Feat: support dataflow run. (#10182)

### What problem does this PR solve?


### Type of change

- [x] New Feature (non-breaking change which adds functionality)

rag/flow/base.py

@@ -35,9 +35,9 @@ class ProcessBase(ComponentBase):
     def __init__(self, pipeline, id, param: ProcessParamBase):
         super().__init__(pipeline, id, param)
         if hasattr(self._canvas, "callback"):
-            self.callback = partial(self._canvas.callback, self.component_name)
+            self.callback = partial(self._canvas.callback, id)
         else:
-            self.callback = partial(lambda *args, **kwargs: None, self.component_name)
+            self.callback = partial(lambda *args, **kwargs: None, id)
 
     async def invoke(self, **kwargs) -> dict[str, Any]:
         self.set_output("_created_time", time.perf_counter())

rag/flow/parser/parser.py

@@ -76,7 +76,6 @@ class ParserParam(ProcessParamBase):
         self.setups = {
             "pdf": {
                 "parse_method": "deepdoc",  # deepdoc/plain_text/vlm
-                "llm_id": "",
                 "lang": "Chinese",
                 "suffix": [
                     "pdf",
@@ -98,8 +97,8 @@ class ParserParam(ProcessParamBase):
                 ],
                 "output_format": "json",
             },
-            "markdown": {
-                "suffix": ["md", "markdown", "mdx"],
+            "text&markdown": {
+                "suffix": ["md", "markdown", "mdx", "txt"],
                 "output_format": "json",
             },
             "slides": {
@@ -156,13 +155,10 @@ class ParserParam(ProcessParamBase):
         pdf_config = self.setups.get("pdf", {})
         if pdf_config:
             pdf_parse_method = pdf_config.get("parse_method", "")
-            self.check_valid_value(pdf_parse_method.lower(), "Parse method abnormal.", ["deepdoc", "plain_text", "vlm"])
+            self.check_empty(pdf_parse_method, "Parse method abnormal.")
 
-            if pdf_parse_method not in ["deepdoc", "plain_text"]:
-                self.check_empty(pdf_config.get("llm_id"), "VLM")
-
-            pdf_language = pdf_config.get("lang", "")
-            self.check_empty(pdf_language, "Language")
+            if pdf_parse_method.lower() not in ["deepdoc", "plain_text"]:
+                self.check_empty(pdf_config.get("lang", ""), "Language")
 
             pdf_output_format = pdf_config.get("output_format", "")
             self.check_valid_value(pdf_output_format, "PDF output format abnormal.", self.allowed_output_format["pdf"])
@@ -226,8 +222,7 @@ class Parser(ProcessBase):
             lines, _ = PlainParser()(blob)
             bboxes = [{"text": t} for t, _ in lines]
         else:
-            assert conf.get("llm_id")
-            vision_model = LLMBundle(self._canvas._tenant_id, LLMType.IMAGE2TEXT, llm_name=conf.get("llm_id"), lang=self._param.setups["pdf"].get("lang"))
+            vision_model = LLMBundle(self._canvas._tenant_id, LLMType.IMAGE2TEXT, llm_name=conf.get("parse_method"), lang=self._param.setups["pdf"].get("lang"))
             lines, _ = VisionParser(vision_model=vision_model)(blob, callback=self.callback)
             bboxes = []
             for t, poss in lines:
@@ -236,6 +231,7 @@ class Parser(ProcessBase):
 
         if conf.get("output_format") == "json":
             self.set_output("json", bboxes)
+
         if conf.get("output_format") == "markdown":
             mkdn = ""
             for b in bboxes:
@@ -299,7 +295,6 @@ class Parser(ProcessBase):
 
     def _markdown(self, name, blob):
         from functools import reduce
-
         from rag.app.naive import Markdown as naive_markdown_parser
         from rag.nlp import concat_img
 
@@ -330,22 +325,6 @@ class Parser(ProcessBase):
         else:
             self.set_output("text", "\n".join([section_text for section_text, _ in sections]))
 
