from __future__ import annotations import asyncio import copy import json import os import random import time import traceback import uuid from collections import defaultdict from queue import Queue as ThreadQueue from typing import TYPE_CHECKING import fastapi from typing_extensions import Literal from gradio import route_utils, routes from gradio.data_classes import ( PredictBody, ) from gradio.exceptions import Error from gradio.helpers import TrackedIterable from gradio.server_messages import ( EstimationMessage, EventMessage, LogMessage, ProcessCompletedMessage, ProcessGeneratingMessage, ProcessStartsMessage, ProgressMessage, ProgressUnit, ) from gradio.utils import LRUCache, run_coro_in_background, safe_get_lock, set_task_name if TYPE_CHECKING: from gradio.blocks import BlockFunction class Event: def __init__( self, session_hash: str | None, fn_index: int, request: fastapi.Request, username: str | None, concurrency_id: str, ): self._id = uuid.uuid4().hex self.session_hash: str = session_hash or self._id self.fn_index = fn_index self.request = request self.username = username self.concurrency_id = concurrency_id self.data: PredictBody | None = None self.progress: ProgressMessage | None = None self.progress_pending: bool = False self.alive = True class EventQueue: def __init__(self, concurrency_id: str, concurrency_limit: int | None): self.queue: list[Event] = [] self.concurrency_id = concurrency_id self.concurrency_limit = concurrency_limit self.current_concurrency = 0 self.start_times_per_fn_index: defaultdict[int, set[float]] = defaultdict(set) class ProcessTime: def __init__(self): self.process_time = 0 self.count = 0 self.avg_time = 0 def add(self, time: float): self.process_time += time self.count += 1 self.avg_time = self.process_time / self.count class Queue: def __init__( self, live_updates: bool, concurrency_count: int, update_intervals: float, max_size: int | None, block_fns: list[BlockFunction], default_concurrency_limit: int | None | Literal["not_set"] = "not_set", ): self.pending_messages_per_session: LRUCache[ str, ThreadQueue[EventMessage] ] = LRUCache(2000) self.pending_event_ids_session: dict[str, set[str]] = {} self.pending_message_lock = safe_get_lock() self.event_queue_per_concurrency_id: dict[str, EventQueue] = {} self.stopped = False self.max_thread_count = concurrency_count self.update_intervals = update_intervals self.active_jobs: list[None | list[Event]] = [] self.delete_lock = safe_get_lock() self.server_app = None self.process_time_per_fn_index: defaultdict[int, ProcessTime] = defaultdict( ProcessTime ) self.live_updates = live_updates self.sleep_when_free = 0.05 self.progress_update_sleep_when_free = 0.1 self.max_size = max_size self.block_fns = block_fns self.continuous_tasks: list[Event] = [] self._asyncio_tasks: list[asyncio.Task] = [] self.default_concurrency_limit = self._resolve_concurrency_limit( default_concurrency_limit ) def start(self): self.active_jobs = [None] * self.max_thread_count self.set_event_queue_per_concurrency_id() run_coro_in_background(self.start_processing) run_coro_in_background(self.start_progress_updates) if not self.live_updates: run_coro_in_background(self.notify_clients) def set_event_queue_per_concurrency_id(self): for block_fn in self.block_fns: concurrency_id = block_fn.concurrency_id concurrency_limit: int | None if block_fn.concurrency_limit == "default": concurrency_limit = self.default_concurrency_limit else: concurrency_limit = block_fn.concurrency_limit if concurrency_id not in self.event_queue_per_concurrency_id: self.event_queue_per_concurrency_id[concurrency_id] = EventQueue( concurrency_id, concurrency_limit ) elif ( concurrency_limit is not None ): # Update concurrency limit if it is lower than existing limit existing_event_queue = self.event_queue_per_concurrency_id[ concurrency_id ] if ( existing_event_queue.concurrency_limit is None or concurrency_limit < existing_event_queue.concurrency_limit ): existing_event_queue.concurrency_limit = concurrency_limit def reload(self): self.set_event_queue_per_concurrency_id() def close(self): self.stopped = True def send_message( self, event: Event, event_message: EventMessage, ): if not event.alive: return event_message.event_id = event._id messages = self.pending_messages_per_session[event.session_hash] messages.put_nowait(event_message) def _resolve_concurrency_limit( self, default_concurrency_limit: int | None | Literal["not_set"] ) -> int | None: """ Handles the logic of resolving the default_concurrency_limit as this can be