Better handling of large bag files (backport #178)

rqt_bag/src/rqt_bag/bag_timeline.py

@@ -28,8 +28,12 @@
 
 import rosbag2_py
 import time
+<<<<<<< HEAD
 import threading
 
+=======
+from typing import Callable, Iterable, Iterator, Optional, Tuple, Union
+>>>>>>> c7d3efd (Better handling of large bag files (#178))
 
 from python_qt_binding.QtCore import qDebug, Qt, QTimer, qWarning, Signal
 from python_qt_binding.QtWidgets import QGraphicsScene, QMessageBox
@@ -43,6 +47,11 @@
 from .message_loader_thread import MessageLoaderThread
 from .player import Player
 from .recorder import Recorder
+<<<<<<< HEAD
+=======
+from .rosbag2 import Entry, Rosbag2
+from .timeline_frame import TimelineFrame
+>>>>>>> c7d3efd (Better handling of large bag files (#178))
 
 
 class BagTimeline(QGraphicsScene):
@@ -63,6 +72,7 @@ def __init__(self, context, publish_clock):
         super(BagTimeline, self).__init__()
         self._bags = []
         self._bag_lock = threading.RLock()
+        self._bags_size = 0
 
         self.background_task = None  # Display string
         self.background_task_cancel = False
@@ -153,6 +163,7 @@ def add_bag(self, bag):
         :param bag: ros bag file, ''rosbag2.bag''
         """
         self._bags.append(bag)
+        self._bags_size += bag.size()
 
         bag_topics = bag.get_topics()
         qDebug('Topics from this bag: {}'.format(bag_topics))
@@ -182,8 +193,7 @@ def add_bag(self, bag):
             self._timeline_frame.index_cache_cv.notify()
 
     def file_size(self):
-        with self._bag_lock:
-            return sum(b.size() for b in self._bags)
+        return self._bags_size
 
     # TODO Rethink API and if these need to be visible
     def _get_start_stamp(self):
@@ -249,46 +259,46 @@ def get_datatype(self, topic):
                     datatype = bag_datatype
             return datatype
 
-    def get_entries(self, topics, start_stamp, end_stamp):
+    def get_entries(self, topics: Optional[Union[str, Iterable[str]]],
+                    start_stamp: Time, end_stamp: Time,
+                    progress_cb: Optional[Callable[[int], None]] = None) -> Iterator[Entry]:
         """
         generator function for bag entries
         :param topics: list of topics to query, ''list(str)''
         :param start_stamp: stamp to start at, ''rclpy.time.Time''
-        :param end_stamp: stamp to end at, ''rclpy.time,Time''
+        :param end_stamp: stamp to end at, ''rclpy.time.Time''
+        :param progress_cb: callback function to report progress, called once per each percent.
         :returns: entries the bag file, ''msg''
         """
-        with self._bag_lock:
-            bag_entries = []
-            for b in self._bags:
-                bag_start_time = b.get_earliest_timestamp()
-                if bag_start_time is not None and bag_start_time > end_stamp:
-                    continue
+        for b, entry in self.get_entries_with_bags(topics, start_stamp, end_stamp, progress_cb):
+            yield entry
+        return None
 
-                bag_end_time = b.get_latest_timestamp()
-                if bag_end_time is not None and bag_end_time < start_stamp:
-                    continue
-
-                # Get all of the entries for each topic. When opening multiple
-                # bags, the requested topic may not be in a given bag database
-                for topic in topics:
-                    entries = b.get_entries_in_range(start_stamp, end_stamp, topic)
-                    if entries is not None:
-                        bag_entries.extend(entries)
-
-            for entry in sorted(bag_entries, key=lambda entry: entry.timestamp):
-                yield entry
-
-    def get_entries_with_bags(self, topic, start_stamp, end_stamp):
+    def get_entries_with_bags(self, topics: Optional[Union[str, Iterable[str]]],
+                              start_stamp: Time, end_stamp: Time,
+                              progress_cb: Optional[Callable[[int], None]] = None) \
+            -> Iterator[Tuple[Rosbag2, Entry]]:
         """
+<<<<<<< HEAD
         generator function for bag entries
         :param topics: list of topics to query, ''list(str)''
+=======
+        Get a generator of bag entries.
+
+        :param topics: list of topics to query (if None, all topics are used), ''list(str)''
+>>>>>>> c7d3efd (Better handling of large bag files (#178))
         :param start_stamp: stamp to start at, ''rclpy.time.Time''
-        :param end_stamp: stamp to end at, ''rclpy.time,Time''
-        :returns: tuple of (bag, entry) for the entries in the bag file, ''(rosbag2.bag, msg)''
+        :param end_stamp: stamp to end at, ''rclpy.time.Time''
+        :param progress_cb: callback function to report progress, called once per each percent.
+        :returns: tuple of (bag, entry) for the entries in the bag file, ''(rosbag2.Rosbag2, msg)''
         """
         with self._bag_lock:
+<<<<<<< HEAD
             bag_entries = []
             bag_by_iter = {}
+=======
+            relevant_bags = []
+>>>>>>> c7d3efd (Better handling of large bag files (#178))
             for b in self._bags:
                 bag_start_time = b.get_earliest_timestamp()
                 if bag_start_time is not None and bag_start_time > end_stamp:
@@ -298,11 +308,61 @@ def get_entries_with_bags(self, topic, start_stamp, end_stamp):
                 if bag_end_time is not None and bag_end_time < start_stamp:
                     continue
 
