Upgrade to Synnefo v0.16

Introduction

Starting with version 0.16, we introduce Archipelago as the new storage backend for the Pithos Service. Archipelago will act as a storage abstraction layer between Pithos and NFS, RADOS or any other storage backend driver that Archipelago supports. In order to use the Pithos Service you must install Archipelago on the node that runs the Pithos and Cyclades workers. Additionally, you must install Archipelago on the Ganeti nodes and upgrade snf-image to version 0.16.2 since this is the first version that supports Archipelago.

Until now the Pithos mapfile was a simple file containing a list of hashes that make up the stored file in a Pithos container. After this consolidation the Pithos mapfile had to be converted to an Archipelago mapfile. An Archipelago mapfile is an updated version of the Pithos mapfile, intended to supersede it.

More info about the new mapfile you can find in Archipelago documentation.

Current state regarding ownerships and permissions

In Synnefo v0.15, all Synnefo components run as www-data:www-data. Also, in case Pithos is using the NFS backend store, the files in the shared directory are owned by www-data:www-data with 0640 permissions. Finally, Archipelago, if used, runs as root:root.

Synnefo v0.16 provides more flexibility regarding required users and permissions. Synnefo introduces a dedicated system user and group: synnefo. On the other hand, Archipelago v0.4 is able to run as an arbitrary user/group (defaults to archipelago:archipelago).

As already mentioned, in Synnefo v0.16, Archipelago is becoming the storage backend for Pithos, so we must guarantee that Pithos will have the right permissions to communicate with Archipelago. For this reason we run Archipelago with the synnefo group.

Finally, in case NFS is used as a storage backend, we must also update the permissions and ownerships of all directories and files on the exported directory. Because this procedure may take a while in a production setup with many TB of data, these upgrade notes provide a detailed procedure in order to be able to perform the transition with minimum downtime.

At the end of the day Synnefo (Cyclades, Pithos, Astakos, etc) will run as synnefo:synnefo while Archipelago will run as archipelago:synnefo. The NFS (if any) will be owned by archipelago:synnefo with 2660 permissions.

Upgrade Steps

The upgrade to v0.16 consists of the following steps:

  1. Upgrade snf-image on all Ganeti nodes
  2. Setup Archipelago on all nodes
  3. Ensure intermediate state
  4. Bring down services and backup databases
  5. Upgrade packages, migrate the databases and configure settings
  6. Inspect and adjust resource limits
  7. Tweak Gunicorn settings on Pithos and Cyclades node
  8. Bring up all services
  9. Finalize permissions
  10. Add unique names to disks of all Ganeti instances

Warning

It is strongly suggested that you keep separate database backups for each service after the completion of each step.

0. Upgrade snf-image on all Ganeti nodes

On all Ganeti VM-capable nodes install the latest snf-image package (v0.16.3).

# apt-get install snf-image

1. Setup Archipelago on all nodes

At this point, we will perform some intemediate migration steps in order to perform the upgrade procedure with minimum downtime. To achieve this, we will pass through an intermediate state where:

  • Pithos will run as www-data:synnefo.
  • Archipelago will run as archipelago:synnefo.
  • The NFS shared directory will be owned by www-data:synnefo with 2660 permissions.

To ensure seamless transition we do the following:

  • Create system users and groups in advance

    NFS expects the user and group ID of the owners of the exported directory to be common across all nodes. So we need to guarantee that ID of archipelago user/group and synnefo group will be the same to all nodes. So we modify the archipelago user and group and create the synnefo user (assuming that ids 200 and 300 are available everywhere), by running the following commands to all nodes that have archipelago installed:

    # addgroup --system --gid 200 synnefo
    # adduser --system --uid 200 --gid 200 --no-create-home \
        --gecos Synnefo synnefo
    
    # addgroup --system --gid 300 archipelago
    # adduser --system --uid 300 --gid 300 --no-create-home \
        --gecos Archipelago archipelago
    

    Normally the snf-common and archipelago packages are responsible for creating the required system users and groups.

  • Upgrade/Install Archipelago

    Up until now Archipelago was optional. So, your setup, either has no Archipelago installation or has Archipelago v0.3.5 installed and configured in all VM-capable nodes. Depending on your case refer to:

    Archipelago does not start automatically after installation. Do not start it manually until it is configured properly.

  • Adjust Pithos umask setting

    On the Pithos node, edit the file /etc/synnefo/20-snf-pithos-app-settings.conf and uncomment or add the PITHOS_BACKEND_BLOCK_UMASK setting and set it to value 0o007.

    Then perform a gunicorn restart on both nodes:

    # service gunicorn restart
    

    This way, all files and directories created by Pithos will be writable by the group that Pithos is running (i.e. www-data).

  • Change Pithos data group permissions

    Ensure that every file and folder under Pithos data directory has correct permissions.

