")
+            sys.argv = [
+                "streamlit",
+                "run",
+                main_script,
+                "--global.developmentMode=false",
+                "--client.toolbarMode=viewer",
+                "--browser.gatherUsageStats=false",
+            ]
+            st_cli.main(prog_name="streamlit")

src/promptflow-devkit/promptflow/_sdk/data/executable/app.spec.jinja2

@@ -1,25 +1,47 @@
 # -*- mode: python ; coding: utf-8 -*-
 import sys
-from PyInstaller.utils.hooks import collect_data_files
-from PyInstaller.utils.hooks import copy_metadata
+from PyInstaller.utils.hooks import collect_data_files, collect_all, copy_metadata
 
 sys.setrecursionlimit(sys.getrecursionlimit() * 5)
 
-datas = [('connections', 'connections'), ('flow', 'flow'), ('settings.json', '.'), ('main.py', '.'), ('utils.py', '.'), ('logo.png', '.'), ('config.json', '.'), ('{{runtime_interpreter_path}}', './streamlit/runtime')]
-datas += collect_data_files('streamlit')
-datas += copy_metadata('streamlit')
+datas = [('connections', 'connections'), ('flow', 'flow'), ('settings.json', '.'), ('main.py', '.'), ('utils.py', '.'),
+('logo.png', '.'), ('config.json', '.'), ('{{runtime_interpreter_path}}', './streamlit/runtime')]
+
+
+all_packages = {{all_packages}}
+meta_packages = ['opentelemetry-api'] + {{meta_packages}}
+
+for package in all_packages:
+    datas += collect_data_files(package)
+
+for package in meta_packages:
+    datas += copy_metadata(package)
+
+opentelemetry_datas, opentelemetry_binaries, opentelemetry_hiddenimports = collect_all('opentelemetry')
+promptflow_datas, promptflow_binaries, promptflow_hiddenimports = collect_all('promptflow')
+
+datas += opentelemetry_datas
+datas += promptflow_datas
+datas += collect_data_files('streamlit_quill')
 datas += collect_data_files('keyrings.alt', include_py_files=True)
 datas += copy_metadata('keyrings.alt')
-datas += collect_data_files('streamlit_quill')
+
+hidden_imports = ['win32timezone', 'opentelemetry.context.contextvars_context', 'streamlit.runtime.scriptrunner.magic_funcs'] + {{hidden_imports}}
+hidden_imports += opentelemetry_hiddenimports
+hidden_imports += promptflow_hiddenimports
+
+binaries = []
+binaries += opentelemetry_binaries
+binaries += promptflow_binaries
 
 block_cipher = None
 
-a = Analysis(
+pfcli_a = Analysis(
     ['app.py', 'main.py', 'utils.py'],
     pathex=[],
-    binaries=[],
+    binaries=binaries,
     datas=datas,
-    hiddenimports={{hidden_imports}},
+    hiddenimports=hidden_imports,
     hookspath=[],
     hooksconfig={},
     runtime_hooks=[],
@@ -29,26 +51,34 @@ a = Analysis(
     cipher=block_cipher,
     noarchive=False,
 )
-pyz = PYZ(a.pure, a.zipped_data, cipher=block_cipher)
-
-exe = EXE(
-    pyz,
-    a.scripts,
-    a.binaries,
-    a.zipfiles,
-    a.datas,
+pfcli_pyz = PYZ(pfcli_a.pure, pfcli_a.zipped_data, cipher=block_cipher)
+
+pfcli_exe = EXE(
+    pfcli_pyz,
+    pfcli_a.scripts,
     [],
+    exclude_binaries=True,
     name='app',
     debug=False,
     bootloader_ignore_signals=False,
     strip=False,
     upx=True,
-    upx_exclude=[],
-    runtime_tmpdir=None,
     console=True,
     disable_windowed_traceback=False,
     argv_emulation=False,
     target_arch=None,
     codesign_identity=None,
     entitlements_file=None,
+    contents_directory='.',
+)
+
+coll = COLLECT(
+    pfcli_exe,
+    pfcli_a.binaries,
+    pfcli_a.zipfiles,
+    pfcli_a.datas,
+    strip=False,
+    upx=True,
+    upx_exclude=[],
+    name='app',
 )

