GitLab

Return ServiceUnavailable, instead of OverLimit, when pools of bridges,
mac-prefixes or IPs run out of resources.

Modify snf-manage {server, network, flavor}-list to list by default only
non-deleted entries. The `--deleted` flag can be used  to also include
deleted ones.

Conflicts:
	snf-cyclades-app/setup.py

Implement a new command which consumes all the messages from the
dead-letter queues and re-sends the messages back to the snf-dispatcher
in order to handle them again.

Messages are annotated with "from-dead-letter" field in order to detect
looping between the original exchange and the dead-letter exchange. If a
loop is detect the message is discarder, unless the --keep-zombies
option is used.

Dummy-save the Network when processing messages from Ganeti hooks
about a VM NICs. This is needed, because the UI uses the HTTP changes-sinces
when requesting VMs and Networks, and combines the result in order to show
the new NICs. Not updating the Network object, resulted in VMs IPv4
and IPv6 shown as not set.

Modify snf-dispatcher callbacks in order to reject messages that caused
an unknown exception. Because a dead-letter-exchange is associated
with the queues that snf-dispatcher is using, these messages will be
transfered to the dead-letter-exchange.

Modify snf-dispatcher in order to declare the dead-letter exchanges and
queues

to reflect support for multiple Ganeti backends and minor fixes.
Also remove it from the general Admin Guide and give it a separate
section in the index.rst

Also add support for custom headers when publishing messages.

Basic_nack method is not support by puka. We are implementing
basic_nack as basic_ack since they have the same effect when
multiple and requeue options are not used.

Use cache in order to speed-up the output of server-list

Copy mac2eui64 from snf-network to snf-cyclades-app, since it is
now used in the reconciliation process to caclculate the IPv6
address of the VMs.

Modify snf-dispatcher and add a timeout option to basic_wait method.  If
timeout is reached, the connection to the current RabbitMQ broker is
closed and dispatcher tries to connect to a different one.

This commit is supplementary of 6d27ead, which detects connection
failures using the TCP keepalive feature. However, while the connection
is active, the RabbitMQ broker may have a problem. This commit tries
to detect such problems, and considers idle connections such a case.

Use the TCP keepalive feature in amqp_puka in order to detect connection
errors to the RabbitMQ broker, and retry to a different one. Also setup
the keepalive parameters to relative small values, to detect failures
relatively soon.

Also, modify and-dispatcher to use the AMQP prefetch_count option of the
basic_consume method, which makes the RabbitMQ broker to not deliver
more than prefetch_count messages to the snf-dispatcher, before
receiving acknowledgement for the previous messages.

To reflect:

 * New wording of Astakos commands
 * Latest snf-image-{host, helper} version

When only detecting, reconciliation can not continue, if the
BackendNetwork object does not exist.

Create the IPPool associated with each Network, when creating it. Until
now, IPPools were created on first demand. However, because the Network
object is not accessed in exclusive mode, concurrent access to the
Network model resulted in a race condition. More than one workers may
find that no Pool is associated with the Network, and try to create a
new one.

Fix bug when connecting a network to a nodegroup in network
reconciliation. Use connect_network instead of obsolete
connect_network_group method. Also fix job dependency handling in this
method.