a SIGCHLD race in signal handler setup (regression since runc 1.3.0) #5208
Description
Description
Since signal.Notify runs in a goroutine, a fast-exiting process
could trigger SIGCHLD before signal registration completes, causing
the signal to be silently discarded. This left forward() blocking
forever on a signal that would never arrive.
Original posted by: @bitoku
Steps to reproduce the issue
No response
Describe the results you received and expected
There is a chance causes runc exec or runc run(without -d) stuck:
runc exec <container_id> true
What version of runc are you using?
1.3.5
Host OS information
No response
Host kernel information
No response