-                for entry in b.get_entries_in_range(start_stamp, end_stamp):
-                    bag_entries.append((b, entry))
+                relevant_bags.append(b)
+
+            generators = {}
+            last_entries = {}
+            for b in relevant_bags:
+                generators[b] = b.entries_in_range_generator(start_stamp, end_stamp, topics)
+                try:
+                    last_entries[b] = next(generators[b])
+                except StopIteration:
+                    last_entries[b] = None
+
+            to_delete = []
+            for b in last_entries:
+                if last_entries[b] is None:
+                    to_delete.append(b)
+
+            for b in to_delete:
+                del last_entries[b]
+                del generators[b]
+                relevant_bags.remove(b)
+
+            if progress_cb is not None:
+                progress = 0
+                num_entries = 0
+                estimated_num_entries = 0
+                for b in relevant_bags:
+                    estimated_num_entries += b.estimate_num_entries_in_range(
+                        start_stamp, end_stamp, topics)
+
+            while any(last_entries.values()):
+                min_bag = None
+                min_entry = None
+                for b, entry in last_entries.items():
+                    if entry is not None:
+                        if min_entry is None or entry.timestamp < min_entry.timestamp:
+                            min_bag = b
+                            min_entry = entry
+                if min_bag is None:
+                    return
+
+                if progress_cb is not None:
+                    num_entries += 1
+                    new_progress = int(100.0 * (float(num_entries) / estimated_num_entries))
+                    if new_progress != progress:
+                        progress_cb(new_progress)
+                        progress = new_progress
+
+                yield min_bag, min_entry
+
+                try:
+                    last_entries[min_bag] = next(generators[min_bag])
+                except StopIteration:
+                    last_entries[min_bag] = None
 
-            for bag, entry in sorted(bag_entries, key=lambda item: item[1].timestamp):
-                yield bag, entry
+        return
 
     def get_entry(self, t, topic):
         """

rqt_bag/src/rqt_bag/index_cache_thread.py

@@ -45,30 +45,25 @@ def __init__(self, timeline):
         self.start()
 
     def run(self):
+        # Delay start of the indexing so that the basic UI has time to be loaded
+        time.sleep(2.0)
+
         while not self._stop_flag:
             with self.timeline.index_cache_cv:
                 # Wait until the cache is dirty
                 while len(self.timeline.invalidated_caches) == 0:
                     self.timeline.index_cache_cv.wait()
                     if self._stop_flag:
                         return
-                # Update the index for one topic
-                total_topics = len(self.timeline.topics)
-                update_step = max(1, total_topics / 100)
-                topic_num = 1
-                progress = 0
-                updated = False
-                for topic in self.timeline.topics:
-                    if topic in self.timeline.invalidated_caches:
-                        updated |= (self.timeline._update_index_cache(topic) > 0)
-                    if topic_num % update_step == 0 or topic_num == total_topics:
-                        new_progress = int(100.0 * (float(topic_num) / total_topics))
-                        if new_progress != progress:
-                            progress = new_progress
-                            if not self._stop_flag:
-                                self.timeline.scene().background_progress = progress
-                                self.timeline.scene().status_bar_changed_signal.emit()
-                    topic_num += 1
+
+                # Update the index for all invalidated topics
+                def progress_cb(progress: int) -> None:
+                    if not self._stop_flag:
+                        self.timeline.scene().background_progress = progress
+                        self.timeline.scene().status_bar_changed_signal.emit()
+
+                topics = self.timeline.invalidated_caches.intersection(set(self.timeline.topics))
+                updated = (self.timeline._update_index_cache(topics, progress_cb) > 0)
 
             if updated:
                 self.timeline.scene().background_progress = 0

rqt_bag/src/rqt_bag/rosbag2.py

@@ -31,17 +31,29 @@
 
 from collections import namedtuple
 import os
+from typing import Callable, Iterable, Iterator, List, Optional, Union
 
 from rclpy.clock import Clock, ClockType
 from rclpy.duration import Duration
 from rclpy import logging
 from rclpy.serialization import deserialize_message
+<<<<<<< HEAD
+=======
+from rclpy.time import Time
+
+>>>>>>> c7d3efd (Better handling of large bag files (#178))
 import rosbag2_py
 from rosidl_runtime_py.utilities import get_message
 
+<<<<<<< HEAD
 from rosbag2_py import get_default_storage_id, StorageFilter
 
 WRITE_ONLY_MSG = "open for writing only, returning None"
+=======
+from rqt_bag import bag_helper
+
+WRITE_ONLY_MSG = 'open for writing only, returning None'
+>>>>>>> c7d3efd (Better handling of large bag files (#178))
 
