When I use SandyBridge as custom cpu in my testing, vm failed to start due to following error:
```
org.libvirt.LibvirtException: unsupported configuration: guest and host CPU are not compatible: Host CPU does not provide required features: avx, xsave, aes, tsc-deadline, x2apic, pclmuldq
```
With this patch, it works with the following setting in agent.properties:
```
guest.cpu.mode=custom
guest.cpu.model=SandyBridge
guest.cpu.features=-avx -xsave -aes -tsc-deadline -x2apic -pclmuldq
```
vm cpu is defined as below:
```
<cpu mode='custom' match='exact'>
<model fallback='allow'>SandyBridge</model>
<feature policy='disable' name='avx'/>
<feature policy='disable' name='xsave'/>
<feature policy='disable' name='aes'/>
<feature policy='disable' name='tsc-deadline'/>
<feature policy='disable' name='x2apic'/>
<feature policy='disable' name='pclmuldq'/>
</cpu>
```
On first startup, the management server creates and saves a random
ssh keypair using ssh-keygen in the database. The command does
not specify keys in PEM format which is not the default as generated
by latest ssh-keygen tool.
The systemvmtemplate always needs re-building whenever there is a change
in the cloud-early-config file. This also tries to fix that by introducing a
stage 2 bootstrap.sh where the changes specific to hypervisor detection
etc are refactored/moved. The initial cloud-early-config only patches
before the other scripts are called.
Signed-off-by: Rohit Yadav <rohit.yadav@shapeblue.com>
This fixes the issue that VM with VMsnapshots fails to start after
extract volume is done on a stopped VM, on VMware.
Signed-off-by: Rohit Yadav <rohit.yadav@shapeblue.com>
- Fixes PR #3146 db cleanup to the correct 4.12->4.13 upgrade path
- Fixes failing unit test due to jdk specific changes after forward
merging
Signed-off-by: Rohit Yadav <rohit.yadav@shapeblue.com>
This introduces a new patching script for patching systemvms on KVM
using qemu-guest-agent that runs inside the systemvm on startup. This
also removes the vport device which was previously used by the legacy
patching script and instead uses the modern and new uniform guest
agent vport for host-guest communication.
Also updates the sytemvmtemplate build config to use the latest Debian
9.9.0 iso.
Signed-off-by: Rohit Yadav <rohit.yadav@shapeblue.com>
* ubuntu16: fix unable to add host if cloudbrX is not configured
while add a ubuntu16.04 host with native eth0 (cloudbrX is not configured),
the operation failed and I got the following error in /var/log/cloudstack/agent/setup.log
```
DEBUG:root:execute:ifconfig eth0
DEBUG:root:[Errno 2] No such file or directory
File "/usr/lib/python2.7/dist-packages/cloudutils/serviceConfig.py", line 38, in configration
result = self.config()
File "/usr/lib/python2.7/dist-packages/cloudutils/serviceConfig.py", line 211, in config
super(networkConfigUbuntu, self).cfgNetwork()
File "/usr/lib/python2.7/dist-packages/cloudutils/serviceConfig.py", line 108, in cfgNetwork
device = self.netcfg.getDefaultNetwork()
File "/usr/lib/python2.7/dist-packages/cloudutils/networkConfig.py", line 53, in getDefaultNetwork
pdi = networkConfig.getDevInfo(dev)
File "/usr/lib/python2.7/dist-packages/cloudutils/networkConfig.py", line 157, in getDevInfo
elif networkConfig.isBridge(dev) or networkConfig.isOvsBridge(dev):
```
The issue is caused by commit 9c7cd8c2485412bc847b2c2473b962fa01435b24
2017-09-19 16:45 Sigert Goeminne ● CLOUDSTACK-10081: CloudUtils getDevInfo function will now return "bridge" instead o
* ubuntu16: Stop service libvirt-bin.socket while add a host
service libvirt-bin.socket will be started when add a ubuntu 16.04 host
DEBUG:root:execute:sudo /usr/sbin/service libvirt-bin start
However, libvirt-bin service will be broken by it after restarting
Stopping service libvirt-bin.socket will fix the issue.
An example is given as below.