    # find /srv/pithos/data -type d -exec chmod g+rwxs '{}' \;
    # find /srv/pithos/data -type f -exec chmod g+rw '{}' \;
    

    This way, we prepare NFS to be fully accessible either via the user or the group.

  • Change gunicorn group

    On the Pithos node, edit the file /etc/gunicorn.d/synnefo and set group to synnefo. Then change the ownership of all configuration and log files:

    # chgrp -R synnefo /etc/synnefo
    # chgrp -R synnefo /var/log/synnefo
    # /etc/init.d/gunicorn restart
    

    This way, Pithos is able to access NFS via gunicorn user (www-data). We prepare Pithos to be able to access the synnefo group.

  • Change Pithos data group owner

    Make synnefo group the group owner of every file under the Pithos data directory.

    # chgrp synnefo /srv/pithos/data
    # find /srv/pithos/data -type d -exec chgrp synnefo '{}' \;
    # find /srv/pithos/data -type f -exec chgrp synnefo '{}' \;
    

    From now on, every file or directory created under the Pithos data directory will belong to the synnefo group because of the directory SET_GUID bit that we set on a previous step. Plus the synnefo group will have full read/write access because of the adjusted Pithos umask setting.

  • Make archipelago run as synnefo group

    Change the Archipelago configuration on all nodes, to run as archipelago:synnefo, since it no longer requires root priviledges. For each Archipelago node:

    • Stop Archipelago

      # archipelago stop
      
    • Change the USER and GROUP configuration option to archipelago and synnefo respectively. The configuration file is located under /etc/archipelago/archipelago.conf

    • Change the ownership of Archipelago log files:

      # chown -R archipelago:synnefo /var/log/archipelago
      
    • Start Archipelago

      # archipelago start
      

2. Ensure intermediate state

Please verify that Pithos runs as www-data:synnefo and any file created in the exported directory will be owned by www-data:synnefo with 660 permissions. Archipelago runs as archipelago:synnefo so it can access NFS via the synnefo group. NFS (blocks, maps, locks and all other subdirectories under /srv/pithos/data or /srv/archip) will be owned by www-data:synnefo with 2770 permissions.

3. Bring web services down, backup databases

  1. All web services must be brought down so that the database maintains a predictable and consistent state during the migration process:

    $ service gunicorn stop
    $ service snf-dispatcher stop
    $ service snf-ganeti-eventd stop
  2. Backup databases for recovery to a pre-migration state.

  3. Keep the database servers running during the migration process.

4. Upgrade Synnefo and configure settings

4.1 Install the new versions of packages

astakos.host$ apt-get install \
                        python-objpool \
                        snf-common \
                        python-astakosclient \
                        snf-django-lib \
                        snf-webproject \
                        snf-branding \
                        snf-astakos-app

cyclades.host$ apt-get install \
                        python-objpool \
                        snf-common \
                        python-astakosclient \
                        snf-django-lib \
                        snf-webproject \
                        snf-branding \
                        snf-pithos-backend \
                        snf-cyclades-app

pithos.host$ apt-get install \
                        python-objpool \
                        snf-common \
                        python-astakosclient \
                        snf-django-lib \
                        snf-webproject \
                        snf-branding \
                        snf-pithos-backend \
                        snf-pithos-app \
                        snf-pithos-webclient

ganeti.node$ apt-get install \
                        python-objpool \
                        snf-common \
                        snf-cyclades-gtools \
                        snf-pithos-backend \
                        snf-network \
                        snf-image

Note

Make sure snf-webproject has the same version with snf-common

Note

Installing the packages will cause services to start. Make sure you bring them down again (at least gunicorn, snf-dispatcher)

Note

If you are using qemu-kvm from wheezy-backports, note that qemu-kvm package 2.1+dfsg-2~bpo70+2 has a bug that is triggered by snf-image. Check snf-image installation for a workaround.

4.2 Sync and migrate the database

Note

If you are asked about stale content types during the migration process, answer ‘no’ and let the migration finish.

astakos-host$ snf-manage syncdb
astakos-host$ snf-manage migrate

cyclades-host$ snf-manage syncdb
cyclades-host$ snf-manage migrate

pithos-host$ pithos-migrate upgrade head

4.3 Configure snf-vncauthproxy

Synnefo 0.16 replaces the Java VNC client with an HTML5 Websocket client and the Cyclades UI will always request secure Websocket connections. You should, therefore, provide snf-vncauthproxy with SSL certificates signed by a trusted CA. You can either copy them to /var/lib/vncauthproxy/{cert,key}.pem or inform vncauthproxy about the location of the certificates (via the DAEMON_OPTS setting in /etc/default/vncauthproxy).

