revive would only fail if rbd had errors due to the loop being in shell.
Now we expand the modules list as arguments to revive which then
invokes revive only once and properly handles errors.
Signed-off-by: John Mulligan <jmulligan@redhat.com>
This should basically run similarly to the travis ci configuration
but use the native github ci infrastructure.
In order to avoid messing around with various env vars I had to use the
v2-beta version of the go setup library. Once v2 has been released
we can switch away from the beta.
Signed-off-by: John Mulligan <jmulligan@redhat.com>
For more consisistency use getError where possible.
Changes produced using `gofmt -r 'RadosError(int(x)) -> getError(x)`,
followed by a few manual fixups.
Signed-off-by: John Mulligan <jmulligan@redhat.com>
Mention the env vars that the go test code looks for but does not change
from the "outer" environment as a way of documenting those variables.
Signed-off-by: John Mulligan <jmulligan@redhat.com>
Previously, calling Release more than once for the same MountInfo would
abort due to a double free in the ceph libs. As this is somewhat user
hostile we add some simple state tracking in our wrapper type such that
one can make redundant calls to Release without crashing the
application.
Signed-off-by: John Mulligan <jmulligan@redhat.com>
The test function that exercises ChangeDir was creating a directory
and leaving it behind. Add some basic cleanup to the test to try
and avoid leaking stuff between test runs.
Signed-off-by: John Mulligan <jmulligan@redhat.com>
In a previous commit we added fsDisconnect to match with fsConnect. For
tests that do not use fsConnect but still acquire resources we need to
free up those resources with direct calls to the appropriate functions.
Signed-off-by: John Mulligan <jmulligan@redhat.com>
For every (most?) call to fsConnect we should pair it with a (deferred)
call to fsDisconnect to close connection and release resources
associated with the mount object we have created for the test.
Signed-off-by: John Mulligan <jmulligan@redhat.com>
The defer function in TestMountWithRoot was always trying to call
unmount before release. This is not correct if the fs was not mounted
so we reorganize this code to more closely match what should be done
in normal practice.
Signed-off-by: John Mulligan <jmulligan@redhat.com>
Somehow we managed to introduce extra code into TestCreateMount that
had it create two mounts. This is undesirable for this test so this
change eliminates the extra unwanted invocation.
Signed-off-by: John Mulligan <jmulligan@redhat.com>
Previously, the mds name was hardcoded in the tests to match that set up
by "micro-osd.sh". We continue to default to that value, but allow the
test to be supplied a custom value by the environment.
Signed-off-by: John Mulligan <jmulligan@redhat.com>
When running cephfs tests locally there is no strong urge to require
an environment exactly like that of the test container. However, we do
want tests run under the container image (especially our default ci) to
use the cephfs mountpoint as they were previously. Set a default value
in the entrypoint script to require the mountpoint so we don't
accidentally miss the tests if something unexpected goes wrong.
Signed-off-by: John Mulligan <jmulligan@redhat.com>
Add a utility function to test cases that expect the cephfs file system
to be mounted locally and accessible.
Add environment vars to configure the location of the mount point as
well whether the function causes the test to be skipped (the default)
or force an early but clear failure.
Signed-off-by: John Mulligan <jmulligan@redhat.com>
Within the last 24hrs the "octopus" release of ceph is available, and
matching tags have been created in the ceph containers project. Consume
the release name rather than master.
Signed-off-by: John Mulligan <jmulligan@redhat.com>
We had a lot of "exported" badness previously but that either been
resolved or disabled on a case-by-case basis. Make the rule fail the
check so that new cases like this are not brought in.
Signed-off-by: John Mulligan <jmulligan@redhat.com>
Previously, the types were named RbdImageOptions and RbdImageOption.
Now they are named ImageOptions and ImageOption.
Type aliases have been added to prevent existing code from
breaking. As these types were new in v0.2, I suspect we can drop
the aliases post-v0.3.
Signed-off-by: John Mulligan <jmulligan@redhat.com>
Image option constants are now named ImageOptionX, without an Rbd
prefix. Aliases to the old names are provided and doc comments for both.
Signed-off-by: John Mulligan <jmulligan@redhat.com>
This further improves the ability to use the script outside the
container itself and against alternate ceph cluster configs.
Signed-off-by: John Mulligan <jmulligan@redhat.com>
Currently there's no container image named "octopus" but its imminent.
Once ceph container starts tagging images with "latest-octopus" we
should switch to that.
Signed-off-by: John Mulligan <jmulligan@redhat.com>
It appears that ceph octopus (currently testing as master) is stricter
about some inputs and the old formatting in the test was failing.
Use formatting that all versions should be happy with.
Signed-off-by: John Mulligan <jmulligan@redhat.com>
Ceph containers [1] are built by the ceph team and for our purposes,
only lack build tools and Go making the delta we need to do smaller.
1: https://github.com/ceph/ceph-container
Signed-off-by: John Mulligan <jmulligan@redhat.com>
Instead of duplicating some of the actions in the travis yaml file and
the Makefile, have travis rules make use of the Makefile.
Signed-off-by: John Mulligan <jmulligan@redhat.com>
