Detail: Cloudstack tries to stop VMs all the time, for all sorts of reasons,
but usually just to get into a known state. Libvirt throws an exception of
'Domain not found' when attempting to stop a VM that doesn't exist. This causes
problems for troubleshooting real issues. Domain not found should equate
to success if trying to stop.
BUG-ID: CLOUDSTACK-4011
Bugfix-for: 4.2
Signed-off-by: Marcus Sorensen <marcus@betterservers.com> 1375825281 -0600
- Move vnetBridge clean up function from LibvirtComputingResource to BridgeVifDriver
-- since only BridgeVifDriver have to handle this event
- LibvirtComputingResource now properly call VifDriver.unplug() when it receives UnPlugCommand
- Remove not working and no longer used method getVnet(String) from VirtualMachineName
- Remove not working and no longer used method getVnet() from StopCommand
- Remove unused constructer StopCommand(VirtualMachine, String, boolean) from StopCommand
- Remove unused member vnet from StopCommand
- Remove unused member _modifyVlanPath from OvsVifDriver
Tested with 2 KVM hosts and confirmed it correctly manipulate vnetBridge with start, stop, migrate, plug, and unplug event
Signed-off-by: Hugo Trippaers <htrippaers@schubergphilis.com>
Description:
a) Fixing NPE when wrong path is provided for primary datastore.
b) No error dialog shows up in GUI when wrong path is provided,
after NPE fix - propagating exception upward.
c) If the KVM agent is down, an invalid datastore gets logged in
storage_pool table and doesn't get removed, so it shows up
in the GUI in the list of datastores - fixing this as well.
- writeToFile removed since no references to it
- readFileAsString replaced with FileUtils.readFileToString
- minor code duplication removed in dependent method getNicStats
- unit test added
Signed-off-by: Laszlo Hornyak <laszlo.hornyak@gmail.com>
Renaming the method in the command objects to be uniform with
PlugNicCommand/UnplugNicCommand.getVmName
Signed-off-by: Prasanna Santhanam <tsp@apache.org>
inactive VM definitions block a new VM starting. Definitions aren't supposed to
be persistent, but sometimes a crash or failed migration can leave behind a
definition.
Signed-off-by: Marcus Sorensen <marcus@betterservers.com> 1370299734 -0600
RBD format 2 supports cloning (aka layering) where one base image can serve
as a parent image for multiple child images.
This enables fast deployment of a large amount of virtual machines, but it also
saves spaces on the Ceph cluster and improves performance due to better caching.
Qemu-img doesn't support RBD format 2 (yet), so to enable these functions the
RADOS/RBD Java bindings are required.
This patch also enables deployment of System VMs on RBD storage pools. Since we
no longer require a patchdisk for passing the boot arguments we are able to deploy
these VMs on RBD.
Detail: We do two strange things, #1, when a vm is created, we create the uuid
for the domain by converting the name into a uuid, then subsequently, any time
we look for a domain, we convert its name to a uuid and domainLookupByUUID. This
is an unnecessary obfuscation, so instead we just do a domainLookupByName. As
a bonus, we are now setting the domain's uuid to be the cloudstack uuid. It is
no longer used anywhere, but will be less confusing to admins. This shouldn't
affect upgrades, since we can always lookup existing VMs by name.
Signed-off-by: Marcus Sorensen <marcus@betterservers.com> 1369287049 -0600
Detail: Undefine VM after migration. Lingering domain definitions cause
migrations back to the original host to fail, since domain already exists.
BUG-ID: CLOUDSTACK-2640
Bugfix-for: 4.1.0,4.2.0
Signed-off-by: Marcus Sorensen <marcus@betterservers.com> 1369285950 -0600
so that callers of startVM in LibvirtComputingResource know that a vm failed
to start
Signed-off-by: Marcus Sorensen <marcus@betterservers.com> 1366224775 -0600
pools the proper way won't cause problems for KVM HA Monitor, this patch closes
holes. Call the KVMStoragePool deleteStoragePool that properly removes it from
the KVMHA hashmap, instead of the pools direct delete() call.