-    def _text(self, name, blob):
-        from deepdoc.parser.utils import get_text
-
-        self.callback(random.randint(1, 5) / 100.0, "Start to work on a text.")
-        conf = self._param.setups["text"]
-        self.set_output("output_format", conf["output_format"])
-
-        # parse binary to text
-        text_content = get_text(name, binary=blob)
-
-        if conf.get("output_format") == "json":
-            result = [{"text": text_content}]
-            self.set_output("json", result)
-        else:
-            result = text_content
-            self.set_output("text", result)
 
     def _image(self, from_upstream: ParserFromUpstream):
         from deepdoc.vision import OCR
@@ -367,7 +346,7 @@ class Parser(ProcessBase):
 
         else:
             # use VLM to describe the picture
-            cv_model = LLMBundle(self._canvas.get_tenant_id(), LLMType.IMAGE2TEXT, llm_name=conf["llm_id"], lang=lang)
+            cv_model = LLMBundle(self._canvas.get_tenant_id(), LLMType.IMAGE2TEXT, llm_name=conf["llm_id"],lang=lang)
             img_binary = io.BytesIO()
             img.save(img_binary, format="JPEG")
             img_binary.seek(0)
@@ -519,11 +498,10 @@ class Parser(ProcessBase):
     async def _invoke(self, **kwargs):
         function_map = {
             "pdf": self._pdf,
-            "markdown": self._markdown,
+            "text&markdown": self._markdown,
             "spreadsheet": self._spreadsheet,
             "slides": self._slides,
             "word": self._word,
-            "text": self._text,
             "image": self._image,
             "audio": self._audio,
             "email": self._email,

rag/flow/pipeline.py

@@ -18,7 +18,7 @@ import json
 import logging
 import random
 import time
-
+from timeit import default_timer as timer
 import trio
 
 from agent.canvas import Graph
@@ -38,25 +38,26 @@ class Pipeline(Graph):
 
     def callback(self, component_name: str, progress: float | int | None = None, message: str = "") -> None:
         log_key = f"{self._flow_id}-{self.task_id}-logs"
+        timestamp = timer()
         try:
             bin = REDIS_CONN.get(log_key)
             obj = json.loads(bin.encode("utf-8"))
             if obj:
-                if obj[-1]["component_name"] == component_name:
-                    obj[-1]["trace"].append({"progress": progress, "message": message, "datetime": datetime.datetime.now().strftime("%H:%M:%S")})
+                if obj[-1]["component_id"] == component_name:
+                    obj[-1]["trace"].append({"progress": progress, "message": message, "datetime": datetime.datetime.now().strftime("%H:%M:%S"), "timestamp": timestamp, "elapsed_time": timestamp-obj[-1]["trace"][-1]["timestamp"]})
                 else:
-                    obj.append({"component_name": component_name, "trace": [{"progress": progress, "message": message, "datetime": datetime.datetime.now().strftime("%H:%M:%S")}]})
+                    obj.append({"component_id": component_name, "trace": [{"progress": progress, "message": message, "datetime": datetime.datetime.now().strftime("%H:%M:%S"), "timestamp": timestamp, "elapsed_time": 0}]})
             else:
-                obj = [{"component_name": component_name, "trace": [{"progress": progress, "message": message, "datetime": datetime.datetime.now().strftime("%H:%M:%S")}]}]
+                obj = [{"component_id": component_name, "trace": [{"progress": progress, "message": message, "datetime": datetime.datetime.now().strftime("%H:%M:%S"), "timestamp": timestamp, "elapsed_time": 0}]}]
             REDIS_CONN.set_obj(log_key, obj, 60 * 30)
             if self._doc_id:
                 percentage = 1./len(self.components.items())
                 msg = ""
                 finished = 0.
                 for o in obj:
-                    if o['component_name'] == "END":
+                    if o['component_id'] == "END":
                         continue
-                    msg += f"\n[{o['component_name']}]:\n"
+                    msg += f"\n[{o['component_id']}]:\n"
                     for t in o["trace"]:
                         msg += "%s: %s\n"%(t["datetime"], t["message"])
                         if t["progress"] < 0:

rag/flow/tests/client.py

@@ -30,7 +30,7 @@ def print_logs(pipeline: Pipeline):
     while True:
         time.sleep(5)
         logs = pipeline.fetch_logs()
-        logs_str = json.dumps(logs)
+        logs_str = json.dumps(logs, ensure_ascii=False)
         if logs_str != last_logs:
             print(logs_str)
         last_logs = logs_str