src/promptflow-devkit/promptflow/_sdk/data/executable/main.py

@@ -48,17 +48,17 @@ def post_process_dump_result(response, session_state_history, *, generator_recor
         # Just use BasicMultimediaProcessor to keep the original logic here.
         # TODO: Add support for other multimedia types
         multimedia_processor = BasicMultimediaProcessor()
+        multimedia_response = {
+            k: multimedia_processor.load_multimedia_data_recursively(v) for k, v in response.output.items()
+        }
         resolved_outputs = {
-            k: multimedia_processor.convert_multimedia_data_to_base64_dict(v) for k, v in response.output.items()
+            k: multimedia_processor.convert_multimedia_data_to_base64_dict(v) for k, v in multimedia_response.items()
         }
         st.session_state.messages.append(("assistant", resolved_outputs))
         session_state_history.update({"outputs": response.output})
         st.session_state.history.append(session_state_history)
         if is_chat_flow:
             dump_path = Path(flow_path).parent
-            response.output = multimedia_processor.persist_multimedia_data(
-                response.output, base_dir=dump_path, sub_dir=Path(".promptflow/output")
-            )
             dump_flow_result(flow_folder=dump_path, flow_result=response, prefix="chat")
         return resolved_outputs
 
@@ -71,14 +71,14 @@ def submit(**kwargs) -> None:
         session_state_history.update({"inputs": kwargs})
         with container:
             render_message("user", kwargs)
-        # Force append chat history to kwargs
-        if is_chat_flow:
-            kwargs[chat_history_input_name] = get_chat_history_from_session()
 
         flow = load_flow(flow_path)
         with TestSubmitter(flow=flow, flow_context=flow.context).init(stream_output=is_streaming) as submitter:
             # can't exit the context manager before the generator is fully consumed
-            response = submitter.flow_test(inputs=kwargs, allow_generator_output=is_streaming)
+            response = submitter.flow_test(
+                inputs={chat_history_input_name: get_chat_history_from_session(), **kwargs},
+                allow_generator_output=is_streaming,
+            )
 
             if response.run_info.status.value == "Failed":
                 raise Exception(response.run_info.error)

src/promptflow-devkit/promptflow/_sdk/data/executable/pf

@@ -0,0 +1,15 @@
+#!/bin/bash
+
+MAIN_EXE="./app"
+
+# Check if the first argument is 'service'
+if [ "$1" == "service" ]; then
+    # Check if the second argument is 'start'
+    if [ "$2" == "start" ]; then
+        nohup "$MAIN_EXE" pf "${@:1}" > output.txt 2>&1 &
+    else
+        "$MAIN_EXE" pf "${@:1}"
+    fi
+else
+    "$MAIN_EXE" pf "${@:1}"
+fi

src/promptflow-devkit/promptflow/_sdk/data/executable/pf.bat

@@ -0,0 +1,21 @@
+@echo off
+setlocal
+
+
+set MAIN_EXE=%~dp0.\app.exe
+REM Check if the first argument is 'start'
+if "%~1"=="service" (
+    REM Check if the second argument is 'start'
+    if "%~2"=="start" (
+        echo Starting service...
+        echo %*
+        cscript //nologo %~dp0.\start_pfs.vbs """%MAIN_EXE%"" pf %*"
+        REM since we won't wait for vbs to finish, we need to wait for the output file to be flushed to disk
+        timeout /t 5 >nul
+        type "%~dp0output.txt"
+    ) else (
+        "%MAIN_EXE%" pf %*
+    )
+) else (
+    "%MAIN_EXE%" pf %*
+)

src/promptflow-devkit/promptflow/_sdk/data/executable/start_pfs.vbs

@@ -0,0 +1,4 @@
+DIM objshell
+set objshell = wscript.createobject("wscript.shell")
+cmd = WScript.Arguments(0)
+iReturn = objshell.run(cmd, 0, false)