DAEMON_OPTS="--pid-file=$PIDFILE --cert-file=<path_to_cert> --key-file=<path_to_key>"

Both files should be readable by the vncauthproxy user or group.

Note

When installing snf-vncauthproxy on the same node as Cyclades and using the default settings for snf-vncauthproxy, the certificates should be issued to the FQDN of the Cyclades worker. Refer to the admin guide, for more information on how to setup vncauthproxy on a different host / interface.

For more information on how to setup snf-vncauthproxy check the snf-vncauthproxy documentation and upgrade notes.

4.4 Re-register service and resource definitions

The Cyclades service definition has been updated and needs to be registered again. On the Astakos node, run:

astakos-host$ snf-component-register cyclades

This will detect that the Cyclades component is already registered and ask to re-register. Answer positively. You need to enter the base URL and the UI URL for Cyclades, just like during the initial registration.

Note

You can run snf-manage component-list -o name,base_url,ui_url to inspect the currently registered base and UI URLs.

5. Inspect and adjust resource limits

Synnefo 0.16 brings significant changes at the project mechanism. Projects are now viewed as a source of finite resources, instead of a means to accumulate quota. They are the single source of resources, and quota are now managed at a project/member level.

System-provided quota are now handled through special purpose user-specific system projects, identified with the same UUID as the user. These have been created during the database migration process. They are included in the project listing with:

snf-manage project-list --system-projects

All projects must specify quota limits for all registered resources. Default values have been set for all resources, listed with:

astakos-host$ snf-manage resource-list

Column system_default (previously known as default_quota) provides the skeleton for the quota limits of user-specific system projects. Column project_default is new and acts as skeleton for applied (non-system) projects (i.e., for resources not specified in a project application). Project defaults have been initialized during migration based on the system default values: they have been set to inf if system_default is also inf, otherwise set to zero.

This default, affecting all future projects, can be modified with:

astakos-host$ snf-manage resource-modify <name> --project-default <value>

Till now a project definition contained one quota limit per resource: the maximum that a member can get from the project. A new limit is introduced: the grand maximum a project can provide to its members. This new project limit is initialized during migration as max members * member limit (if max members is not set, the double of current active members is assumed).

Existing projects can now be modified directly through the command line. In order to change a project’s resource limits, run:

astakos-host$ snf-manage project-modify <project_uuid> --limit <resource_name> <member_limit> <project_limit>

With the new mechanism, when a new resource is allocated (e.g., a VM or a Pithos container is created), it is also associated with a project besides its owner. The migration process has associated existing resources with their owner’s system project. Note that users who had made use of projects to increase their quota may end up overlimit on some resources of their system projects and will need to reassign some of their reserved resources to another project in order to overcome this restriction.

6. Tweak Gunicorn settings

First we make Gunicorn run as synnefo:synnefo, by setting the user and group option in Gunicorn configuration file (/etc/gunicorn.d/synnefo).

Also on the Pithos and Cyclades node you also have to set the following:

  • --config=/etc/synnefo/gunicorn-hooks/gunicorn-archipelago.py

Warning

If you have already installed Synnefo v0.16rc1 or v0.16rc2 you should replace pithos.conf.py with gunicorn-archipelago.py located under /etc/synnefo/gunicorn-hooks directory. Afterwards you can freely delete pithos.conf.py conf file.

After setting the user/group that Gunicorn will run as, we must also make sure that configuration and log files are accessible:

# chgrp -R synnefo /etc/synnefo/
# chown -R synnefo:synnefo /var/log/synnefo/

On the Cyclades node, the snf-dispatcher must run as synnefo:synnefo. In /etc/default/snf-dispatcher verify that SNF_USER and SNF_DSPTCH_OPTS settings are:

SNF_USER="synnefo:synnefo"
SNF_DSPTCH_OPTS=""

Finally, verify that snf-dispatcher can access its log file (e.g. /var/log/synnefo/synnefo.log):

# chown synnefo:synnefo /var/log/synnefo/dispatcher.log

7. Bring all services up

After the upgrade is finished, we bring up all services:

astakos.host  # service gunicorn start
cyclades.host # service gunicorn start

pithos.host   # service gunicorn start

cyclades.host # service snf-dispatcher start

8. Finalize permissions

At this point, and while the services are running, we will finalize the permissions of existing directories and files in the NFS directory to match the user/group that Archipelago is running:

# chown -R archipelago:synnefo /srv/pithos/data

9. Add unique names to disks of all Ganeti instances

Synnefo 0.16 introduces the Volume service which can handle multiple disks per Ganeti instance. Synnefo assigns a unique name to each Ganeti disk and refers to it by that unique name. After upgrading to v0.16, Synnefo must assign names to all existing disks. This can be easily performed with a helper script that is shipped with version 0.16:

cyclades.host$ /usr/lib/synnefo/tools/add_unique_name_to_disks