The previous algorithm for comparing files used python's
dircmp and is considered to be a shallow comparision. This
allowed distinctly small possibilities that two files being
compared could be different but not caught.
We now use go-mtree to do the comparison. This can emulate the
shallow comparison we had before but we can also adding a
sha256digest as part of the comparison using the new --keywords
option.
Also, made slight tweaks to gomtree functions in Atomic.util
so we debug and influence the return of JSON data.
This solves https://github.com/projectatomic/atomic/issues/761
Closes: #777
Approved by: rhatdan
Covers all but verify and generate. This is a refactoring of the
images subverbs (i.e. info, version, delete, ...)
Added in a unittest for list and info.
Closes: #771
Approved by: baude
Change --storage flag to not conflict with -f, and update some tests
to not error on some duplicate storage.
Signed-off-by: Yu Qi Zhang <jerzhang@redhat.com>
Closes: #768
Approved by: giuseppe
Add --rollback flag to atomic update, which switches a system
container to the other deployment if one exists.
Signed-off-by: Yu Qi Zhang <jerzhang@redhat.com>
Closes: #762
Approved by: rhatdan
If the has_image|has_container evals to True, then return
the proper object; else None
Add unittest structure for backends/backendutils.py
We want to make sure we have a good backend/backendutils
unittest. This should simplify unittests for all backends.
refactor.py: Remove the file used to help mini-test the
refactoring work. It is not needed by upstream and was
only including during the development phases.
Closes: #757
Approved by: baude
The unmount function for system containers was failing
for images that had regular files in its / directory
because shutil.rmtree was being used. The rmtree method
requires its input to a path of type dir. Added a simple
condition to check the input path and use os.remove(path)
on files.
Closes: #759
Approved by: rhatdan
With the -m switch, we can now compare the metadata between
the two diff objects. We report only the differences between them. All
common data is deleted from the object structures.
This solves --> https://github.com/projectatomic/atomic/issues/760
Closes: #764
Approved by: baude
Add an optional --storage flag to the following commands:
- images delete
- info
- mount
- verify
- version
If specified, the command will only look at the specified storage
(ostree/docker) for the image to perform the action. If not
specified, the command will look through both ostree/docker for
the image (as it was before). However, if the storage is not
specified and the image exists in both ostree and docker, the
command will error and prompt the user to specify. Image inspection
also no longer forces the user to delete/rename one or the other.
This is meant to address the duplicate naming issue (where a user
can have an image in both ostree and docker with the same name).
Signed-off-by: Yu Qi Zhang <jerzhang@redhat.com>
Closes: #720
Approved by: giuseppe
Several long-term issues existed for verify including some
munging of information between layers under certain
conditions. This PR addresses these issues by simplifying
the way things are to be inspected.
Example output as follows:
```
$ atomic verify docker.io/library/busybox
$ atomic verify docker.io/ubuntu
docker.io/ubuntu:latest contains the following images:
NAME LOCAL VERSION REMOTE VERSION DIFFERS
docker.io/ubuntu:latest 44776f55294a30e89d6a f753707788c5c100f194 Yes
registry.access.redhat.com/rhel7/rsyslog:latest contains the following images:
NAME LOCAL VERSION REMOTE VERSION DIFFERS
rhel7/rsyslog rhel7/rsyslog-7.2-21 rhel7/rsyslog-7.2-29 Yes
rhel7/rhel rhel7/rhel-7.2-56 rhel7/rhel-7.2-104 Yes
$
```
Closes: #714
Approved by: baude
The atomic image version subcommand was broken. Among the issues
was it called the wrong method, attempted to obtain the image
if not locally present, and outputted incorrect information. Some
of the problems were described in:
https://github.com/projectatomic/atomic/issues/708
The version method has been redone to obtain the correct information. It
no longer attempts to obtain the image if not present. It raises an
error.
Also, moved the version method into info and performed a pep8 clean up.
Example output now appears like:
```
atomic --debug version -r docker.io/busybox
IMAGE NAME VERSION IMAGE ID
10.3.11.254:5000/baude/busybox:latest None e02e811dd08f
Tag: docker.io/busybox:latest
Tag: localhost:5000/busybox:latest
atomic --debug version -r registry.access.redhat.com/rhel7/rsyslog
IMAGE NAME VERSION IMAGE ID
registry.access.redhat.com/rhel7/rsyslog:latest rhel7/rsyslog-7.2-21 53f20e902da7
registry.access.redhat.com/rhel7.2:7.2-56 rhel7/rhel-7.2-56 1a9b3357bac5
```
Closes: #710
Approved by: baude
Atomic Host variants do not have groff (and most likely never will since
it would pull in a lot of Perl packages). Print out a nicer error if
we're on AH, and make sure to skip that test.