specified via a combination of the `default_concurrency_limit` parameter of the `Blocks.queue()` or the `GRADIO_DEFAULT_CONCURRENCY_LIMIT` environment variable. The parameter in `Blocks.queue()` takes precedence over the environment variable. Parameters: default_concurrency_limit: The default concurrency limit, as specified by a user in `Blocks.queu()`. """ if default_concurrency_limit != "not_set": return default_concurrency_limit if default_concurrency_limit_env := os.environ.get( "GRADIO_DEFAULT_CONCURRENCY_LIMIT" ): if default_concurrency_limit_env.lower() == "none": return None else: return int(default_concurrency_limit_env) else: return 1 def __len__(self): total_len = 0 for event_queue in self.event_queue_per_concurrency_id.values(): total_len += len(event_queue.queue) return total_len async def push( self, body: PredictBody, request: fastapi.Request, username: str | None ) -> tuple[bool, str]: if body.fn_index is None: return False, "No function index provided." if self.max_size is not None and len(self) >= self.max_size: return ( False, f"Queue is full. Max size is {self.max_size} and size is {len(self)}.", ) event = Event( body.session_hash, body.fn_index, request, username, self.block_fns[body.fn_index].concurrency_id, ) event.data = body if body.session_hash is None: body.session_hash = event.session_hash async with self.pending_message_lock: if body.session_hash not in self.pending_messages_per_session: self.pending_messages_per_session[body.session_hash] = ThreadQueue() if body.session_hash not in self.pending_event_ids_session: self.pending_event_ids_session[body.session_hash] = set() self.pending_event_ids_session[body.session_hash].add(event._id) event_queue = self.event_queue_per_concurrency_id[event.concurrency_id] event_queue.queue.append(event) self.broadcast_estimations(event.concurrency_id, len(event_queue.queue) - 1) return True, event._id def _cancel_asyncio_tasks(self): for task in self._asyncio_tasks: task.cancel() self._asyncio_tasks = [] def set_server_app(self, app: routes.App): self.server_app = app def get_active_worker_count(self) -> int: count = 0 for worker in self.active_jobs: if worker is not None: count += 1 return count def get_events(self) -> tuple[list[Event], bool, str] | None: concurrency_ids = list(self.event_queue_per_concurrency_id.keys()) random.shuffle(concurrency_ids) for concurrency_id in concurrency_ids: event_queue = self.event_queue_per_concurrency_id[concurrency_id] if len(event_queue.queue) and ( event_queue.concurrency_limit is None or event_queue.current_concurrency < event_queue.concurrency_limit ): first_event = event_queue.queue[0] block_fn = self.block_fns[first_event.fn_index] events = [first_event] batch = block_fn.batch if batch: events += [ event for event in event_queue.queue[1:] if event.fn_index == first_event.fn_index ][: block_fn.max_batch_size - 1] for event in events: event_queue.queue.remove(event) return events, batch, concurrency_id async def start_processing(self) -> None: try: while not self.stopped: if len(self) == 0: await asyncio.sleep(self.sleep_when_free) continue if None not in self.active_jobs: await asyncio.sleep(self.sleep_when_free) continue # Using mutex to avoid editing a list in use async with self.delete_lock: event_batch = self.get_events() if event_batch: events, batch, concurrency_id = event_batch self.active_jobs[self.active_jobs.index(None)] = events event_queue = self.event_queue_per_concurrency_id[concurrency_id] event_queue.current_concurrency += 1 start_time = time.time() event_queue.start_times_per_fn_index[events[0].fn_index].add( start_time ) process_event_task = run_coro_in_background( self.process_events, events, batch, start_time ) set_task_name( process_event_task, events[0].session_hash, events[0].fn_index, batch, ) self._asyncio_tasks.append(process_event_task) if self.live_updates: self.broadcast_estimations(concurrency_id) else: await asyncio.sleep(self.sleep_when_free) finally: self.stopped = True self._cancel_asyncio_tasks() async def start_progress_updates(self) -> None: """ Because progress updates can be very frequent, we do not necessarily want to send a message per update. Rather, we check for progress updates at regular intervals, and send a message if there is a pending update. Consecutive progress updates between sends will overwrite each other so only the most recent update will be sent. """ while not self.stopped: events = [ evt for job in self.active_jobs if job is not None for evt in job ] + self.continuous_tasks if len(events) == 