Database Reference.listen() is very slow for large responses

Comienza con la implementación de sseclient del Firebase Admin SDK utilizada por Reference.listen() y, después, compara su comportamiento al leer respuestas con el de btubbs/sseclient.py, enlazado. Reproduce el problema usando la prueba proporcionada para Raspberry Pi y confirma que las respuestas grandes de listen() ya no tardan unos 10 segundos ni utilizan casi el 100 % de la CPU, mientras que el callback sigue recibiendo el evento completo.

Describe your environment

• Operating System version: Linux raspberrypi 4.19.97+ (On a Raspberry Pi Zero W)
• Firebase SDK version: 5.2.0
• Firebase Product: database
• Python version: 3.7.3
• Pip version: 18.1

Describe the problem

Reading a large event in response to calling Reference.listen() is extremely slow on a Raspberry Pi Zero. This looks very similar to https://github.com/nhorvath/Pyrebase4/issues/45 and I noticed that the sseclient here is based on that from Pyrebase.

From reading the sseclient code, the Firebase Admin SDK seems to have solved the quadratic regex scanning issue that the sseclient in Pyrebase4 has, but it still reads the response one character at a time, which I suspect is the root issue. It looks like this is solved in https://github.com/btubbs/sseclient/blob/master/sseclient.py by reading in chunks when short reads are supported.

Steps to reproduce:

Modify the test code below to add a working databaseURL and run it. On my Raspberry Pi Zero W, it outputs:

0 ms: Initializing app...
1 ms: Getting DB reference...
17 ms: Putting 20000 bytes...
1122 ms: Getting data...
1347 ms: Got 20000 bytes with get().
1349 ms: Listening for data...
11686 ms: Got 20000 bytes with listen().

That is, listen() takes about 10 000 ms (with close to 100% CPU usage) to get and read the response, while get() takes 200 ms.

Relevant Code:

import firebase_admin
import firebase_admin.db
import sys
import time

count = 2000
nonsense = "abcedefghi"
start_time = time.time()

def log(msg):
  t = int(1000 * (time.time() - start_time))
  print(str(t) + " ms: " + msg)

log("Initializing app...")
firebase_admin.initialize_app(options={'databaseURL': 'https://XXXXXXXXXX.firebaseio.com/'})

log("Getting DB reference...")
ref = firebase_admin.db.reference("bringert_test_streaming_speed")

data = nonsense * count

log("Putting " + str(len(data)) + " bytes...")
new_data=ref.push(data)

log("Getting data...")
log("Got " + str(len(new_data.get())) + " bytes with get().")

def callback(event):
  if event.event_type == "put":
    log("Got " + str(len(event.data)) + " bytes with listen().")
    sys.exit(0)

log("Listening for data...")
new_data.listen(callback)