runc create failed

[YRXING]

When I use podman to create container, it blocked and an error occurred in the runc service. The flowing is the error log.
Nov 18 12:27:08 dc06-p3-t628-n013 conmon[396051]: conmon 90a1aa9dc23f95a46e5e : container 397582 exited with status 137
Nov 18 12:45:19 dc06-p3-t628-n013 conmon[386855]: conmon d05ca534cae8fa505f1b : container 396038 exited with status 137
Nov 18 14:55:52 dc06-p3-t628-n013 conmon[2154175]: conmon cdfba70f9e2dbb5a18c2 : runtime stderr: runc create failed: unable to start container process: can't get final child's PID from pipe: EOF
Nov 18 14:55:52 dc06-p3-t628-n013 conmon[2154175]: conmon cdfba70f9e2dbb5a18c2 : Failed to create container: exit status 1
Nov 18 14:55:52 dc06-p3-t628-n013 conmon[2715470]: conmon e65dbd8ac78061834562 : runtime stderr: runc create failed: unable to start container process: error during container init: unable to store init state: %!w()
Nov 18 14:55:52 dc06-p3-t628-n013 conmon[2715470]: conmon e65dbd8ac78061834562 : Failed to create container: exit status 1
Nov 19 16:09:00 dc06-p3-t628-n013 conmon[23966]: conmon d8e7b9fba2987d386bde : runtime stderr: runc create failed: unable to start container process: can't get final child's PID from pipe: EOF
Nov 19 16:09:00 dc06-p3-t628-n013 conmon[23966]: conmon d8e7b9fba2987d386bde : Failed to create container: exit status 1
Nov 19 17:08:37 dc06-p3-t628-n013 conmon[1271628]: conmon 0c86f25a68b7d1d0086d : runtime stderr: runc create failed: unable to start container process: can't get final child's PID from pipe: EOF
Nov 19 17:08:37 dc06-p3-t628-n013 conmon[1271628]: conmon 0c86f25a68b7d1d0086d : Failed to create container: exit status 1
Nov 19 21:23:19 dc06-p3-t628-n013 conmon[951854]: conmon 37764cef06e9a5925286 : runtime stderr: runc create failed: unable to create new parent process: unable to setup exec fifo: open /run/runc/37764cef06e9a59252863e19915cd940b2c01e2a4e91eec9339a41178ccf9751/exec.fifo: no such file or directory
Nov 19 21:23:19 dc06-p3-t628-n013 conmon[951854]: conmon 37764cef06e9a5925286 : Failed to create container: exit status 1
Nov 20 03:33:18 dc06-p3-t628-n013 systemd[1]: carrier.service: Dependency After=carrier.service dropped
Nov 20 03:33:19 dc06-p3-t628-n013 systemd[1]: carrier.service: Dependency After=carrier.service dropped
Nov 20 15:33:20 dc06-p3-t628-n013 conmon[582965]: conmon 795834d7c475dc1da9b9 : runtime stderr: runc create failed: unable to create new parent process: unable to setup exec fifo: open /run/runc/795834d7c475dc1da9b9a8c6e79575728df6aade1d2538666e5205b05771fb7e/exec.fifo: no such file or directory

And libpod created a large number of cgroup directories at the same time.

Does anyone know why?

[lifubang]

Could you please provide more informations. For example:
Steps to reproduce the issue
What version of runc/podman are you using?
Host OS information
Host kernel information

[YRXING]

Could you please provide more informations. For example: Steps to reproduce the issue What version of runc/podman are you using? Host OS information Host kernel information
This is the same question I raised in the podman community. containers/podman#27565
This issule contains version information for podman and runc.
os & kernel version
Debian 5.4.250.bsk.3
Description: Debian GNU/Linux 10 (buster)
Release: 10
Codename: buster

[cyphar]

Is it possible for you to use a runc built with #4928 applied and then provide the logs you see? Higher-level runtimes like podman strip out our error logging which means you only really see errors like can't get final child's PID from pipe: EOF which only tell you that runc init died (which could happen for a hundred different reasons) and nothing else useful.

(Of course if you can reproduce this with runc directly, that would be ideal.)

[YRXING]

Is it possible for you to use a runc built with #4928 applied and then provide the logs you see? Higher-level runtimes like podman strip out our error logging which means you only really see errors like can't get final child's PID from pipe: EOF which only tell you that runc init died (which could happen for a hundred different reasons) and nothing else useful.

(Of course if you can reproduce this with runc directly, that would be ideal.)

Okay, I'll rebuild a runc version using this PR first.