```
root@node32:~# /etc/init.d/libvirt-bin restart
[ ok ] Restarting libvirt-bin (via systemctl): libvirt-bin.service.
root@node32:~# virsh list
error: failed to connect to the hypervisor
error: no valid connection
error: Failed to connect socket to '/var/run/libvirt/libvirt-sock': No such file or directory
root@node32:~# systemctl stop libvirt-bin.socket
root@node32:~# /etc/init.d/libvirt-bin restart
[ ok ] Restarting libvirt-bin (via systemctl): libvirt-bin.service.
root@node32:~# virsh list
Id Name State
----------------------------------------------------
```
* ubuntu16: Diable libvirt default network
By default, libvirt will create default network virbr0 on kvm hypervisors.
If vm uses the same ip range 192.168.122.0/24, there will be some issues.
In some cases, if we run tcpdump inside vm, we will see the ip of kvm hypervisor as source ip.
* Mock Scanner, instead of scan the computer running the test.
This allows non linux machines to run the tests without scanning for a
non existing /proc/meminfo.
* test fixes on 'other' platforms libvirt wrapper unit tests (#3)
* Fix XenServer Security Groups 'vmops' script
- fix tokens = line.split(':') to tokens = line.split(';')
- fix expected tokens size from 5 to 4
- enhance logs
- remove unused vmops script. The XCP patch points to the vmops script
on the parent folder [1]. Thus, all XenServer versions are considering
the vmops script located at [2].
- fix UI ipv4/ipv6 cidr validator to allow a list of cidirs.
Fixing issue: #3192 Security Group rules not applied at all for
XenServer 6.5 / Advanced Zone
https://github.com/apache/cloudstack/issues/3192
* Update security group rules after VM migration
Add security group rules on target host
Cause: vmops script expected secondary IPs as "0;" but received "0:"
Remove security group network rules on source host.
Cause: destroy_network_rules_for_vm function on vmops script was not
called when migrating VM
* Add unit tests and address reviewers
* Keep iotune section in the VM's XML after live migration
When live migrating a KVM VM among local storages, the VM loses the
<iotune> section on its XML, therefore, having no IO limitations.
This commit removes the piece of code that deletes the <iotune> section
in the XML.
* Add test for replaceStorage in LibvirtMigrateCommandWrapper
Signed-off-by: Wido den Hollander <wido@widodh.nl>
* Fix Javadoc for method replaceIpForVNCInDescFile
* feature: add libvirt / qemu io bursting
Adds the ability to set bursting features from libvirt / qemu
This allows you to utilize the iops and bytes temporary "burst" mode
introduced with libvirt 2.4 and improved upon with libvirt 2.6.
https://blogs.igalia.com/berto/2016/05/24/io-bursts-with-qemu-2-6/
* updates per rafael et al
* api: add command to list management servers
* api: add number of mangement servers in listInfrastructure command
* ui: add block for mangement servers on infra page
* api name resolution method cleanup
* - Offline VM and Volume migration on Vmware hypervisor hosts
- Also add VM disk consolidation call on successful VM migrations
* Fix indentation of marvin test file and reformat against PEP8
* * Fix few comment typos
* Refactor debug messages to use String.format() when debug log level is enabled.
* Send list of commands returned by hypervisor Guru instead of explicitly selecting the first one
* Fix unhandled NPE during VM migration
* Revert back to distinct event descriptions for VM to host or storage pool migration
* Reformat test_primary_storage file against PEP-8 and Remove unused imports
* Revert back the deprecation messages in the custom StringUtils class to favour the use of the ApacheUtils
The KVM Agent had two mechanisms for reporting its capabilities
and memory to the Management Server.
On startup it would ask libvirt the amount of Memory the Host has
and subtract and add the reserved and overcommit memory.
When the HostStats were however reported to the Management Server
these two configured values on the Agent were no longer reported
in the statistics thus showing all the available memory in the
Agent/Host to the Management Server.
This commit unifies this by using the same logic on Agent Startup
and during statistics reporting.
memory=3069636608, reservedMemory=1073741824
This was reported by a 4GB Hypervisor with this setting:
host.reserved.mem.mb=1024
The GUI (thus API) would then show:
Memory Total 2.86 GB
This way the Agent properly 'lies' to the Management Server about its
capabilities in terms of Memory.
This is very helpful if you want to overprovision or undercommit machines
for various reasons.
Overcommitting can be done when KSM or ZSwap or a fast SWAP device is
installed in the machine.