0: await asyncio.sleep(self.progress_update_sleep_when_free) continue for event in events: if event.progress_pending and event.progress: event.progress_pending = False self.send_message(event, event.progress) await asyncio.sleep(self.progress_update_sleep_when_free) def set_progress( self, event_id: str, iterables: list[TrackedIterable] | None, ): if iterables is None: return for job in self.active_jobs: if job is None: continue for evt in job: if evt._id == event_id: progress_data: list[ProgressUnit] = [] for iterable in iterables: progress_unit = ProgressUnit( index=iterable.index, length=iterable.length, unit=iterable.unit, progress=iterable.progress, desc=iterable.desc, ) progress_data.append(progress_unit) evt.progress = ProgressMessage(progress_data=progress_data) evt.progress_pending = True def log_message( self, event_id: str, log: str, level: Literal["info", "warning"], ): events = [ evt for job in self.active_jobs if job is not None for evt in job ] + self.continuous_tasks for event in events: if event._id == event_id: log_message = LogMessage( log=log, level=level, ) self.send_message(event, log_message) async def clean_events( self, *, session_hash: str | None = None, event_id: str | None = None ) -> None: for job_set in self.active_jobs: if job_set: for job in job_set: if job.session_hash == session_hash or job._id == event_id: job.alive = False async with self.delete_lock: events_to_remove: list[Event] = [] for event_queue in self.event_queue_per_concurrency_id.values(): for event in event_queue.queue: if event.session_hash == session_hash or event._id == event_id: events_to_remove.append(event) for event in events_to_remove: self.event_queue_per_concurrency_id[event.concurrency_id].queue.remove( event ) async def notify_clients(self) -> None: """ Notify clients about events statuses in the queue periodically. """ while not self.stopped: await asyncio.sleep(self.update_intervals) if len(self) > 0: for concurrency_id in self.event_queue_per_concurrency_id: self.broadcast_estimations(concurrency_id) def broadcast_estimations( self, concurrency_id: str, after: int | None = None ) -> None: wait_so_far = 0 event_queue = self.event_queue_per_concurrency_id[concurrency_id] time_till_available_worker: int | None = 0 if event_queue.current_concurrency == event_queue.concurrency_limit: expected_end_times = [] for fn_index, start_times in event_queue.start_times_per_fn_index.items(): if fn_index not in self.process_time_per_fn_index: time_till_available_worker = None break process_time = self.process_time_per_fn_index[fn_index].avg_time expected_end_times += [ start_time + process_time for start_time in start_times ] if time_till_available_worker is not None and len(expected_end_times) > 0: time_of_first_completion = min(expected_end_times) time_till_available_worker = max( time_of_first_completion - time.time(), 0 ) for rank, event in enumerate(event_queue.queue): process_time_for_fn = ( self.process_time_per_fn_index[event.fn_index].avg_time if event.fn_index in self.process_time_per_fn_index else None ) rank_eta = ( process_time_for_fn + wait_so_far + time_till_available_worker if process_time_for_fn is not None and wait_so_far is not None and time_till_available_worker is not None else None ) if after is None or rank >= after: self.send_message( event, EstimationMessage( rank=rank, rank_eta=rank_eta, queue_size=len(event_queue.queue) ), ) if event_queue.concurrency_limit is None: wait_so_far = 0 elif wait_so_far is not None and process_time_for_fn is not None: wait_so_far += process_time_for_fn / event_queue.concurrency_limit else: wait_so_far = None def get_status(self) -> EstimationMessage: return EstimationMessage( queue_size=len(self), ) async def call_prediction(self, events: list[Event], batch: bool) -> dict: body = events[0].data if body is None: raise ValueError("No event data") username = events[0].username body.event_id = events[0]._id if not batch else None try: body.request = events[0].request except ValueError: pass if batch: body.data = list(zip(*[event.data.data for event in events if event.data])) body.request = events[0].request body.batched = True app = self.server_app if app is None: raise Exception("Server app has not been set.") api_name = "predict" fn_index_inferred = route_utils.infer_fn_index( app=app, api_name=api_name, body=body ) gr_request = route_utils.compile_gr_request( app=app, body=body, fn_index_inferred=fn_index_inferred, username=username, request=None, ) assert body.request is not None # noqa: S101 