Skip to content

Latest commit

 

History

History
347 lines (256 loc) · 16.7 KB

README.md

File metadata and controls

347 lines (256 loc) · 16.7 KB

Midonet Sandbox

Getting started

Clone the git repository and install the package with:

$ python setup.py install

You may want to use a virtual env for it:

$ mkdir venv
$ virtualenv venv/
$ cd venv ; source bin/activate

Once you installed it, you can check the available flavours:

$ sandbox-manage flavours-list
[07-02 10:51:54] INFO - Cannot read ~/.midonet-sandboxrc
[07-02 10:51:54] INFO - Using default settings
+-----------------+
| Flavours        |
|-----------------|
| master+kilo     |
| master+liberty  |
| 2015.03+kilo    |
+-----------------+

You may want to check which components these flavours provide:

$ sandbox-manage flavours-list --details
[07-02 10:54:01] INFO - Cannot read ~/.midonet-sandboxrc
[07-02 10:54:01] INFO - Using default settings
+--------------+-------------------------------------------------------------------------------------------------------------------------------------------------+
| Flavour      | Components                                                                                                                                      |
|--------------+-------------------------------------------------------------------------------------------------------------------------------------------------|
| master+kilo  | 1x sandbox/keystone:kilo, 2x sandbox/cassandra:master, 1x sandbox/midolman:master, 1x sandbox/midonet-api:master, 3x sandbox/zookeeper:master   |
| 2015.03+kilo | 2x sandbox/cassandra:master, 1x sandbox/midolman:2015.03, 1x sandbox/keystone:kilo, 1x sandbox/midonet-api:2015.03, 3x sandbox/zookeeper:master |
+--------------+-------------------------------------------------------------------------------------------------------------------------------------------------+

You can check if the required images are available in your docker installation:

$ sandbox-manage images-list
[07-02 10:58:04] INFO - Cannot read ~/.midonet-sandboxrc
[07-02 10:58:04] INFO - Using default settings
[07-02 10:58:06] ERROR - A <class 'requests.exceptions.ConnectionError'> occured: ('Connection aborted.', error(2, 'No such file or directory'))

As default, the sandbox will try to connect to a local Docker daemon via the unix socket. In this case Docker was running in a VM, so let's specify this in the config file:

$ cat config.cfg
[sandbox]
docker_socket = tcp://192.168.33.10:4243

 sandbox-manage -c config.cfg images-list
[07-02 11:03:11] INFO - Loading configuration file: config.cfg
[07-02 11:03:11] INFO - No images found

Ok, so let's build the required images for the flavour:

$ sandbox-manage -c config.cfg build-all master+kilo
[07-02 11:04:28] INFO - Loading configuration file: config.cfg
[07-02 11:04:28] INFO - Building the following components: cassandra:master, midolman:master, midonet-api:master, zookeeper:master
[07-02 11:04:28] INFO - Build started for cassandra:master, publish is False
[07-02 11:04:28] INFO - Base image cassandra:base not found, skipping
[07-02 11:04:28] INFO - Now building sandbox/cassandra:master
Step 0 : FROM ubuntu:14.04
[... cut ...]

If you modify some of the images (dockerfiles or associated files), be sure to do python setup.py install, then you can build them independently. :

$ python setup.py install
$ sandbox-manager -c config.cfg build midonet:master
[07-10 16:18:16] INFO - Loading configuration file: config.cfg
[07-10 16:18:16] INFO - Build started for midolman:master, publish is False
[07-10 16:18:16] INFO - Now building sandbox/midolman:base
Step 0 : FROM ubuntu-upstart:14.04
[... cut ...]

NOTE: If creating the images is very slow, try setting the nofile limit to 1024 in /etc/init/docker.conf and restart docker (bug report).