Underprovisioning is done when the Host might run other tasks then a KVM
hypervisor, for example when it runs in a hyperconverged setup with Ceph.
In addition internally many values have been changed from a Double to a Long
and also store the amount of bytes instead of Kilobytes.
Signed-off-by: Wido den Hollander <wido@widodh.nl>
* security group: Replace deprecated optparse by argparse
Starting with Python 2.7 the library optparse has been replaced by
argpase.
This commit replaces the use of optparse by argparse
Signed-off-by: Wido den Hollander <wido@widodh.nl>
* security group: Remove LXC support from security_group.py
LXC does not work and has been partially removed from CloudStack already
Signed-off-by: Wido den Hollander <wido@widodh.nl>
* security group: Refactor libvirt code
Use a single function which properly throws an Exception when the
connection to libvirt fails.
Also simplify some logic, make it PEP-8 compatible and remove a unused
function from the code.
Signed-off-by: Wido den Hollander <wido@widodh.nl>
* security group: Raise Exception on execute() failure
If the executed command exists with a non-zero exit status we should
still return the output to the command, but also raise an Exception.
Signed-off-by: Wido den Hollander <wido@widodh.nl>
* security group: Use a function to determin the physical device of a bridge
We can not safely assume that the first device listed under a bridge is the
physical device.
With VXLAN isolation a vnet device can be attached to a bridge prior to the
vxlanXXXX device being attached.
We need to filter out those devices and then fetch the physical device attached
to the bridge.
In addition use the 'bridge' command instead of 'brctl'. 'bridge' is part of the
iproute2 utils just like 'ip' and should be considered as the new default.
This command is also available on EL6 and does not break any backwards compat.
Signed-off-by: Wido den Hollander <wido@widodh.nl>
* security group: --set is deprecated, use --match-set
These messages are seen in the KVM Agent log:
--set option deprecated, please use --match-set
Functionality does not change
Signed-off-by: Wido den Hollander <wido@widodh.nl>
* security group: PEP-8 and indentation fixes
There were a lot of styling problems in the code:
- Missing whitespace or exess whitespace
- CaMelCaSe function names and variables
- 2-space indentation instead of 4 spaces
This commit addresses those issues.
Signed-off-by: Wido den Hollander <wido@widodh.nl>
The additional queues can enhance the performance of the VirtIO SCSI disk
and it is recommended to set this to the amount of vCPUs a Instance is assigned.
The optional queues attribute specifies the number of queues for the
controller. For best performance, it's recommended to specify a value matching
the number of vCPUs. Since 1.0.5 (QEMU and KVM only)
Source: https://libvirt.org/formatdomain.html#elementsVirtio
Signed-off-by: Wido den Hollander <wido@widodh.nl>
The static method syncVolumeToRootFolder() from VmwareStorageLayoutHelper.java:146 has been incorrectly called and leads to an infinite recursive call that ends up in a StackOverflowError. This PR fixes this.
public static void syncVolumeToRootFolder(DatacenterMO dcMo, DatastoreMO ds, String vmdkName, String vmName) throws Exception { syncVolumeToRootFolder(dcMo, ds, vmdkName, null); } -> public static void syncVolumeToRootFolder(DatacenterMO dcMo, DatastoreMO ds, String vmdkName, String vmName) throws Exception { syncVolumeToRootFolder(dcMo, ds, vmdkName, vmName, null); }
* Allow KVM VM live migration with ROOT volume on file
* Allow KVM VM live migration with ROOT volume on file
- Add JUnit tests
* Address reviewers and change some variable names to ease future
implementation (developers can easily guess the name and use
autocomplete)
Added dummy and lo devices to be treated as a normal bridge slave devs.
Fixes#2998
Added two more device names (lo* and dummy*). Implemented tests. Code was refactored.
Improved paths concatenation code from "+" to Paths.get.
If a host has many routes this can be a magnitude faster then printing
all the routes and grepping for the default.
In some situations the host might have a large amount of routes due to
dynamic routing being used like OSPF or BGP.
In addition fix a couple of loglines which were throwing messages on
DEBUG while WARN and ERROR should be used there.
Signed-off-by: Wido den Hollander <wido@widodh.nl>
When vxlan://untagged is used for public (or guest) network, use the
default public/guest bridge device same as how vlan://untagged works.
