Views
Release Infrastructure/IMG
Contents |
What is IMG
IMG (Image Me Give) is a small python client/server application suite, its sole job is to get a kickstart file from a user, via a web interface, command line or BOSS process. Then it runs Moblin-Image-Creator, either in a host or in a VM. IMG also has the capability to run as a BOSS participant and thus be a part of a build process.
Design
- Asynchronous
- Fire and forget about it, check back later with the cli client for results
- Queue view with django,
- Show a nice html view of currently running or standby MIC jobs in the queue
- Submit a ks via a web interface
- Upload via a form
- No security
- No https or file filter restrictions :)
- Two ways of communication
- BOSS and pure AMQP messages
(FIXME: Which rpms provide which services and where should they be installed? What is the expected network layout - in particular does a single img-web control a pool of img workers? What about the participant? lbt 27/Dec)
Example workflow
The application consists of two parts, django frontend and the actual image creation application. Both of them are connected via AMQP, RabbitMQ is used as the server.
So an example workflow, in AMQP:
- User posts an image creation request with email address and sample kickstart file (or a package overlay), using the web application or the command line client.
- Django application receives the request and imports the information on a model object and saves it (Even for the command line client? lbt 27/Dec)
- A message is sent via an AMQP broker to the image creation worker, containing the details for the required image. (Does this use BOSS? lbt 27/Dec)
- Image creation commences
- Virtual machine disk image containing mic2 is engaged (as a kvm overlay image), with ssh access.
- When a package overlay is received, a kickstart file is constructed from a ready-made template and passed to kickstart part
- When a kickstart file is received, it is passed directly to worker
- Worker starts up and connects to the virtual machine using ssh and passes the required arguments to mic2.
- Worker copies image on success from the virtual machine to well-known www-root, if there is an error, only a message about it is sent via AMQP.
- When mic2 has finished, the virtual machine is shut down by the worker.
- When image is created, the virtual machine disk image is deleted
Installation
Installing IMG via RPM for OpenSUSE 11.2.
The following repos are needed : MeeGo-Infra, opensuse python and MeeGo tools:
zypper ar http://download.opensuse.org/repositories/Maemo:/MeeGo-Infra/openSUSE_11.2-plus/Maemo:MeeGo-Infra.repo zypper ar http://download.opensuse.org/repositories/devel:/languages:/python/openSUSE_11.2/devel:languages:python.repo zypper ar http://repo.meego.com/MeeGo/tools/repos/opensuse/11.2/meego-tools.repo zypper ref
Then install the core IMG package and the amqp listener:
zypper in img-core img-amqp img-boss
Worker / img-core
The worker code is provided by img-core and is responsible for running mic2 either locally or on a virtual machine. The initialisation code takes care of creating the necessary command stubs for mic2 use for either implementations. It also handles the KVM creation command stubs.
In KVM mode, it first creates an overlay image from the base image (not distributed) and then starts it using the overlay image as the hard disk. Virtio is used as a speed optimisation method for KVM. When a specific amount of time has passed, it will copy the kickstart and mic2 config file to the guest VM. Then it runs mic2 in the VM with parameters specified in the init method. After mic2 has run, the image is copied from the guest using scp.
In local mode, only mic2 is run on the host machine. It will create the image based on the options in the init method along with the kickstart file.
A worker needs a queueing mechanism to accept jobs. The primary solution is expected to be the BOSS participant (img-boss) but there is a plain amqp client too (img-amqp).
BOSS participant / img-boss
The general design of the client is pretty much the same as the AMQP server, since they both get activated and then call the worker code to create the image, with or without KVM. The worker code is described in the section above.
More information about BOSS Infrastructure/BOSS.
See boss_build_image.py for the source code.
BOSS Client / img-amqp
See boss_img_client.py for example BOSS client.
Usage
(Is this the img-amqp usage? lbt 27/Dec)
Usage: boss_client.py -n|--name <name> -t|--type <imagetype> -e|--email <author@email> -r|--release <release> -a|--arch <arch >-s|--submit -k <kickstart_file.ks>
boss_client.py Sends a message (poll for result later) to the BOSS, using <kickstart.ks> as the kickstart file.
Options:
-h, --help show this help message and exit
-s, --submit Submit to BOSS, takes no options
-k KICKSTART, --kickstart=KICKSTART
Kickstart file
-t TYPE, --type=TYPE Image type
-n NAME, --name=NAME Image name
-e EMAIL, --email=EMAIL
Author email
-r RELEASE --release Image release, goes directly to mic2
-a ARCH --arch Architecture to use
Server
See build_image.py for example server.