Now all the required images are available:

$ sandbox-manage -c config.cfg images-list
[07-02 11:10:56] INFO - Loading configuration file: config.cfg
+----------------------------+---------------+
| Image                      | Created       |
|----------------------------+---------------|
| sandbox/zookeeper:master   | a minute ago  |
| sandbox/midonet-api:master | 2 minutes ago |
| sandbox/midonet-api:base   | 4 minutes ago |
| sandbox/midolman:master    | 4 minutes ago |
| sandbox/midolman:base      | 7 minutes ago |
| sandbox/cassandra:master   | 7 minutes ago |
+----------------------------+---------------+

and we can start a sandbox based on this flavour, let's call it staging.

NOTE: Before running the midolman component make sure the openvswitch module has been loaded on the host

(where the docker daemon runs)
$ sudo modprobe openvswitch

$ sandbox-manage -c config.cfg run master+kilo --name staging
[07-02 11:20:36] INFO - Loading configuration file: config.cfg
[07-02 11:20:36] INFO - Spawning master+kilo sandbox
Creating mnsandboxstaging_cassandra1_1...
Creating mnsandboxstaging_cassandra2_1...
Creating mnsandboxstaging_zookeeper1_1...
Creating mnsandboxstaging_zookeeper2_1...
Creating mnsandboxstaging_zookeeper3_1...
Creating mnsandboxstaging_api_1...
Creating mnsandboxstaging_midolman_1...
+-----------+-------------------------------+----------------------------+----------------------------------------------+-------------+
| Sandbox   | Name                          | Image                      | Ports                                        | Ip          |
|-----------+-------------------------------+----------------------------+----------------------------------------------+-------------|
| staging   | mnsandboxstaging_midolman_1   | sandbox/midolman:master    | tcp/22                                       | 172.17.0.28 |
| staging   | mnsandboxstaging_api_1        | sandbox/midonet-api:master | tcp/22,tcp/8080                              | 172.17.0.27 |
| staging   | mnsandboxstaging_zookeeper3_1 | sandbox/zookeeper:master   | tcp/2181,tcp/2888,tcp/3888                   | 172.17.0.26 |
| staging   | mnsandboxstaging_zookeeper2_1 | sandbox/zookeeper:master   | tcp/2181,tcp/2888,tcp/3888                   | 172.17.0.25 |
| staging   | mnsandboxstaging_zookeeper1_1 | sandbox/zookeeper:master   | tcp/2181,tcp/2888,tcp/3888                   | 172.17.0.24 |
| staging   | mnsandboxstaging_cassandra2_1 | sandbox/cassandra:master   | tcp/7000,tcp/7001,tcp/7199,tcp/9042,tcp/9160 | 172.17.0.23 |
| staging   | mnsandboxstaging_cassandra1_1 | sandbox/cassandra:master   | tcp/7000,tcp/7001,tcp/7199,tcp/9042,tcp/9160 | 172.17.0.22 |
+-----------+-------------------------------+----------------------------+----------------------------------------------+-------------+

To stop the sandbox:

$ sandbox-manage -c config.cfg stop staging      # or stop-all to stop all the running sandbox
[07-02 11:23:19] INFO - Loading configuration file: config.cfg
[07-02 11:23:19] INFO - Sandbox staging - Stopping container api_1
[07-02 11:23:20] INFO - Sandbox staging - Stopping container zookeeper3_1
[07-02 11:23:30] INFO - Sandbox staging - Stopping container zookeeper2_1
[07-02 11:23:40] INFO - Sandbox staging - Stopping container zookeeper1_1
[07-02 11:23:50] INFO - Sandbox staging - Stopping container cassandra2_1
[07-02 11:23:52] INFO - Sandbox staging - Stopping container cassandra1_1

The stop command will try to gracefully stop the container. However, this might be a slow operation as the host waits for the container command to return. In a development session, you might not care about the state of the container after stopping it so you have the option to just kill it without waiting for the container to return.