Signed-off-by: Rohit Yadav <rohit.yadav@shapeblue.com>
These additional RBD features allow for faster lookups of how much space a RBD
image is using, but with the exclusive locking we prevent two VMs from writing
to the same RBD image at the same time.
These are the default features used by Ceph for any new RBD image.
Signed-off-by: Wido den Hollander <wido@widodh.nl>
This adds a new API updateVmwareDc that allows admins to update the
VMware datacenter details of a zone. It also recursively updates
the cluster_details for any username/password updates
as well as updates the url detail in cluster_details table and guid
detail in the host_details table with any newly provided vcenter
domain/ip. The update API assumes that there is only one vCenter per
zone. And, since the username/password for each VMware host could be different
than what gets configured for vcenter at zone level, it does not update the
username/password in host_details.
Previously, one has to manually update the db with any new vcenter details for the zone.
Signed-off-by: Rohit Yadav <rohit.yadav@shapeblue.com>
On actual testing, I could see that kvmheartbeat.sh script fails on NFS
server failure and stops the agent only. Any HA VMs could be launched
in different hosts, and recovery of NFS server could lead to a state
where a HA enabled VM runs on two hosts and can potentially cause
disk corruptions. In most cases, VM disk corruption will be worse than
VM downtime. I've kept the sleep interval between check/rounds but
reduced it to 10s. The change in behaviour was introduced in #2722.
Signed-off-by: Rohit Yadav <rohit.yadav@shapeblue.com>
Prevents errors while migrating VM from ISO:
Test 1: Deploy VM from ISO -> Live migrate VM to another host -> ERROR
Test 2: Register ISO using Direct Download on KVM -> Deploy VM from ISO -> Live migrate VM to another host -> ERROR
- Prevent NullPointerException migrating VM from ISO
- Prevent mount secondary storage on ISO direct downloads on KVM
Since we support only Ubuntu 16.04+ on master/4.12+, we can now use
the libvirt service name `libvirtd` for all distributions. This also
fixes an optional package name for libvirtd installation on Debian 9+.
Fixes#2909
Signed-off-by: Rohit Yadav <rohit.yadav@shapeblue.com>
Windows has support for several paravirt features that it will use when running on Hyper-V, Microsoft's hypervisor. These features are called enlightenments. Many of the features are similar to paravirt functionality that exists with Linux on KVM (virtio, kvmclock, PV EOI, etc.)
Nowadays QEMU/KVM can also enable support for several Hyper-V enlightenments. When enabled, Windows VMs running on KVM will use many of the same paravirt optimizations they would use when running on Hyper-V.
A number of years ago, a PR was introduced that added a good portion of the code to enable this feature set, but it was never completed. This PR enables the existing features. The previous patch set detailed in #1013 also included the tests.
By selecting Windows PV, the enlightenment additions will be applied to the libvirt configuration. This is support on Windows Server 2008 and beyond, so all currently supported versions of Windows Server.
In our testing, we've seen benchmark improvements of around 20-25% running on Centos 7 hosts and it is also supported on Centos/RHEL 6.5 and later. Testing on Ubuntu would be appreciated.
This tries to provide a threshold based fix for #2873 where swappinness of VR is not used until last resort. By limiting swappiness unless actually needed, the VR system degradation can be avoided for most cases. The other change is around not starting baremetal-vr by default on all VRs, according to the spec https://cwiki.apache.org/confluence/display/CLOUDSTACK/Baremetal+Advanced+Networking+Support only vmware VRs need to run it and that too only as the last step of the setup/completion, so we don't need to run it all the time.
Signed-off-by: Rohit Yadav <rohit.yadav@shapeblue.com>
[VMware] VM is not accessible after migration across clusters.
Once a VM is successfully started, don't delete the files associated with the unregistered VM, if the files are in a storage that is being used by the new VM.
Attempt to unregister a VM in another DC, only if there is a host associated with a VM.
This closes#556
When a Instance is (attempted to be) started in KVM Host the Agent
should not worry about the allocated memory on this host.
To make a proper judgement we need to take more into account:
- Memory Overcommit ratio
- Host reserved memory
- Host overcommit memory
The Management Server has all the information and the DeploymentPlanner
has to make the decision if a Instance should and can be started on a
Host, not the host itself.
Signed-off-by: Wido den Hollander <wido@widodh.nl>
