ensuring ES gracefully finish and save metadata in every system shutdown

Yahmez

@aphex The only thing I did was make the killes.service script executable. The code you linked is the shutdown script I am using. Although I am assuming you made the modifications to the NESPi as outlined in my PDF...

meleu

@yahmez said in ensuring ES gracefully finish and save metadata in every system shutdown:

The only thing I did was make the killes.service script executable

The script is killes.sh, and, yes, it must be executable. There's nothing wrong with the .service file. ;-)

Yahmez

@meleu
Ha, sorry. My error.

cyperghost

@meleu As I mentioned in my latest script!
To trigger ES events we can use incron job.

you can set it up as I done in my github script and maybe you can add a new user (afaik you are not using user PI on your system)

Then there can be two ways! To locate in which status ES shutdown now we have to use a kind of logger as incron just writes events to syslog....

So just write echo $# in incronjob results in an entry of syslog...

So we would write this to icrontab... maybe sudo?? sudo /etc/log.sh $#

/tmp IN_CREATE /etc/log.sh $#

log.shwill contain the redirection from incron

#!/bin/bash
# incron-logger to determine EmulationStation-quit-status
    esstatus=("es-restart" "es-sysrestart" "es-shutdown")

    for status in "${esstatus[@]}"; do
               [[ "$status" == "$1" ]] && echo "$1" > /etc/esstatus.log
    done

Important! That needs a bit of fine tuning! Because every file written will be logged! Unix uses tons of temporary files ;)
FIXED!

EDIT:

/tmp IN_CREATE cp $@$# /dev/shm/

would be much nicer ... then presence of ES status files have just to be tested ;) As these files are zero files, there is no chance to break copy process. And /dev/shm is "self cleaning" after shutdown ;)

Then we need to modify @meleu s script to determine the generated logfile and with this we can generate action based on ES-quit-status ;)

The solution I presnted yesterday lacks of speed, it works on a Raspberry 3 but maybe if the RPi 4 is available the shutdown is performed so fast my script will just terminated without action!

So I think systemd gots a more bright future.

@meleu
It's your turn now ;)
I hope that this gives you encourage to modify your script. Me and lot's of users will appreciate this. As I did not care about user rights managment, there maybe must be made sudo calls from time to time.

The king way would be to integrate the logger into the killes.sh and I vote for a rename of your script.... it sounds so destructive ;)

meleu

Hey bro. I appreciate your effort and I'm sure you're enjoying to learn all this stuff while working on your solution. Sadly I'm currently I'm away from my pi to test this stuff.

But let's just try to put very clear what problem we are trying to solve. As far as I understood we want to detect if the system is going to actually poweroff or just reboot. Is it right?

cyperghost

@meleu Yes...
Then we can set a trigger to this transistor GPIO

But what happens then? Will be a "second" shutdown performed? Because the triggered transistor will act as button press and call the regular shutdown script. Therefore I made a difference between software and hardware based actions ;) What's your opinion about?

What happens if you press shutdown button in ES? Then also a es-shutdown will be generated and the killes script will be triggered ;)

So you see it's a challenging game ;) And you see why the script from yesterday is so endless long. So yes, we need a to differ between shutdown and reboot and what initiated it? software or hardware.

The easiest solution seems to just to edit the emulationstation.sh and set the calls here but that's far away from generic solution.

But as final note:

Personally it's not my responsibility to offer scripts here and 
to satisfy someones setup. It's an encouragement to learn something 
new and to share this knowledge for free as volunteer.

Furthermore:
I see this as interaction between enthusiasts to make a great development even
better.

caver01

@cyperghost @meleu You guys are amazing! I really appreciate the work you put into this.

You guys have already vastly improved the button-triggered shutdown by re-writing the script to ensureES saves metadata. That alone is a very nice improvement. However, there has always been an unrelated problem with the Mausberry circuit: it simply ignores software shutdowns/reboots. In order for it to cut power it requires a physical button press. The fact that it ignores a reboot is fine--might even be designed that way. But software shutdown puts the circuit into a locked state and does not cut power.

For me, the goal was to fix a locking Mausberry circuit on soft shutdown. @meleu's service idea not only simplified the Mausberry script allowing us to revert back to the original version, but it also fixed the ES metadata problem for EVERY CIRCUIT OR SHUTDOWN SOLUTION. I love this because it takes the solution public, so to speak. It works no matter how you are triggering your shutdown.

Of course, it also gave me a place to insert something specific--a GPIO call to trigger my transistor. Doing so, now eliminates the Mausberry lockup problem. It does it at the expense of rebooting (make all reboots lead to poweroff), but this is a good compromise in my opinion.

If I could wrap that call inside a conditional to some system-level indicator to detect a shutdown (and ignore a reboot) my transistor trigger would be perfect.

@cyperghost's idea goes a long way toward detecting reboots from ES, but what about a reboot from the RetroPie config menus? What about SSH and a sudo reboot issued from commandline?

I love the ideas here. This puzzle has been a real challenge to enjoy! Thanks for all of the input.

cyperghost

OUTDATED: Take a look to this code piece

@caver01

but what about a reboot from the RetroPie config menus? What about SSH and a sudo reboot issued from commandline?

That's the issue I was working on. As long as you set a GPIO command in the killes.sh script as long it is triggered on every reboot and shutdown. A reboot is important... Just to test scripts and to test interaction.

So my bad workaround I would suggest is. Edit the /opt/retropie/supplementary/emulationstation/emulationstation.shand add something like

rm -f /tmp/es-shutdown && touch /dev/shm/es-shutdown

and then in killes.sh

[[ -f /dev/shm/es-shutdown ]] && trigger GPIO

That's the easiest way without installing additional services, but will work only for ES or you need to remember to add file es-sysrestart to location that needs to be detected and then perform a reboot.

Of course, it also gave me a place to insert something specific--a GPIO call to trigger my transistor. Doing so, now eliminates the Mausberry lockup problem. It does it at the expense of rebooting (make all reboots lead to poweroff), but this is a good compromise in my opinion.

Yes this solution is so far the best one - indeed!

If I could wrap that call inside a conditional to some system-level indicator to detect a shutdown (and ignore a reboot) my transistor trigger would be perfect.

That's my hope... Afaik reboot and poweroff use different runlevels to terminate programs. If there is a chance to detect runlevel for poweroff (0) or to detect runlevel for reboot (6) then there is a generel solution available.

meleu

The runcommand.sh script changed a bit 3 days ago and users that are uptodate maybe are facing those issues with metadata not being saved when powering off while there's an emulator running.

I'm already taking a look at it. Wait a little. ;-)

meleu

UPDATE

I've updated the killes.sh script in the OP. The issue I talked about in the post above seems to be fixed.

Let me know if you guys find some problem.

Cheers!

snarc

@meleu

Thanks for your awesome shutdown script. It works fantastic! :D

I only have 2 minor suggestions for improvement:

#1 I think it would be better if you add a "sudo" to "chmod a+x /etc/killes.sh" because some people will encouter a "permission denied error" without the sudo command.

#2 In the commented code in your 2nd post is a tiny error: "ExecStop=/home/pi/bin/killes.sh" hast to be replaced by "ExecStop=/etc/killes.sh" ;)

Regards!

Mafu

I'm using a Nespi Case with a mosfet switch mod and this works excellent for shutting down the system or using the reset button to do a full reset of the system. Currently however I am trying to get the reset button to just go back to the emulation station from a game but it is not saving the data. Is there any way to cannablize part of your method to save the game data and patch it into the code the maker of the mod I'm using provided? My current solution is to have retroarch save the game every few seconds which can't be very healthy for my sd card.

Sorry if the answer is obvious, I am a complete noob at python. (code below)

#!/usr/bin/python
import RPi.GPIO as GPIO
import os, time

GPIO.setmode(GPIO.BCM)
GPIO.setup(23, GPIO.IN, pull_up_down=GPIO.PUD_UP) #Reset switch
GPIO.setup(24, GPIO.IN, pull_up_down=GPIO.PUD_UP) #Power switch
GPIO.setup(25, GPIO.OUT) #ON control
GPIO.output(25, GPIO.HIGH)

def exitEmulator(channel):
print('exitEmulator')
pids = [pid for pid in os.listdir('/proc') if pid.isdigit()]

for pid in pids:
    try:
        commandpath = open(os.path.join('/proc', pid, 'cmdline'), 'rb').read()
        if commandpath[0:24] == '/opt/retropie/emulators/':
            os.system('kill -QUIT %s' % pid)
            print('kill -QUIT %s' % pid)
    except IOError:
        continue

GPIO.add_event_detect(23, GPIO.FALLING, callback=exitEmulator, bouncetime=500)

while True:
if (GPIO.input(24)):
time.sleep(0.25)
else:
print ("Shutting down...")
os.system("sudo shutdown -h now")
break

Sano

I would suggest to use Network Control Interface to take a savestate and properly quit Retroarch.
See here : https://buildbot.libretro.com/.docs/tech/network-control-interface/
Edit : yes I'm fond of this feature since I discovered it :)

Mafu

The reset button I have on my case is currently being controlled by the script I posted. How would I integrate the Network Control Interface with it?

caver01

@mafu it looks like, you would need to enable the option in retroarch.cfg, then insert a command such as echo -n "SAVE_STATE" | nc -u -w1 127.0.0.1 55355 before your kill command which would tell retroarch to save the game. Of course, this is limited to retroarch emulators only. If you were, for example, running AdvanceMAME, it would probably generate an error.

Mafu

I think we're talking past one another, I can't find any information on echo in python. I gave it a shot and put it in the script anyway but it just crashed the console instead of safe shutdown of the power button and did nothing when using the reset.

caver01

@mafu Sorry, I am not a python expert either, but what about borrowing the example from the script's working shutdown command and doing this:
os.system("echo -n "SAVE_STATE" | nc -u -w1 127.0.0.1 55355")

Mafu

That was an excellent idea carver01 (I facepalmed a little when I saw it for not thinking of it myself). It works partially but only on save states, not ingame save data (games that have their own dedicated save feature).

Whats frustrating is when using the OP's solution and just shutting the system down with the power button it will

Exit the game.
pause at the game selection screen (I assume this is where the "sleep 5" comes into play).
And then shut down.

What I need is a way to get it to stop at step 2 and still save the game's data(save files, save states, etc) when pressing the reset button. As it is now it goes back to the game selection screen but does not save any data.

zobilla

Hey guys

Anyone having problems with systemd refusing to load with no ExecStart line in the service?

--
pi@retropie:~ $ dmesg | grep killes
[ 2.660196] systemd[1]: killes.service lacks ExecStart setting. Refusing.
[ 2.679020] systemd[1]: Cannot add dependency job for unit killes.service, ignoring: Unit killes.service failed to load: Invalid argument. See system logs and 'systemctl status killes.service' for details.

I pulled the logs from systemctl status killes.service but it just says invalid args.

Sano

@mafu Could you precise what you mean by "save the game data" ?
Is it that .srm files are not correctly written to the sd when retroarch is exited with the script ?

It seems to me very difficult to script game saves beyond the savestates as explained above, because each game is quite specific about it's own saves (different menu, number of save slots, use of savepoints...)