(I asked this question in robotics.stackexchange.com but did not received any answer! That is why asking in here with some modifications.) I have the following ROS 2 scripts for a publisher node:

import rclpy
from rclpy.node import Node
from std_msgs.msg import String
import time
class MinimalPublisher(Node):
    def __init__(self):
        super().__init__('minimal_publisher')
        self.publisher_ = self.create_publisher(String, 'topic', 1)
        self.timer1 = self.create_timer(1, self.timer1_callback)
        self.timer2 = self.create_timer(2, self.timer2_callback)
        self.timer3 = self.create_timer(3, self.timer3_callback)
        self.i = 1
        self.j = 1
        self.k = 1
    def timer3_callback(self):
        msg = String()
        msg.data = 'timer3:  %d' % self.k
        self.publisher_.publish(msg)
        self.get_logger().info('Publishing: "%s"' % msg.data)
        self.k += 1
    def timer2_callback(self):
        msg = String()
        msg.data = 'timer2: %d' % self.j
        self.publisher_.publish(msg)
        self.get_logger().info('Publishing: "%s"' % msg.data)
        self.j += 1
    def timer1_callback(self):
        msg = String()
        msg.data = 'timer1: %d' % self.i
        self.publisher_.publish(msg)
        self.get_logger().info('Publishing: "%s"' % msg.data)
        self.i += 1
        # self.create_rate(4).sleep()

def main(args=None):
    rclpy.init(args=args)
    minimal_publisher = MinimalPublisher()
    time.sleep(2)
    msg = String()
    msg.data = 'Publishing garbage 1'
    minimal_publisher.publisher_.publish(msg)
    msg = String()
    msg.data = 'Publishing garbage 2'
    minimal_publisher.publisher_.publish(msg)
    msg = String()
    msg.data = 'Publishing garbage 3'
    minimal_publisher.publisher_.publish(msg)
    msg = String()
    msg.data = 'Publishing garbage 4'
    minimal_publisher.publisher_.publish(msg)
    msg = String()
    msg.data = 'Publishing garbage 5'
    minimal_publisher.publisher_.publish(msg)
    msg = String()
    msg.data = 'Publishing garbage 6'
    minimal_publisher.publisher_.publish(msg)    
    time.sleep(3)
    # rate = minimal_publisher.create_rate(5)
    # rate.sleep()
    try:
        rclpy.spin(minimal_publisher)
    except KeyboardInterrupt:
        pass    
    # minimal_publisher.destroy_node()
    # rclpy.shutdown()

if __name__ == '__main__':
    main()

I used timer to hold after publishing some garbage messages, and then started spinning using 3 timers. I have a simple subscriber node as follows:

import rclpy
from rclpy.node import Node
from std_msgs.msg import String
import time

class MinimalSubscriber(Node):
    def __init__(self):
        super().__init__('minimal_subscriber')
        self.subscription = self.create_subscription(String,'topic',self.listener_callback, 4)
        self.subscription  # prevent unused variable warning

    def listener_callback(self, msg):
        self.get_logger().info('I heard: "%s"' % msg.data)

def main(args=None):
    rclpy.init(args=args)
    minimal_subscriber = MinimalSubscriber()
    # delay = minimal_subscriber.create_rate(2.5)
    rclpy.spin(minimal_subscriber)

if __name__ == '__main__':
    main()

I used a launch file to run both publisher and subscriber at once. I did some experiments by varying the publisher and subscriber queue size in the above code in order to understand topic size, publisher queue, subscriber queue as follows:

Experiment 1: pub queue size 1, sub queue size 1. Results: Only the 6th garbage message is shown. After that every message is received from all 3 timers (5 garbage messages lost)

Experiment 2: pub queue size 100, sub queue size 1. Results: EXACTLY SAME AS EXPERIMENT 1 (also checked for pub size 10, 5, 6 and results are same)

Experiment 3: pub queue size 1, sub queue size 100. Results: All 6 garbage messages are shown (nothing lost). After that every message is received from all 3 timers.

Experiment 4: pub queue size 1, sub queue size 4. Results: Last 4 garbage messages are shown (First two messages are lost!). After that every message is received from all 3 timers.

It seems like only the subscriber queue size is the one that matters. The publisher size does not matter at all. I haven't been able to generate any case where increasing the publisher size when the subscriber size is fixed (say 1) is making any difference! Can anybody explain? Moreover, does the topic has any internal size, or only the publisher and subscriber sizes are the one that exists.