$ sandbox-manage -c config.cfg kill staging     # or kill-all to kill all the running sandbox

Provide your own flavour

To provide your own flavour, just create a folder, place your flavour yml file in it (you may want to copy and modifify a base one you find Here, and specify it in the configuration file:

$ ls /workplace/midonet-sandbox/venv/extra
zk-only.yml

$ cat config.cfg
[sandbox]
extra_flavours=/workplace/midonet-sandbox/venv/extra
docker_socket = tcp://192.168.33.10:4243

$ sandbox-manage -c config.cfg flavours-list --details
[07-02 11:41:30] INFO - Loading configuration file: config.cfg
+--------------+-------------------------------------------------------------------------------------------------------------------------------------------------+
| Flavour      | Components                                                                                                                                      |
|--------------+-------------------------------------------------------------------------------------------------------------------------------------------------|
| zk-only      | 3x sandbox/zookeeper:master                                                                                                                     |
| master+kilo  | 1x sandbox/keystone:kilo, 2x sandbox/cassandra:master, 1x sandbox/midolman:master, 1x sandbox/midonet-api:master, 3x sandbox/zookeeper:master   |
| 2015.03+kilo | 2x sandbox/cassandra:master, 1x sandbox/midolman:2015.03, 1x sandbox/keystone:kilo, 1x sandbox/midonet-api:2015.03, 3x sandbox/zookeeper:master |
+--------------+-------------------------------------------------------------------------------------------------------------------------------------------------+

The flavour file will inherit and reuse base components. You refer to base yml files using the $BASE variable:

$  cat /workplace/midonet-sandbox/venv/extra/zk-only.yml
zookeeper1:
  extends:
    file: $BASE/zookeeper.yml
    service: zookeeper
  image: sandbox/zookeeper:master
  volumes:
  - /zoo/conf/
  environment:
  - ZOO_ID=1
  ports:
  - 9000:2181

[ .. cut ...]

Provide your own component

Same way, you can provide your own components creating a components directory and passing it in the configuration file using the extra_components parameter.

The component directory must follow a structure like ($component/$version/$(component-version).dockerfile). For example, if want to provide your zookeeper component tagging it as master, the associated components directory will be something like:

extra-components
└── myzookeeper
    └── master
        ├── bin
        │       └── run-zookeeper.sh
        └── myzookeeper-master.dockerfile

This specify the myzookeeper:master component (docker image) that you can use in your flavours.

The configuration file will look similar to:

$ cat config.cfg
[sandbox]
extra_flavours=/workplace/midonet-sandbox/venv/extra
extra_components=/workplace/midonet-sandbox/venv/extra-components
docker_socket = tcp://192.168.33.10:4243

Provide an override

An override is a directory that contains a directory for each component to override (the name must match the 'service' parameter defined in the flavour yml file)

$ tree /workplace/midonet-sandbox/venv/myoverride
/workplace/midonet-sandbox/venv/myoverride
├── midolman
│   ├── midolman-master.deb
│   └── override.sh
└── zookeeper
    └── override.sh

Each component folder contains the override.sh script and may contain files.

When an override is applied to a sandbox each container mounts the specific override folder in the /override path and change its CMD to /override/override.sh.

Typical uses of an override is to pass a new debian package and install it, or change a configuration file before starting the service:

$ cat /workplace/midonet-sandbox/venv/myoverride/midolman/override.sh
#!/bin/sh
dpkg -i /override/midolman-master.deb
exec ./run-midolman.sh

NOTE: if you call another script within your override to start an upstart service (e.g. run-midolman.sh calls /sbin/init to spawn an upstart service), do it through the exec command (not a regular invocation).

