Lib installer hangs if project is on NFS mount

[augonis]

Configuration

Operating system: Ubuntu 18.04
PlatformIO Version: 4.1.0b3

Description of problem

If the project is within a mounted NFS share library install hangs and must be killed

Steps to Reproduce

1. Make new project within a mounted NFS share
2. Try to install a library e.g. pio lib install "ESPAsyncTCP"
3. ctrl+C when you are tired of waiting
4. Wait some more for the process to terminate

Actual Results

Library Storage: [somewhere in NFS mmount]
Looking for ESPAsyncTCP library in registry
Found: https://platformio.org/lib/show/305/ESPAsyncTCP [stuck here til I give up]
^C [waiting here for another good minute]
Aborted!

Expected Results

Library successfully installed

If problems with PlatformIO Build System:

The content of platformio.ini:

[env:esp12e]
platform = espressif8266
board = esp12e
framework = arduino

[ivankravets]

Could you reproduce this issue with the latest PIO Core 4.2?

[augonis]

Same results with PIO Core 4.2.0
I've tried to pause execution with pydev debugger and no reaction => python interpreter itself blocking on some system call?

[ivankravets]

Could you help to debug this issue? I don't have a similar environment. The PIO Core source code is located in

python -c "import platformio; print(platformio)"

[augonis]

I'm open to help, with some guidance.

[augonis]

I hooked up debugger an this time got some results.
In lockfile.py line 64 fcntl.flock(self._fp.fileno(), fcntl.LOCK_EX | fcntl.LOCK_NB) always raises OSError No locks available·

Now seeing that error I remember another python based project (Django) spitting that error under similar circumstances. Could be python bug.

[ivankravets]

Could you try to replace in this file all

fcntl.flock(

by

fcntl.lockf(

?

Does it work now?

See https://stackoverflow.com/questions/22409780/flock-vs-lockf-on-linux

[augonis]

Same result.
Reading https://serverfault.com/questions/66919/file-locks-on-an-nfs and https://unix.stackexchange.com/questions/396886/local-file-locking-on-nfs-being-linux-kernel-dependent I gather locks should work on NFS, but depends on server and client setup.
As workaround I remounted the share with nolock option with positive results.

[ivankravets]

As workaround I remounted the share with nolock option with positive results.

I think this is the right solution? If we drop the blocking process of package installer, it will lead to a different problem when 1> concurrent instances try to install the same package or manager registry.

Can we close this issue?

[augonis]

Agreed. Close it.

[ivankravets]

Happy coding with PlatformIO!