Age | Commit message (Collapse) | Author | Files | Lines |
|
* use raw devices instead of qcow2
* disable i/o caching, it reduces data copies and bus traffic
* set native aio instead of thread based
JIRA: FUEL-201
Change-Id: I8ac728cbb5be5948080fab8ae4d3d311e0a5c09c
Signed-off-by: Michael Polenchuk <mpolenchuk@mirantis.com>
|
|
Change-Id: I71790b8abb25936045f777a78ed903f3b8ce4303
Signed-off-by: wu.zhihui <wu.zhihui1@zte.com.cn>
|
|
Give more attempts for zte hardware.
Change-Id: I0906b6fe9e046ddd6436e5246494b89c812a40a2
Signed-off-by: wu.zhihui <wu.zhihui1@zte.com.cn>
|
|
|
|
We should allow the nodes some time to stabilize after sending a command
to them. Currently we are checking the status too fast.
Change-Id: I26a60d2bb6a43edbec842b727f825057e2778981
Signed-off-by: Josep Puigdemont <josep.puigdemont@enea.com>
|
|
|
|
Change-Id: I0103082672259eaf4b1c2be8ca3a1236fc6137eb
Signed-off-by: Peter Barabas <peter.barabas@ericsson.com>
|
|
Change-Id: I328432e6b30437e8a83065859cd80358d4c3c468
Signed-off-by: Peter Barabas <peter.barabas@ericsson.com>
|
|
With this patch it should be possible to create a fuel VM on a remote
libvirt server by properly defining the LIBVIRT_DEFAULT_URI [1]
environment variable. If the variable is not defined, then there should
be no percievable change in behaviour for the script.
This patch introduces the ability to create volumes (images) on a
remote libvirt host where the Fuel VM is to be deployed. For now
the volumes are created by default in a pool named jenkins, but
the idea is to allow this to be configured, probably in the POD's
DHA file.
Since all virsh commands honor LIBVIRT_DEFAULT_URI, we use this
environment variable to detect wheter we should create a volume or not.
The rationale being that the variable will only be set if the user wants
to to do the VM deployment on a remote libvirt host.
We need to create a volume because we can not rely on being able to
access the remote server's file system directly.
The images are then transferred to the libvirt host using virsh
commands. All this could also be done using scp and a user directory
on the host machine, but using pools allows us to take advantage of
libvirt's policies and file permissions.
CHANGE: when LIBVIRT_DEFAULT_URI is defined, the script will not check
for the presence of the required PXE bridge. This will still be checked
when the Fuel VM is started and the bridge not found, but this happens
at a later point than it does today.
CHANGE: before this patch, the file system image was named like the VM:
vm_name.raw. This patch introduces a change and adds a timestamp suffix
to the image: vm_name-timestamp.raw. This is so to avoid collisions with
an image with the same name on the remote pool (two PODs may be using
the same pool). It may also be useful to keep around old file system
images.
FIXME: This patch requires a pool named "jenkins" in the remote libvirt
server, and it will fail if it is not present. This should be
configurable.
Notice though that we can still define LIBVIRT_DEFAULT_URI as
"qemu:///system" to create the Fuel VM on the local host.
[1] https://libvirt.org/remote.html#Remote_URI_reference
Change-Id: I40925ed31337d3ad9cf505f284f5c3d14e9129a0
Signed-off-by: Josep Puigdemont <josep.puigdemont@enea.com>
|
|
The method get_node_state has been added to the the IpmiAdapter class.
In addition, now the power on/off methods will try several times to
perform their IPMI command before giving up, instead of bailing out at
the first error.
After the power on/off command is completed, the method will wait until
the node is in the desired state.
NOTE: a command could potentially take several minutes if the defaults
are used; each IPMI command can take up to 1 minute, and there can be 3
commands issued per operation, one of them may be retried 20 times with
the current defaults. Ideally we would use eventlet or something similar
to allow each command a limited time to execute, instead:
with eventlet.timeout.Timeout(seconds) as t:
power_on/off_command
Signed-off-by: Josep Puigdemont <josep.puigdemont@enea.com>
|
|
Sometimes the IPMI lanplus protocol listens on a non-standard
remote port, e.g. when target nodes are interfaced through a
fake IPMI BMC application that listens on multiple ports on the
same IP address.
Therefore, allow setting IPMI port in the DHA using a new
property named `ipmiPort`, and pass it along to `ipmitool` when set.
CHANGE: get_access_info now also supports specifying the IPMI
port to use with `ipmitool` by configuring the `ipmiPort`
property in the DHA.
hp_adapter.py: updated `get_access_info` return signature with
the new (unused there) `ipmiport`.
Change-Id: I620176bd7f466aa460518cf12d15ccbe86a22560
Signed-off-by: Alexandru Avadanii <Alexandru.Avadanii@enea.com>
|
|
This reverts commit 5926bcddca0eca28a33ae43aadf90b9263ae6b84.
Change-Id: Ibea4ca5d38ce2e10ab6d5bb1d7ccdf666ee149d5
Signed-off-by: Stefan K. Berg <stefan.k.berg@ericsson.com>
|
|
Change-Id: I1f6d38da24d5a4e66061b9bdfe576fbbda7fb624
Signed-off-by: Peter Barabas <peter.barabas@ericsson.com>
|
|
ZTE hardware need use IPMI command "chassis power cycle"
on function node_reset().
JIRA: FUEL-127
Change-Id: I286fd9cda43d2e1799b134f0a391f57d08cd1928
Signed-off-by: wu.zhihui <wu.zhihui1@zte.com.cn>
|
|
This commit enables the full configuration
of the VM(fuel/controller/compute)
defintion through the dha file.
Change-Id: I4e78334d1e5aec1e98667343390283587f0b3ea5
|
|
Some compones of openstack produce a lot of CPU load.
With this commit it is possible to
make more use of the Hypervisor where the virtual
nodes runs on.
Change-Id: Ide567dd0823c5526171c29073f2a36aa5f27d4b6
|
|
A lot of diskspace is allocated although never used
this commit changes to thin provisioning
for the virtual enviroment.
Change-Id: Ia128143a3bae9f3d85a9e749319c2f5343c1df4e
|
|
Use Python's tempfile instead of forking a shell for mktemp.
Change-Id: Icba442ea08d4098d67fb51575da16e85d4363beb
Signed-off-by: Peter Barabas <peter.barabas@ericsson.com>
|
|
Restructure of the directory layout due to move of Fuel into it's own repo
JIRA: FUEL-85
Change-Id: I3647e1992a508f29dce06a5d6c790725c527f6f5
Signed-off-by: Jonas Bjurel <jonas.bjurel@ericsson.com>
|