This section documents the server component of IMG.
Functionality
Consumer functions
- kickstarter_callback:
- Receives messages as described [#Kickstarter here], decodes JSON data to variables and creates a kickstart file with the received list of packages and then submits them to the mic-image-creator [#MoblinImageCreator here].
- mic2_callback:
- Receives messages as described [#MoblinImageCreator here], decodes JSON data to variables and runs moblin-image-creator (NB! this has to exist in /usr/bin/moblin-image-creator, will be changeable in the future) against the supplied configuration file and image type.
- The images are saved to to a configured location, needs to be changed in the python subprocess call in order to get downloadable images (eg. to a webserver media path).
- If the image creation fails (is caught with "CalledProcessError" exception in the code, this happens with nonzero exit status), a message is sent to "error_queue" queue, with the following JSON encoded dictionary:
'error': Contains the error message of the exception received from the python subprocess call. 'id': identification string of the original message sent to "image_queue" queue 'url': Contains the URL from which one can download the log
- If the image creation succeeds, a message is queued to "link_queue" queue, with the following JSON encoded dictionary:
'url': Contains the URL from which one can download the image
'id': identification string of the original message sent to "image_queue" queue
Exchanges
- image_exchange, The exchange for image creationg related queues
- django_result_exchange, Exchange for results, picked up by django
Queues
Queues for exchange "image_exchange":
- image_queue
- Stores messages related to Moblin Image Creator runtime.
- kickstarter_queue
- Stores messages related to kickstarter runtime.
Queues for exchange "django_result_exchange":
- status_queue
- Contains messages in the following format (JSON encoded dictionary):
- 'status': status of the image build, compulsory
- 'id': the identification string of the original message sent to kickstarter_queue, compulsory
- 'url': Contains the URL from which one can download the image
- 'log': specifies the log file to read from
- 'error': any error messages faced
Client
For demonstration see img_client.py for example client.
Usage
img_client.py -p|--poll <id> -n|--name <name> -t|--type <imagetype> -e|--email <author@email> -r|--release -a|--arch s|--submit -k <kickstart_file.ks>"
where:
- -p|--poll <message_id>, poll the AMQP server for messages relating to the id, see -a|--async
- -t|--type <imagetype>, image type, can be of: livecd, liveusb, loop, raw, nand, mrstnand, vdi or vmdk
- -n|--name Image name
- -a|--arch Image architecture
- -r|--release Image release number
- -c|--config Configuration file to use
Documentation
This client version uses AMQP as the message bus to communicate with the server. The format for the messages is described in the next part of this document.
Messages
The general message format is JSON encoded dictionary.
Moblin Image Creator
To initialize the actual image building, send a JSON formatted message as follows:
{'email':email, 'id':id, 'imagetype':imagetype, 'ksfile':ksfile, 'name': name, 'arch':arch, 'release':release}
where:
- email, email address of the submitter
- id, identification string of the original message (passed on from kickstarter process)
- imagetype, image type of values: livecd, liveusb, loop, raw, nand, mrstnand, vdi or vmdk
- ksfile, the actual kickstar file to upload (in text)
- name, the name for the image
- arch, architecture to use for the image building
- release, release numbering, $VERTICAL-$ARCH-$VARIANT like in mic2
Django client / img-web
Views
- submit
- If this view receives a POST request, it constructs a bound form with the data from POST array. If the form is valid it extracts the data, as specified in the forms section.
- Applying the overlay is handled in this form, in the following way.
- Get the overlay text from the form field
- use python strings split() method to split the overlay text in to a list, with a comma delimiter
- construct a kickstart file using this list and append the packages to the kickstart files package list
- queue
- This view is responsible of two things. Firstly it checks wether there are any messages in the "status_queue" message queue, if there are, then it updates the !ImageJob model object (identified by the identification string from the message) to include the ready image URL.
- Secondly, it checks for possible error messages in "status_queue" queue, if there are any, it assigns a special variable to indicate that there has been an error.
- If no messages are received, it simply redirects to a template with all the !ImageJob model objects.
- job
- Job view is responsible of providing the information about the log file. The information about the logfile is formed in the server side, such as the URL to the log file. If there is a URL, it will open the URL and read its contents and return the resulting text to the template. The template renders the log message with a textarea html widget.