 Entry = namedtuple('Entry', ['topic', 'data', 'timestamp'])
 
@@ -152,25 +164,111 @@ def get_entry_after(self, timestamp, topic=None):
         self.reader.reset_filter()
         return result
 
-    def get_entries_in_range(self, t_start, t_end, topic=None):
+    def get_entries_in_range(self, t_start: Time, t_end: Time,
+                             topic: Optional[Union[str, Iterable[str]]] = None,
+                             progress_cb: Optional[Callable[[int], None]] = None) \
+            -> Optional[List[Entry]]:
+        """
+        Get a list of all entries in a given time range, sorted by receive stamp.
+
+        Do not use this function for large bags. It will load all entries into memory. Use
+        entries_in_range_generator() instead and process the data as they are returned.
+
+        :param t_start: stamp to start at, ''rclpy.time.Time''
+        :param t_end: stamp to end at, ''rclpy.time.Time''
+        :param topic: topic or list of topics to query (if None, all topics are), ''list(str)''
+        :param progress_cb: callback function to report progress, called once per each percent.
+        :returns: entries in the bag file, ''list(Entry)''
+        """
         if not self.reader:
             self._logger.warn("get_entries_in_range - " + WRITE_ONLY_MSG)
             return None
 
+        return list(self.entries_in_range_generator(t_start, t_end, topic, progress_cb))
+
+    def entries_in_range_generator(self, t_start: Time, t_end: Time,
+                                   topic: Optional[Union[str, Iterable[str]]] = None,
+                                   progress_cb: Optional[Callable[[int], None]] = None) \
+            -> Iterator[Entry]:
+        """
+        Get a generator of all entries in a given time range, sorted by receive stamp.
+
+        :param t_start: stamp to start at, ''rclpy.time.Time''
+        :param t_end: stamp to end at, ''rclpy.time.Time''
+        :param topic: topic or list of topics to query (if None, all topics are), ''list(str)''
+        :param progress_cb: callback function to report progress, called once per each percent.
+        :returns: generator of entries in the bag file, ''Generator(Entry)''
+        """
+        if not self.reader:
+            self._logger.warn('entries_in_range_generator - ' + WRITE_ONLY_MSG)
+            return
+
+        if isinstance(topic, Iterable) and not isinstance(topic, str):
+            topics = topic
+        else:
+            topics = [topic] if topic is not None else []
+
         self.reader.set_read_order(rosbag2_py.ReadOrder(reverse=False))
-        self.reader.set_filter(rosbag2_py.StorageFilter(topics=[topic] if topic else []))
+        self.reader.set_filter(rosbag2_py.StorageFilter(topics=topics))
         self.reader.seek(t_start.nanoseconds)
-        entries = []
+        if progress_cb is not None:
+            num_entries = 0
+            progress = 0
+            estimated_num_entries = self.estimate_num_entries_in_range(t_start, t_end, topic)
+
         while self.reader.has_next():
             next_entry = self.read_next()
             if next_entry.timestamp <= t_end.nanoseconds:
-                entries.append(next_entry)
+                if progress_cb is not None:
+                    num_entries += 1
+                    new_progress = int(100.0 * (float(num_entries) / estimated_num_entries))
+                    if new_progress != progress:
+                        progress_cb(new_progress)
+                        progress = new_progress
+                yield next_entry
             else:
                 break
 
         # No filter
         self.reader.reset_filter()
-        return entries
+
+        if progress_cb is not None and progress != 100:
+            progress_cb(100)
+
+        return
+
+    def estimate_num_entries_in_range(self, t_start: Time, t_end: Time,
+                                      topic: Optional[Union[str, Iterable[str]]] = None) -> int:
+        """
+        Estimate the number of entries in the given time range.
+
+        The computation is only approximate, based on the assumption that messages are distributed
+        evenly across the whole bag on every topic.
+
+        :param t_start: stamp to start at, ''rclpy.time.Time''
+        :param t_end: stamp to end at, ''rclpy.time.Time''
+        :param topic: topic or list of topics to query (if None, all topics are), ''list(str)''
+        :returns: the approximate number of entries, ''int''
+        """
+        if not self.reader:
+            self._logger.warn('estimate_num_entries_in_range - ' + WRITE_ONLY_MSG)
+            return 0
+
+        if isinstance(topic, Iterable) and not isinstance(topic, str):
+            topics = topic
+        else:
+            topics = [topic] if topic is not None else []
+
+        range_duration = t_end - t_start
+        bag_duration = self.get_latest_timestamp() - self.get_earliest_timestamp()
+        fraction = bag_helper.to_sec(range_duration) / bag_helper.to_sec(bag_duration)
+
+        num_messages = 0
+        for t_info in self.metadata.topics_with_message_count:
+            if t_info.topic_metadata.name in topics:
+                num_messages += t_info.message_count
+
+        return int(fraction * num_messages)
 
     def read_next(self):
         return Entry(*self.reader.read_next())