Running multiple mpv instances #126
Comments
|
You can create multiple MPV instances in the same thread. The observers are called from a per-instance event handling thread that is automatically spawned in the background when you construct the instance. Here's an example of running two at the same time: #!/usr/bin/env python3
import time
import sys
import mpv
player1 = mpv.MPV(log_handler=print, loglevel='debug', vo='x11')
time.sleep(0.2)
player2 = mpv.MPV(log_handler=print, loglevel='debug', vo='x11')
player1.loop = True
player2.loop = True
player1.play('test.webm')
time.sleep(0.2)
player2.play('test.webm')
def make_observer(player_name):
def observer(_prop_name, prop_val):
print(f'{player_name}: {prop_val}')
return observer
player1.observe_property('time-pos', make_observer('player1'))
player2.observe_property('time-pos', make_observer('player2'))
player1.wait_for_playback()Testing this I got some crashes in the default opengl vo when launching both players at the same time, but I suspect that is an upstream bug specific to my setup. |
|
Thanks again. I'll see if I can get this running. I'm writing a test application to see if new Wi-Fi 7 APs and routers are really better at handling multiple video streams. |
|
Instead of actually playing these streams you could just dump them to disk using mpv, then analyze variables like momentary throughput and packet latency using tcpdump/wireshark. |
|
That's good thought. Could I retrieve video stats too like frame and bit rate and A/V sync? I've got four players launched so far using your suggested code. I'm working through getting both the player log messages and decorated property observer output also written to a log per player. Your interface and help have been invaluable! Thank you! |
|
I'm trying to write a log file per player. Do I need to define a loghandler per player or is there a way I can pass a logfile name to one handler? Or is it easier to direct all player logging to a single file, with logging tagged with a player ID? |
You could e.g. use from functools import partial
def log_handler(logfile, level, prefix, text):
print(level, prefix, text, file=logfile)
some_logfile = open('some.log', 'w+')
player = mpv.MPV(log_handler=partial(log_handler, some_logfile), loglevel='debug') |
Bitrate should definitely be possible with wireshark's statistics. Does A/V sync depend on the network? I thought you'd transport some container format over RTSP and that would have a fixed A/V sync? If audio and video are separate RTSP streams you might be able to measure that if there are some frame numbers/timestamps in the packet that wireshark can understand. |
|
Thanks for both replies. Both are helpful suggestions. |
|
Having a problem implementing the partial log_handler. Here's the code When I run it, mpv errors all are sent to player_0_vidtest.log. What did I do wrong? |
The code you posted is fine! I added the missing parts and got the script below. The output is as expected. I'm guessing you maybe have a global somewhere that's shadowing something? Are you testing this as a script or in an interactive interpreter? If you're testing this in ipython or similar make sure you're actually running the code you posted by restarting the session. #!/usr/bin/env python3
import mpv
import atexit
import time
import datetime
import functools
nice_cols = [49, 118, 27, 196, 208, 206, 229, 87]
@functools.lru_cache(maxsize=None)
def col(seed):
global nice_cols
rv, *nice_cols = nice_cols
nice_cols = [*nice_cols, rv]
return rv
def logtofile(logfile, level, prefix, text):
# print to terminal
print(f'{logfile} {datetime.datetime.now():%Y-%m-%d %H:%M:%S} \033[38;5;246m{level:>8} \033[38;5;{col(prefix)}m{prefix}:\033[0m {text}')
# print to logfile
print(f'{logfile}{datetime.datetime.now():%Y-%m-%d %H:%M:%S} [{level}] {prefix}: {text}', file=logfile)
# Create players and logfiles
logfilelist = []
vidplayerlist = []
for n in range(5):
print('>>> Creating player %s logfile' % n)
logfilename = '/tmp/player_'+str(n)+'_vidtest.log'
logfile = open(logfilename, 'w+')
atexit.register(logfile.close)
logfilelist.append(logfile)
print('>>> Creating player %s' % n)
playerInstance = mpv.MPV(log_handler=functools.partial(logtofile,logfile), loglevel='warn')
playerInstance.window_scale = 0.25
vidplayerlist.append(playerInstance)
print('>>> Starting player %s' % n)
playerInstance.play('test.webm')
time.sleep(2)Output: dev-pypi~/python-mpv <3 python3 demo6.py master
>>> Creating player 0 logfile
>>> Creating player 0
>>> Starting player 0
<_io.TextIOWrapper name='/tmp/player_0_vidtest.log' mode='w+' encoding='UTF-8'> 2020-07-21 18:45:20 error file: Cannot open file 'test.webm': No such file or directory
<_io.TextIOWrapper name='/tmp/player_0_vidtest.log' mode='w+' encoding='UTF-8'> 2020-07-21 18:45:20 error stream: Failed to open test.webm.