- index
- Index only returns a template in which one can click a link about uploading the image.
URLs
url(r'submit/$', 'meego_img.app.views.submit', name='img-app-submit'), url(r'queue/$', 'meego_img.app.views.queue', name='img-app-queue'), url(r'job/(?P<msgid>\S+)$', 'meego_img.app.views.job', name='img-app-job'), url(r'images/(?P<msgid>\S+)$', 'meego_img.app.views.download',name='img-app-download'),
Models
IMG currently has only one model, !ImageJob, here is the Django code for it:
# Create your models here.
class ImageJob(models.Model):
email = models.CharField(max_length=40)
filename = models.CharField(max_length=40)
logfile = models.CharField(max_length=50)
task_id = models.CharField(max_length=30)
imagefile = models.CharField(max_length=50)
created = models.DateTimeField(auto_now_add=True)
error = models.CharField(max_length=500)
type = models.CharField(max_length=10)
status = models.CharField(max_length=30)
def delete(self, *args, **kwargs):
if self.logfile:
if os.path.exists(self.logfile):
os.remove(self.logfile)
os.remove(self.logfile.replace("-log", ""))
print "Removed %s"%self.logfile
super(ImageJob, self).delete(*args, **kwargs)
As can be seen in the delete method, this model cleans up all image creation related files, like the kickstarter file and log file.
Forms
UploadFileForm
Contains the following fields
- email, email field
- overlay, text input field
- name, text input field
- platform, select field for the platform, parsed from the default template
- release, a text input field for release
- arch, architecture selection field
- imagetype, select field for the image type, with values: livecd, liveusb, loop, raw, nand, mrstnand, vdi, vmdk
- ksfile, the raw kickstart file
Worker configuration
The configuration file is to be installed in /etc/imger/img.conf and a sample file is here.
Configuration file
base_url: A base URL that has a direct access to the base_dir via HTTP, example http://127.0.0.1 base_dir: A directory to put the finished images, example /var/www/images num_workers: Number of workers to start, OBSOLETE, USE INITSCRIPT OR MANUALLY RUN MANY PARTICIPANTS post_creation: Path to a script to run after the image is created use_kvm: Wether to use a virtual machine, values are "yes" or "no"
mic_opts: Example, mic_opts = --save-kernel, --use_comps, so comma separated options
amqp_host: Host that runs BOSS amqp_user: Username to tap into the BOSS virtual host amqp_pwd: Password amqp_vhost: The actual virtual host to connect to
KVM Image
The KVM image must be created in such a way that it has openssh and mic2, along with having ssh-key based login for root, which uses a certain ssh public key.
The format of the image is recommended to be qcow2, with preallocation and cluster_size=2M as the qemu-img options. This is because the worker code uses qcow2 to create the overlay, thus saving time if the image is in the same format.
If the host is using openSUSE, please enable the following things for zypper repositories:
http://download.opensuse.org/repositories/Kernel:/openSUSE-11.3/openSUSE_11.2/ http://download.opensuse.org/repositories/Virtualization/openSUSE_11.2/
And add kvm-intel to /etc/sysconfig/kernel.
Virtual machine
Currently, IMGer can run the MIC2 jobs inside a KVM virtual machine, the only prerequisites are that image is located in /usr/share/img/base.img (configurable in the future) path and that the virtual machine has MIC2 installed and configured.
Suse studio is one way to create the images, just make sure that you have a rpm repository which contains mic2, python-xml and qemu-static-arm to make mic2 run properly on the image.
One must also confirm that the image SSH-keys in the source distribution are configured in the virtual machine and both keys exist in /usr/share/img (configurable in the future). During the runtime of IMGer, it is possible that ssh-keys inside the image may change, this is already handled by IMGer by ignoring the host key checking.
BOSS Workitem definition
Input
The following defines the input dictionary for the participant:
"kickstart" : a kickstart file, in raw, not a path "email": email address, not parsed so can be a dummy address "id": a simple UUID, eg. from pythons uuid library "type":type, from values: livecd, liveusb, loop, raw, nand, mrstnand, vdi or vmdk "name":name, a simple string for the image name "arch":architecture, for image "release":release
Output
The work item can contain the following dictionary entries:
Status: Status, is either "DONE" or "ERROR" in the final workitem, depending on success/failure URL: URL to a image/log/kickstart file location on the worker Error: An error string with details of the error, if any Image: A direct URL to the finished image Log: A direct URL to the log file of the mic2 output