Bug #10182
open
Provide more reasonable error messages for memory issues during container dispatch
Description
A customer received these two errors (logged in RT #136):
Error response from daemon: Cannot start container f1082c7aeec7be8b8ad1e45b5cf1273457f262e71a7909e7485a3c280c8c4dd4: [8] System error: open /sys/fs/cgroup/memory/system.slice/docker-f1082c7aeec7be8b8ad1e45b5cf1273457f262e71a7909e7485a3c280c8c4dd4.scope/memory.memsw.limit_in_bytes: no such file or directory
Error response from daemon: Cannot start container 6a0f0f39045bad7032893f3f926f4b1a2caee3d7caa4f061e46d41f52a763965: [8] System error: write /sys/fs/cgroup/memory/system.slice/docker-6a0f0f39045bad7032893f3f926f4b1a2caee3d7caa4f061e46d41f52a763965.scope/memory.memsw.limit_in_bytes: invalid argument
Which Ward interpreted as "is missing runtime_contraints, the job ran out of memory" and "has a runtime_constraint for ram but it seems to be set too low, the job ran out of memory."
We should make the logged error messages be more like Ward's version than the current cryptic versions.
Updated by Joshua Randall about 8 years ago
I ran into pretty much exactly these two error messages (from the same job) after upgrading systemd to the latest version (v230 from jessie-backports in my case), which appears to have issues with docker. The underlying problem seems to be that the system.slice directory is no longer present in that version.
The workaround was to switch docker to not use systemd for managing cgroups: https://github.com/docker/docker/issues/17653#issuecomment-155609224
If the fix for this issue obscured the error messages that come from docker, I'd never have figured out the real problem, so whatever the fix is here should probably make sure the errors are (also) logged.
Updated by Ward Vandewege about 8 years ago
Joshua Randall wrote:
I ran into pretty much exactly these two error messages (from the same job) after upgrading systemd to the latest version (v230 from jessie-backports in my case), which appears to have issues with docker. The underlying problem seems to be that the system.slice directory is no longer present in that version.
The workaround was to switch docker to not use systemd for managing cgroups: https://github.com/docker/docker/issues/17653#issuecomment-155609224
If the fix for this issue obscured the error messages that come from docker, I'd never have figured out the real problem, so whatever the fix is here should probably make sure the errors are (also) logged.
I agree - note that this ticket was a bit out of date - I also figured out a couple weeks ago that the other failure that can lead to this error is the cgroup thing you identified. Interpreting errors could be useful to give users a hint of what may be going on, but we shouldn't obscure the underlying errors.
Updated by Tom Morris over 7 years ago
- Target version set to Arvados Future Sprints
Updated by Ward Vandewege over 3 years ago
- Target version deleted (
Arvados Future Sprints)
Updated by