The benefit of this script is very limited, and after removing most of
the __all__ definitions it was even more limited. It was also buggy and
was confused by the use of __all__ in the signals.testing file, which is
why that could only be removed when the script was
Remove all of the uses of __all__ from almost all of the files
All of the unused definitions of __all__ are removed by this change. The definition of "unused" here is that all the tests still pass without them. They were also left in the public files for the apiclient.
LP #1900623: Secure erase may false-fail due to Linux page cache
Use direct I/O for read/write when checking secure erase success
Currently maas_wipe performs a check on secure erase effectiveness
by writing a buffer filled with 'M' to the device before the secure
erase and checking after the secure erase if, by reading the same
position/size from the device, the buffer of 'M' could be retrieved
(showing that the secure erase operation didn't work).
These read/write operations are achieved using regular Python
primitives, but Linux counts with page cache, so this test is
compromised (may show a false-negative, due to caching).
This patch fixes that by performing the read/write operations in such
routine using direct I/O. It was tested in user's environment and indeed
it worked by eliminating the false "failure" of secure erase operation.
Signed-off-by: Guilherme G. Piccoli <email address hidden>
Retry IPMI power commands without K_g key on K_g key failure.
Some BMCs will claim they are IPMI 2.0 complainant, allow you to set an
IPMI K_g key, allow you to verify an IPMI K_g key is set, yet will deny
any IPMI requests that use an IPMI K_g key. The error given is the K_g
key is invalid. This can be very confusing to users as MAAS allows you
to set an K_g key leaving them to think the only fix is to reset the
machine in person. In this case simply not using an K_g key allows all
IPMI commands to work again.
When there is an IPMI failure which is due to an K_g key and an K_g key
is set MAAS will retry without the K_g key. This is logged as a event
so the user knows what is happening but can still use their machine.
