[v1.2] regression in v1.2.12 leaves container/shim hanging #4023
Description
reported in moby/moby#40514
bisected to #3366
f71f6d39b6131a72bcc93a667f72c1ed722ef3e4 is the first bad commit
commit f71f6d39b6131a72bcc93a667f72c1ed722ef3e4
Author: Michael Crosby <crosbymichael@gmail.com>
Date: Fri Jun 21 15:28:16 2019 -0400
Robust pid locking for shim processes
Closes #2832
Signed-off-by: Michael Crosby <crosbymichael@gmail.com>
(cherry picked from commit 719a2c594e4aad6a2de5cd9c298ab95309c2135c)
Signed-off-by: Sebastiaan van Stijn <github@gone.nl>
:040000 040000 fc5e0f9feb773e73730dc857137f93ca9241674f 2c6ee1b0341db99193cf61da77cada9994ec0f27 M runtime
Only seems to happen on some machines. The race could be performance-related and requires a slower machine. One known configuration where it reproduces is DigitalOcean $10 1cpu/2GB ram instance.
The container is left hanging with containerd-shim still running. Unknown if it helps but https://gist.github.com/tonistiigi/b1d7a3ad3811a7ce1c1ac96c4be04009 is a trace of shim when it hangs.