root_path = route_utils.get_root_url( request=body.request, route_path="/queue/join", root_path=app.root_path ) try: output = await route_utils.call_process_api( app=app, body=body, gr_request=gr_request, fn_index_inferred=fn_index_inferred, root_path=root_path, ) except Exception as error: show_error = app.get_blocks().show_error or isinstance(error, Error) traceback.print_exc() raise Exception(str(error) if show_error else None) from error # To emulate the HTTP response from the predict API, # convert the output to a JSON response string. # This is done by FastAPI automatically in the HTTP endpoint handlers, # but we need to do it manually here. response_class = app.router.default_response_class if isinstance(response_class, fastapi.datastructures.DefaultPlaceholder): actual_response_class = response_class.value else: actual_response_class = response_class http_response = actual_response_class( output ) # Do the same as https://github.com/tiangolo/fastapi/blob/0.87.0/fastapi/routing.py#L264 # Also, decode the JSON string to a Python object, emulating the HTTP client behavior e.g. the `json()` method of `httpx`. response_json = json.loads(http_response.body.decode()) if not isinstance(response_json, dict): raise ValueError("Unexpected object.") return response_json async def process_events( self, events: list[Event], batch: bool, begin_time: float ) -> None: awake_events: list[Event] = [] fn_index = events[0].fn_index try: for event in events: if event.alive: self.send_message( event, ProcessStartsMessage( eta=self.process_time_per_fn_index[fn_index].avg_time if fn_index in self.process_time_per_fn_index else None ), ) awake_events.append(event) if not awake_events: return try: response = await self.call_prediction(awake_events, batch) err = None except Exception as e: response = None err = e for event in awake_events: self.send_message( event, ProcessCompletedMessage( output={ "error": None if len(e.args) and e.args[0] is None else str(e) }, success=False, ), ) if response and response.get("is_generating", False): old_response = response old_err = err while response and response.get("is_generating", False): old_response = response old_err = err for event in awake_events: self.send_message( event, ProcessGeneratingMessage( output=old_response, success=old_response is not None, ), ) awake_events = [event for event in awake_events if event.alive] if not awake_events: return try: response = await self.call_prediction(awake_events, batch) err = None except Exception as e: response = None err = e for event in awake_events: relevant_response = response or err or old_err self.send_message( event, ProcessCompletedMessage( output={"error": str(relevant_response)} if isinstance(relevant_response, Exception) else relevant_response or {}, success=( relevant_response is not None and not isinstance(relevant_response, Exception) ), ), ) elif response: output = copy.deepcopy(response) for e, event in enumerate(awake_events): if batch and "data" in output: output["data"] = list(zip(*response.get("data")))[e] self.send_message( event, ProcessCompletedMessage( output=output, success=response is not None, ), ) end_time = time.time() if response is not None: self.process_time_per_fn_index[events[0].fn_index].add( end_time - begin_time ) except Exception as e: traceback.print_exc() finally: event_queue = self.event_queue_per_concurrency_id[events[0].concurrency_id] event_queue.current_concurrency -= 1 start_times = event_queue.start_times_per_fn_index[fn_index] if begin_time in start_times: start_times.remove(begin_time) try: self.active_jobs[self.active_jobs.index(events)] = None except ValueError: # `events` can be absent from `self.active_jobs` # when this coroutine is called from the `join_queue` endpoint handler in `routes.py` # without putting the `events` into `self.active_jobs`. # https://github.com/gradio-app/gradio/blob/f09aea34d6bd18c1e2fef80c86ab2476a6d1dd83/gradio/routes.py#L594-L596 pass for event in events: # Always reset the state of the iterator # If the job finished successfully, this has no effect # If the job is cancelled, this will enable future runs # to start "from scratch" await self.reset_iterators(event._id) async def reset_iterators(self, event_id: str): # Do the same thing as the /reset route app = self.server_app if app is None: raise Exception("Server app has not been set.") if event_id not in app.iterators: # Failure, but don't raise an error return async with app.lock: del app.iterators[event_id] app.iterators_to_reset.add(event_id) return