Closes: #684
Approved by: cgwalters
Adding the ability to decompose an input image "name" that
includes a digest. For example:
docker.io/library/fedora@sha256:64a02df6aac27d1200c2572fe4b9949f1970d05f74d367ce4af994ba5dc3669e
Also, reword the decompose method in a Decompose class. This simplifies the use of
decomposition and allows for growth.
Example usage:
registry, repo, image, tag, digest = Decompose("docker.io/library/busybox:latest").all
repo = Decompose("docker.io/library/busybox:latest").repo
digest = Decompose("docker.io/fedora@sha256:64a02df6aac27d1200c2...67ce4af994ba5dc3669e").digest
Closes: #701
Approved by: rhatdan
We now allow pulls from registries that are not in
the docker configuration file. This altered our
decompose method a bit. We now check the registry
in decompose to see if it resolves on the network.
If so, then we use it.
Closes: #693
Approved by: baude
Add two classes and several new methods for inspecting
registries and images. These methods are helpful
for:
* taking user input (i.e. image names, partial image names) and determining
the fully qualified names.
* determining the fq name when the registry is omitted by the user.
* obtaining image manifests
* mimicing skopeo inspect
Closes: #687
Approved by: baude
Raise an error if there are more images matching the same prefix.
Signed-off-by: Giuseppe Scrivano <gscrivan@redhat.com>
Closes: #683
Approved by: rhatdan
Colleagues asked for decompose to be improved to where it took
an image name and broke it into registry, repo, image, and tag.
It also should mimic docker's implementation where 'library' is a known
exception
Also added unittests for decompose. Removed a singular glob test
as it is no longer valid
Closes: #677
Approved by: rhatdan
Save labels from docker manifest into ostree, so when a user invokes
'atomic version' on a system image, the metadata info is displayed,
same as docker images.
Signed-off-by: Yu Qi Zhang <jerzhang@redhat.com>
Closes: #679
Approved by: giuseppe
Add missing interfaces.
Fix functions to return text, rather then displaying directly.
Change atomic_client.py to take input from command line. These args are translated into function calls in python.
Add tests scripts in tests/integration/dbus directory
Closes: #668
Approved by: rhatdan
Wrap atomic run/stop with systemctl start/stop for system
containers. This way the user can directly use the atomic
CLI for the full container cycle. In addition, the equivalent
functionality already exists for docker containers, so it
makes sense to allow system containers to start/stop in a
similar fashion.
Signed-off-by: Yu Qi Zhang <jerzhang@redhat.com>
Closes: #651
Approved by: rhatdan
store the full image name, registry included, when pulling an image.
Some characters in the url are not usable for the name of an OSTree
ref, so escape them with '_'. This change breaks old images already
pulled.
use "rm -rf /ostree/repo/refs/heads/ociimage" and pull the images
again.
Signed-off-by: Giuseppe Scrivano <gscrivan@redhat.com>
Closes: #620
Approved by: cgwalters
Change the status in "atomic containers list" to display systemd
state instead of runc state. This way system containers match user
containers. Another reason for the switch is that systemctl
start/stop creates/destroys a runc container every time instead of
using the same container, so systemd gives better indication of
container status.
Also expand the states to be:
- running: container is running correctly
- inactive: container is stopped/newly created and ready to run
- failed: either systemd service or container process failed
instead of just "exited" and "running" as before.
Add tests and fix to test-image originally by Giuseppe.
Signed-off-by: Yu Qi Zhang <jerzhang@redhat.com>
Closes: #619
Approved by: giuseppe
It is useful in case of reverting to an old version. We may want to
do this automatically, without any manual intervention.
Signed-off-by: Giuseppe Scrivano <gscrivan@redhat.com>
Closes: #596
Approved by: rhatdan
Interactive users or deployment systems may want to perform
configuration before starting. For example, I may want to configure
the flannel data in etcd before starting flannel.
This topic comes up a lot in the RPM/deb worlds too; most RPM
distributions are inconsistent about this, whereas Debian basically
never starts a service on package install.
I think also it's better to teach admins to use systemd for service
management directly.
(This commit also changes things so we only `enable` on initial install,
which is a minor tweak)
Closes: #584
Approved by: rhatdan
