By pruning machines that are in "unknown" status after each operation,
the Go code path exposed the fact that the Docker provider was not
updating the machine index during an "up" - leaving the state as
"unknown".
This is basically a bug within the Docker provider, so I think it's okay
to update the plugin code to correct this rather than working around the
issue in Go.
All we need to do is call `machine.state` instead of reaching through to
`machine.provider.state` while waiting for the container to be started.
That causes the extra logic for updating the machine index in
`machine.state` to fire.
Prior to this commit, if a created but exited container bound a port,
and a new container grabed that same port (say for an ssh port forward),
when the initial container came back up it would fail because the port
also got bound to the second container. This commit fixes that behavior
by first looking at what containers are already bound prior to creating
a container.
When user is using podman's docker CLI emulation the containers would
fail to enter running state because the docker driver could not catch
the container name. This commit fixes that by adding a check if podman
docker emulation is used and pick the container hash correctly from the
output.
When buildkit is enabled docker will write out
`writing image <image id>`
When buildkit is not enables docker will write out
`Successfully builld <image id>`
It is more likely that searching for the `writing image` will not
clash with build output than when searching for `Successfully built`.
eg. when installing python packages with pip, it is common to use
the verbage `Successfully built`.
Output from recent Docker has changed:
Stderr: Error response from daemon: conflict: unable to delete 0ba49dd235e5 (cannot be forced) - image is being used by running container 250bbe980448
Fixes#7245
Prior to this commit, docker would look for a container ID based on
"Successfully built" string. This output does not exist if a user has
enabled the experimental feature buildkit. This commit updates the build
behavior to match against both kinds of outputs, and instead of using
`scan`, it uses MatchData and groups the container id with match group
name `:id` instead of making hard assumptions with the matches being
contained inside arrays from scan.
At least on macOS combo Catalina + Docker engine 19.03.4 + Docker desktop 2.1.0.4 + vagrant 2.2.6 the original `matches = result.scan(/Successfully built (.+)$/i)` -line fails to generate a match. With this change I can `vagrant up --provider=docker` successfully.
Prior to this commit, the docker action was using the method `prefix` on
an IPv4 and IPv6 address. This works fine for ruby versions 2.5 and
newer, however the ruby shipped with Vagrant is before 2.5, and
therefore the IPv4 and IPv6 classes do not have the prefix method,
resulting in an error. This commit fixes that by using a different
method of determining the prefix.
Prior to this commit, the docker compose build method would not properly
set build_args if given in a Vagrantfile. This commit fixes that by
using the passed in key `extra_args` from the docker build action.
In commit 7980178d194bb21bde2e21858fbc969e4cc7eb3f (#10879) I added a
`usable?` class method to `VagrantPlugins::DockerProvider::Provider`.
However, commit 34e53a5a4b208edcc3e25c1f0e3aa0ef56e8d8d9 (#10890)
incorrectly changed it to an instance method. This rendered it
ineffective because it’s called on the class, not an instance. Change
it back to a class method.
Signed-off-by: Anders Kaseorg <andersk@mit.edu>
Prior to this commit, the docker login action assumed that if there was
a password to authenticate with, Vagrant was using a host vm to run
docker. This is likely due to some legacy decisions with how Vagrant
used to manage running docker. This commit fixes that by only grabbing a
host_vm lock if the host_vm is actually in use, otherwise login
normally.
Prior to this commit, the docker compose driver would _always_ path
expand a host volume no matter what. This is not always the correct
option, for example if that host volume is actually a reference to a key
inside a `volumes` hash instead of a path on disk. This commit changes
that by looking to see if the requested host volume is actually a
defined key inside the compose config, and if not, it will path expand
it like before. Otherwise it will leave the key "as is".
Use `mask` option for defining subnet on network configuration. Allow
options to be passed through using hash scoping and docker_network and
docker_connect prefixes. Enable public networks. Allow configuration
to define pre-existing networks by name.
Container inspection is caching data on first lookup. This will
result in incorrect data being returned on subsequent lookups if
a different `cid` value is provided. Also removed rescue of
the `Exception` class as this generally should never happen; rescue
of StandardError will be enough.
This commit updates the behavior of how the docker provider creates new
docker networks. It looks at each existing network to see if the
requested subnet has already been configured in the docker engine. If
so, Vagrant will use that network rather than creating a new one. This
includes networks not created by Vagrant. Vagrant will not clean up
these networks if created outside of Vagrant.