>>> Creating player 1 logfile
>>> Creating player 1
>>> Starting player 1
<_io.TextIOWrapper name='/tmp/player_1_vidtest.log' mode='w+' encoding='UTF-8'> 2020-07-21 18:45:22 error file: Cannot open file 'test.webm': No such file or directory
<_io.TextIOWrapper name='/tmp/player_1_vidtest.log' mode='w+' encoding='UTF-8'> 2020-07-21 18:45:22 error stream: Failed to open test.webm.
>>> Creating player 2 logfile
>>> Creating player 2
>>> Starting player 2
<_io.TextIOWrapper name='/tmp/player_2_vidtest.log' mode='w+' encoding='UTF-8'> 2020-07-21 18:45:24 error file: Cannot open file 'test.webm': No such file or directory
<_io.TextIOWrapper name='/tmp/player_2_vidtest.log' mode='w+' encoding='UTF-8'> 2020-07-21 18:45:24 error stream: Failed to open test.webm.
^CTraceback (most recent call last):
File "demo6.py", line 40, in <module>
time.sleep(2)
KeyboardInterrupt
dev-pypi~/python-mpv <3 cat /tmp/player_0_vidtest.log (1) master
<_io.TextIOWrapper name='/tmp/player_0_vidtest.log' mode='w+' encoding='UTF-8'>2020-07-21 18:45:20 [error] file: Cannot open file 'test.webm': No such file or directory
<_io.TextIOWrapper name='/tmp/player_0_vidtest.log' mode='w+' encoding='UTF-8'>2020-07-21 18:45:20 [error] stream: Failed to open test.webm.
dev-pypi~/python-mpv <3 cat /tmp/player_1_vidtest.log master
<_io.TextIOWrapper name='/tmp/player_1_vidtest.log' mode='w+' encoding='UTF-8'>2020-07-21 18:45:22 [error] file: Cannot open file 'test.webm': No such file or directory
<_io.TextIOWrapper name='/tmp/player_1_vidtest.log' mode='w+' encoding='UTF-8'>2020-07-21 18:45:22 [error] stream: Failed to open test.webm. |
|
I'm testing as a script. I think I found a problem. When running the script, my streaming source failed. Note the difference in io.TextIOWrapper names for errors from ffmpeg/demuxer and the other sources. |
|
Were you able to determine why ffmpeg doesn't use the correct io.TextIOWrapper name? |
|
No, I tried reproducing it using your code but with the script I put it into everything worked. It must have been something outside of the code you posted above. |
|
Could it be ffmpeg version and/or python-mpv version? I'm using 0.4.0 due to issue 116 |
|
That should definitely not cause this kind of behavior. |
|
Sorry for the long post. tldr; I suspect log_handler is not properly handling multiple instances. No matter what I try, ffmpeg errors are always logged to the first log_handler instance. I took your demo6.py script, ran it on my system and got the same result. File 'test.webm' doesn't exist on my system either. I then changed playerInstance.play to play a valid rtsp stream and saw the same result as before; ffmpeg errors are written to player_0_vidtest.log. I then modified the script to print logfile value in the for loop and again in logtofile: When I try to play a nonexistent file, the _io.TextIOWrapper name values match in both print statements. When I play a valid file, they don't. But all errors come from ffmpeg. I then changed the script to eliminate functools.partial. I created multiple logtofile functions, changing the print statement to include the function name, i.e. All errors still go to the first logfile. I know both players are getting errors because I see them when I launch multiple mpv players manually. |
|
Since you can't reproduce this, it's probably time to close this bug. |
|
ffmpeg has a global log callback, so the first mpv instance or so gets all ffmpeg log messages. This is a problem that needs to be fixed in ffmpeg itself. |
|
Thanks wm4. I thought something like that was happening. But I don't know why jaseg wasn't seeing the same behavior. Maybe this has already been addressed in a newer version of ffmpeg than I'm using? |
|
@smallnetbuilder I probably was not seeing the issue since I did not get any ffmpeg errors. mpv's regular output gets redirected as intended, after all. You might be interested in what we're doing in the unit tests to silence libmpv/ffmpeg: We |
|
Thanks for the follow up @jaseg |
I need to run multiple instances of mpv. Each player will have its own property observer that I would like to log to a separate file.
I think I need to do this by putting the player and its decorated property observer in a separate thread.
Is the ability to do this built into python-mpv or does it need to be done in the script that calls your interface?
The text was updated successfully, but these errors were encountered: