NSIS uninstaller checks the app is still running and kills it by nsProcess::KillProcess before deleting the installed files. However, nsProcess::KillProcess seems neither emitting any window-alll-closed and quit events, nor kills child (and grand child) processes. Hence, child processes spawned by the app keep running and block files are deleted.
Using taskkill instead of nsProcess::KillProcess might solve this problem.
I found some related discussions:
Nice catch, thanks a lot.
Since Electron in any case requires Windows 7, taskkill can be used.
Thank you for the quick response!
@develar @jkawamoto hiļ¼This change brought a new bug. It happens when autoupdate ļ¼ātaskkill /f /t ā ļ¼This will kill the installation processļ¼Because the installation process is the main application of the child processes.
As shown:
pid-2028 is a child processes of pid-3044ļ¼pid-2028 is installation processļ¼pid-3044 is main application


@develar hiļ¼I tested it 20.4.0ļ¼It still does not working. if change 'taskkill /f /t /im' to 'taskkill /f /im',it will working.
Without \t option, I think it wouldnāt fix this issue.
I wonder why the installation process is still a child process of the main process even if it was spawned with {detached: true, stdio: āignoreā}.
@jkawamoto @develar
set {detached: true, stdio: āignoreā}ļ¼This just means āOn Windows, setting options.detached to true makes it possible for the child process to continue running after the parent exits. The child will have its own console window.ā,But if you kill the process treeļ¼with āātaskkill /tāāļ¼ļ¼all the processes will be killed ļ¼because the parent-child relationship between processes will not changeļ¼
moreoverļ¼āātaskkill /tāā a is higher than "PID ne $pid" of priority,so it will still kill installation process.
As shown:
pid-4816 is a child processes of pid-4046ļ¼pid-4816 is installation processļ¼pid-4046 is main application,and "儽čÆåŗ.exe" is APP_EXECUTABLE_FILENAME.
pid-4816, pid-4046, pid-7512ļ¼another child processļ¼ ļ¼They are all killed

Hi @develar ,
I am still seeing the problem of installer apparently being killed during autoupdate with 20.4.1.
Caveat:
2 years or so ago the installer didnt finish a running application on its own yet so I had added the following as custom NSIS include do do this from my end:
!macro customInit
; SHUT DOWN APP IF CURRENTLY RUNNING
${GetProcessInfo} 0 $0 $1 $2 $3 $4
${if} $3 != "${APP_EXECUTABLE_FILENAME}"
${nsProcess::FindProcess} "${APP_EXECUTABLE_FILENAME}" $R0
${If} $R0 == 0
;MessageBox MB_OK "App currently running - going to shutdown to install new version"
${nsProcess::CloseProcess} "${APP_EXECUTABLE_FILENAME}" $R0
Sleep 5000
${nsProcess::KillProcess} "${APP_EXECUTABLE_FILENAME}" $R0
Sleep 3000
${EndIf}
${nsProcess::Unload}
${endIf}
!macroend
this led to the error message
Plugin not found, cannot call ${nsProcess::FindProcess}
Error in macro customInit on macroline 4
Error in script "<stdin>" on line 131 -- aborting creation process
⢠exited command=makensis.exe code=1 pid=1040
⢠async task error error=
Error: \tmp-electron-builder\nsis\nsis-3.0.1.13\Bin\makensis.exe exited with code 1
at ChildProcess.childProcess.once.code (C:\cygwin64\home\sshd\builds\code\screenawar
a few days ago (while having worked fine for close to 2 years).
So I took it out and now see the installer closing the original running install and not installing the new version (no logs being produced).
I.e.: not sure if it is a coincidence (me removing the custom nsis exit-running-app code while this ticket was coming up) or whether the fix 12 hours ago is maybe not complete yet.
Gladly let me know if there is any other info I could provide to shed some light on this.
nsProcess Is not bundled now. Thanks for code snippetā it seems you use more longer timeouts.
@consense "Close" here means that on kill process windows will be closed. I think, it helps app to exit gracefully. But in our case I think proper solution will be still not use nsProcess plugin, but instead try to ask process to exit gracefully.
possible fix:
/f (force) and /t (tree kill)tree kill is not used now ā now app is killed softly at first and it should be enough to terminate child process.
It seems like the installer doesn't kill grandchildren processes. Is this an expected behavior?
@develar so here is our problem:
We spawn LBRY.exe which spawns child process for our daemon, which is lbrynet-daemon.exe . When lbrynet-daemon.exe is spawned, it creates a shell and spawns the actual lbrynet-daemon.exe, which is now a child of the child (or grandchild if you will). When we run the upgrade (or the installer when the app is running), it kills LBRY.exe and the shell lbrynet-daemon.exe, which leaves the grandchild process running. We are still unsure if the current behavior of killing the shell process not killing the child is correct (need to do more research).
Can you share your thoughts on this? I believe this would be fixed by running taskkill /t, but you've stated before that this might kill the installer itself? Is this still the case? In https://github.com/electron-userland/electron-builder/issues/2516#issuecomment-372009092 you alluded to waiting 2 seconds and then killing with /t - don't think this would help us as LBRY.exe would already be killed, but I'm not 100% sure.
We'd be willing to tip if the builder can support our scenario somehow.
Did a little more digging and this only happens when running the installation manually while our app is running. It seems like the child processes and grandchild processes are killed during an in app upgrade, but we did notice that we get 2 UAC alerts to run the updated installer, once with a --updated --force-run parameter and then a second time with --updated /S . When the 2nd UAC prompt comes up, the first is already installing, so if you wait and then proceed when the app is starting/started, it kills it again and then doesn't restart and leaves a child lbrynet-daemon.exe process running.
Can the task killing process (from in-app upgrade process) be replicated when running the installer manually and choosing the "LBRY is running, click OK to close it" option?
We'll try to dig into this behavior a bit more also.
As @IDrinkMoreWater explained at https://github.com/electron-userland/electron-builder/issues/2516#issuecomment-371688191 the current approach is problematic. The /t switch kills the whole tree, even if the /fi switch is supplied, resulting in the installer being killed.
This should also be related to #4891
Most helpful comment
tree kill is not used now ā now app is killed softly at first and it should be enough to terminate child process.