don't use DHCP for network configuration by defaultSetting static IP configuration on interfaces
is more reliable, than using DHCP. At least now
DHCP doesn't work, if you set external gateway on
router (because we don't have full namespaces support
and DHCP replies are being SNAT'ed).
assume we always use neutron's firewallLet's forbid nova's firewall, since it doesn't work
anyway (because we need to change devname in network_info
into appropriate value vethXXX.X or vmeXXXX).
Also let's always connect instances to integration bridge
via intermediate linux bridge.
Let's keep VifOvsEthernet class and all methods, which tune
nova firewall. Maybe someone will need it later.
add firewall class for neutronHere is how neutron with OVS works:
br-int <pcsvoXXX>|--|<pcsvbXXX> pcsbrXXX <veth100.0>|---<eth0 in CT>
First, we have integration bridge br-int and veth pair, one
end of which is inside container, another one (veth100.0) -
in host. We have to connect veth100.0 to bridge br-int.
Also neutron need to setup security group rules between
container and br-int. IptablesFirewallDriver class does it...
don't rename PCS's network devicesInstead of renaming PCS devices 'vethXXX.X' and 'vmeXXX'
we can provide name to netutron's firewall driver with help
of OVS's external_ids.
So first, don't rename devices and second, add iface-name
external id to OVS port (linux_net.create_ovs_vif_port
can't do it, so add function pcs_create_ovs_vif_port).
implement networking with help of prlsdkapiWe can't use vzctl for VMs, so it's needed to manage network in VEs
using prlsdkapi.
It's not possible to get device in host with given name using only
prlsdkapi, so I use 2 hacks:
1. If you specify unexistent virtual network id, device will not be
connected to any bridges, and VE will be started correctly.
2. Interface can be renamed if it's down.
So all works exactly as before: we creat...
add format_mac functionMAC addresses can have different string formats. So we
can't compare them as is. format_mac will convert MAC
address to format XX:XX:XX:XX:XX:XX, so macs returned by
this function could be compared.
pass prlsdkapi.Vm object to vif driverCreating Vm object by given uuid or name of the VE is relatively slow
operation and should be avoided if possible. So let's pass Vm object to
plug and unplug methods of vif driver. Consequent patches will need it.
fix integer options in pcs-glance-tool--min-disk and --min-ram options has type int, so corresponding
'type' kerword argument should be given to @arg decorator.
fix snapshot method for VMsadd VM support to snapshot method. We have to set disk_format to
ploop-vm in this case.
Snapshot for VMs can be created only if VM is powered off, because of
limitation in prlsdkapi.
use disk_format to distinguish containers and VMsUse disk_format to distinguish containers and VMs instead of
container_format. When we will add support of booting from
volumes it will be easier to check, which virtualization type
is supported by specific volume.
ez templates support will be removed soon, so let's also
use another disk_format for them.
fix pcs-glance-tool for VMsget_home_path works differently for VMs and containers, for containers
it returns path to private area, but for VMs - path to cofnig file
config.pvs.
add basic support of VMsAdd support of running VMs as openstack instances. You have to create
image in glance with container format pcs-vm and then launch it,
proper type of virtualization will be automatically selected.
Netwokirng is not supported yet, because vzctl is used to configure
network in container and that place have to be fixed before enabling
networking in VMs.
change disk size correctlyChanging disk size is not a change in a config file, so it should be
outsize of begin_edit/commit. Also this is an asynchronous operation
and return job object, which have to be wait'ed.
add tool for creating glance images from PCS VEsAdd script pcs-glance-tool which creates glance images by given
container or VM name.
It's a sort of an extension to glance command, so it supports all
common arguments, which are supported by glance and environment
variables (like --os-auth-url, --os-tenant-id or OS_USERNAME env var).
Example of usage:
./pcs-image-util.py --os-image-api-version 2 \
image-create my-container --name ne...
rename install-image.py to add-ez-image.pyI'll add another script for creating image from existing container or
VM, which also will do 'install-image', so let's rename this one.
rename image container_format to pcs-containerI'm going to add support for VMs and they should have another
container format since it's should be possible if we have to
create container or VM from given image.
