Calling a contextlib.contextmanager-annotated function which runs losetup / losetup -d (on teardown), twice in a row, will cause a Resource temporarily unavailable error on GCB kernel (as of this writing). It's non-deterministic, succeeding sometimes.
The contextmanager:
@contextmanager
def losetup_ctxmgr():
args = ["-f", "--show", "--offset", "0", "--sizelimit", "50000000000", IMAGE]
device = losetup(*args).stdout.decode("utf8").strip()
yield device
losetup("-d", device)
The relevant build output:
======================================== beginning repro.py in Dockerfile.inner container run by docker-dind in GCB
Traceback (most recent call last):
File "/workspace/repro.py", line 31, in <module>
main()
File "/workspace/repro.py", line 23, in main
with losetup_ctxmgr() as device:
File "/usr/lib/python3.7/contextlib.py", line 112, in __enter__
return next(self.gen)
File "/workspace/repro.py", line 18, in losetup_ctxmgr
device = losetup(*args).stdout.decode("utf8").strip()
File "/usr/lib/python3.7/site-packages/sh.py", line 1427, in __call__
return RunningCommand(cmd, call_args, stdin, stdout, stderr)
File "/usr/lib/python3.7/site-packages/sh.py", line 774, in __init__
self.wait()
File "/usr/lib/python3.7/site-packages/sh.py", line 792, in wait
self.handle_command_exit_code(exit_code)
File "/usr/lib/python3.7/site-packages/sh.py", line 815, in handle_command_exit_code
raise exc
sh.ErrorReturnCode_1:
RAN: /sbin/losetup -f --show --offset 0 --sizelimit 50000000000 /workspace/disk.img
STDOUT:
STDERR:
losetup: /workspace/disk.img: failed to set up loop device: Resource temporarily unavailable
ERROR
ERROR: build step 0 "gcr.io/cloud-builders/docker" failed: exit status 1