Additionally, you can also create a packages directory inside the override that will act as a local debian repository. This way, you could copy the packages inside this folder, sandbox will detect it and mount it inside the container with proper repository metadata files. The main benefit of this approach is automatic package dependency resolution, not available with the dpkg command. To use this local repository, a component override script should execute the following:

$ cat /workplace/midonet-sandbox/venv/myoverride/midolman/override.sh
#!/bin/sh
LOCAL_REPO_FILE/etc/apt/sources.list.d/local-repo.list
echo "deb file:/packages /" > $LOCAL_REPO_FILE
apt-get update -o Dir::Etc::sourcelist=$LOCAL_REPO_FILE

# After that, the override can install the packages normally with apt-get,
# appending /local to the package name. This will take precedence over all
# other repos.
apt-get install -qy midolman/local

**NOTE: only Debian repositories are supported for the moment.

To apply an override to a sandbox you can use the --override parameter:

$ sandbox-manage -c config.cfg run master+kilo --name staging --override /workplace/midonet-sandbox/venv/myoverride
[07-02 11:52:59] INFO - Loading configuration file: config.cfg
[07-02 11:52:59] INFO - Spawning master+kilo sandbox with override /workplace/midonet-sandbox/venv/myoverride
Recreating mnsandboxstaging_cassandra1_1...
Recreating mnsandboxstaging_cassandra2_1...
[ ... cut ... ]

Provide a provisioning script

You may want to configure a virtual topology once the sandbox is up&running. An handy way to achieve this would be passing a provisioning script at the command line. This script will be run just after all the containers are up and will take care of inizializing the virtual topology. A command line will simply looks similar to:

$ sandbox-manage -c config.cfg run master+kilo --name staging --provision venv/myoverride/provision.sh

Eg: The following provisioning script will create a tunnel zone and register two hosts to it:

#!/bin/sh
set -uxe

# Wait that all services are up&running
sleep 30

CLI_HOST="$(docker ps | grep midonet-api | awk '{print $1}')"
CLI_COMMAND="docker exec $CLI_HOST midonet-cli -A -e"

HOST0_ID=$($CLI_COMMAND host list | head -n 1 | awk '{print $2}')
HOST1_ID=$($CLI_COMMAND host list | tail -n 1 | awk '{print $2}')
HOST0_NAME=$($CLI_COMMAND host $HOST0_ID show name)
HOST1_NAME=$($CLI_COMMAND host $HOST1_ID show name)

MIDOLMAN1_IP="$(docker inspect --format '{{ .NetworkSettings.IPAddress }}' $HOST0_NAME)"
MIDOLMAN2_IP="$(docker inspect --format '{{ .NetworkSettings.IPAddress }}' $HOST1_NAME)"

# Create the tunnelzone and add midolman
$CLI_COMMAND tunnel-zone create name default type vxlan

TZONEID=$($CLI_COMMAND tunnel-zone list | awk '{print $2}')

$CLI_COMMAND tunnel-zone $TZONEID add member host $HOST0_ID address $MIDOLMAN1_IP
$CLI_COMMAND tunnel-zone $TZONEID add member host $HOST1_ID address $MIDOLMAN2_IP

Specifying override folder and provisioning scripts in the flavour yaml

You can also specify the override folder and the provisioning scripts in the flavour yaml filed you use to start up a sandbox. For that, just add the relative paths starting where the sandbox-manage command is executed, somewhere in your flavour, at the same level as other service definitions (here's an example, it may differ in your local environment):

[...]
neutron:
  extends:
    file: $BASE/neutron.yml
    service: neutron
  image: sandbox/neutron:kilo
  hostname: neutron
  links:
  - "keystone:keystone"
  - "cluster:midonetapi"

provision: sandbox/provision/all-provisioning.sh

override: sandbox/override_v2

These options can be overridden through the command line as explained in the previous sections. Now, to start a sandbox, you just need to specify the flavour definiton without needing to also specify the provision and override options.

Pushing and pulling from an external registry

By default, sandbox-manage pulls and pushes to the default registry in docker index.docker.io. If you need to interact with a different docker registry, a couple of config parameters should be added to the sandbox.conf file. An example of how the configuration file looks like in that case is displayed below.

$ cat sandbox.conf
[sandbox]
extra_flavours=sandbox/flavors
extra_components=sandbox/components
docker_registry=your.own.artifactory.company.com
docker_insecure_registry=True