Templates of both containers and VMs called 'golden image'
in our terminology, so let's use container formats
'pcs-container' and 'pcs-vm'.
rename PCSDriver._psrv to psrvThere is no need to have name of this variable starting with
underscope, because all parts of the driver need to work with
SDK and have to use prlsdkapi.Server object.
implement methods for firewall managementAdd a set of methods, which do different operations with
firewall. In case of PCS driver they only call the same methods
from nova's firewall driver.
fix bug with setting up VNCget_vnc_port method of Vm class doesn't requests new VNC port from
server, it just returns a value, which was retrieved with
get_vm_config SDK call (0, if VNC was disabled).
So after setting up VNC we have to request new VM config from server.
add helper function for sdk error codes conversionParallels SDK provides error codes in exceptions different from
constants defined. Special function conv_error must be used to
convert condtant to error code in python. So let's add helper function
get_sdk_errcode.
add support of spawning instances from snapshotsAdd ability to create an instance based on snapshot.
GoldenImageTemplate is added for this purpose. get_template will return
proper PCSTemplate object depending on container_type property of an
image.
remove network adapters from VE before saving imageWe don't need to store information about network adapters in snapshot.
Moreover disk data is the only thing we need to store in snapshot, but
it's not possible with parallels SDK to create container from existing
disk.
We will add new network adapters each time in spawn method. So
snapshot method is a good place to clean network configuration.
don't store unneded attributes in EzTemplate classcontext and image_ref are used only in __init__ in EzTemplate class. So
no need to store them as attributes.
use special container types in glanceNow we have to different types of images in glance: ez templates and
tarred containers. So we need to distinguish them. Let's use container
types 'pcs-ez' and 'pcs-golden-image' for this purpose.
refactor code for implementing new template typeAbility to create instances from snapshot is needed, because now you can
only create snapshot, but can't create a new instance from it. This
paths does some preparation before implementing that functinality.
* Make PCSTemplate base class and add EzTemplate class.
* Move code for creating container to Template class
(create_instance method) and add function get_template,
which will select p...
implement snapshot functionPCS support clone-to-template operation, which clones existing container
to a container of special type. It, in turn, can be cloned later and the
new container will be new auto-generated mac addresses on network
adapter, new ID, new name, e.t.c.
Image in glance is an array of data and metadata on form of key:value
pairs. So we can't upload directory with container as-is. Let's use tar
to conve...
call unplug_vifs before plug_vifs in power_onOpenstack thinks that power_off operation will not destroy network
interfaces and VM still will be connected to virtual port. So all
other drivers don't call unplug_vifs in power_off and plug_vifs
in power_on.
If we will not call unplug/plug_vifs - veth devices will be destroyed
on container shutdown, but port in openvswitch will be left and
when veth device will appear this port will not work...
do set_ram_size together with set_resourcephyspages will not change if you set only PHYSPAGES parameter using
ve.set_resource(PCR_PHYSPAGES, x). You also need to call set_ram_size.
apply flavor parameters to the containerTune number of cpus, memory, swap and root disk size.
Since we create containers with vSwap memory model, memory and swap
sizes simply map to the physpages and swappages container parameters.
Containers are created with existing disk, so we have to do resize of
the ploop disk after creating.
implement get_vnc_console operationBy default VNC is not started for containers and VMs in PCS. So we have
to enable it, if it's not done before. First call to this function will
start VNC server, subsequent ones - only get VNC port and return it.
implement suspend/resume operationsThese operations calls corresponding functions from parallels SDK. Also
resume operation calls plug_vifs to ensure, that instance is connected
to network.
implement reboot operationImplement reboot operation. Now it only works if instance is started, so
it has to be extended to all other possible states, because base
ComputeDriver class requires it.
keep host_ifname of veth during dhcp configurationYou have to provide --host_ifname together with --ifname to command,
which tunes veth parameters, or interface name in the host will be
changed to default vethXXX.X.
vif: don't plug iface to bridge, if it already therenova calls plug_vifs after start, but if an instance already started -
iface could be plugged into bridge. So we have to check, if we really
need to plug interface into bridge.
add script for creating glance images from rpmsIt's very unconvenient to execute glance command with a lot
of arguments to upload ez-template to glance. So add helper script,
which will do all that work.
Usage:
./utils/install-image.py <rpm file>
create instances, based on images from glanceGlance will store rpm files of EZ templates. When you try to
create an instance based on a image from glance, pcs driver
will first retrieve info about rpm from glance:
If the image has properties pcs_name, pcs_version and pcs_release -
they will be used, if not - compute driver will download rpm and
get name and version from an rpm file.
If rpm is already installed and has the same or higher...
Dmitry Guryanov