Signed-off-by: Marcus Sorensen <marcus@betterservers.com> 1366172318 -0600
- Supports DHCP, Source NAT, Static NAT, Firewall rules, Port Forwarding
- Renamed MidokuraMidonet to MidoNet
- Related Jira ticket is CLOUDSTACK-996
Signed-off-by: Dave Cahill <dcahill@midokura.com>
Signed-off-by: Hugo Trippaers <htrippaers@schubergphilis.com>
deleted NFS pools, causing failures when defining new storage pools. Sometimes
a storage pool has never been used on a host, and getStoragePool fails when
copying templates or in storage migration. deleteStoragePool(pool) often fails
silently, leaving no pool defined in libvirt, but a mountpoint left behind.
This patch handles some of these exceptions and brings forward any issues via
logging.
Signed-off-by: Marcus Sorensen <marcus@betterservers.com> 1364603486 -0600
KVM to manager. This adds collection of available storage to KVM, not
just used.
Bugfix-for: 4.0.2, 4.1, master
Submitted-by: Ted Smith <darnoth@gmail.com>
Signed-off-by: Marcus Sorensen <marcus@betterservers.com> 1363966235 -0600
The collection of network usage from VPC virtual router on KVM does not work,
because there is no corresponding procedure to deal with VPC virtual router
(cmd.isForVpc() = true).
Reviewed-by: Kishan Kavala <kishan@apache.org>
Reported-by: Wei Zhou <w.zhou@leaseweb.com>
Signed-off-by: Prasanna Santhanam <tsp@apache.org>
cloud-defined resources on the host has caused various problems. As a backward
compatible fix, if an existing pool with a different name collides with a pool
being created (by path), the pool will be redefined with the name cloudstack
knows about. This is actually what brought up the bug, a persisted storage pool
cloudstack wasn't managing.
Signed-off-by: Marcus Sorensen <marcus@betterservers.com> 1363210149 -0600
Detail: When we stop a VM, it's definition is no longer valid. Therefore, we
need to catch the exception thrown from libvirt in trying to lookup a
non-existent domain by UUID while trying to check if it's shut down.
BUG-ID:CLOUDSTACK-600
Signed-off-by: Marcus Sorensen <marcus@betterservers.com> 1363201066 -0600
Detail: A previous patch fixed an issue where we are defining VMs to persist
locally on KVM hosts, which can cause issues if the agent isn't running and
libvirt decides to start the VM unbeknownst to cloudstack. The previous patch
stopped defining VMs as persistent. This patch adds compatibility for existing
cloudstack environments, removing the persistent definition on stop if needed.
BUG-ID: CLOUDSTACK-600
Signed-off-by: Marcus Sorensen <marcus@betterservers.com> 1363194656 -0600
Detail: This gets rid of the patchdisk method of passing cmdline and
authorized_keys to KVM system VMs. It instead passes them to a virtio socket,
which the KVM guest reads from the character device /dev/vport0p1 during
cloud-early-config. Tested to work on CentOS 6.3 and Ubuntu 12.04. Should
work with even older versions of libvirt.
Signed-off-by: Marcus Sorensen <marcus@betterservers.com> 1362691685 -0700
Current kvm agent will silently ignore many exception, and there's no
way to see what really happened. This patch will log in trace level log
that was silently ignored. And also, it will fix huge bare Exception
catch, which is very harmful because it also catches RuntimeException.
Detail: This device can be used for remotely controlling the system vms through
a local socket on the host. We will attempt to replace the KVM patchdisk with
it. Tested, successfully deploys VM, and if system vm has proper driver it
will create a /dev/vport0p1 device within the VM. We will be updating the
system VM in 4.2/5.0 and will support this.
Signed-off-by: Marcus Sorensen <marcus@betterservers.com> 1362527352 -0700
Libvirt-java 0.4.9 works just fine with JNA 3.2.4 which is in
all distributions.
Future libvirt version require at least JNA 3.5.1 due to new methods
and memory management, but that is not our concern now.
By depending on the JNA in the distribution and adding it to the classpath
we can work just fine.
- Building map of {trafficType, vifDriver} at configure time
- Use the relevant VIF driver for the given traffic type when call plug()
- Inform all vif drivers when call unplug(), as we no longer know traffic type
- Refactor VIF driver choosing code and add unit tests
- Basic unit tests, just test default case
- Also slight refactor of unit test code, and use jUnit 4 instead of 3, to match rest of codebase
Signed-off-by: Hugo Trippaers <htrippaers@schubergphilis.com>
Some concepts included:
* the replace.properties location used by maven is parameterized to allow
for a build that does not modify the currently git tracked files
* package naming is updated along the lines of what was discussed on the
-dev mailing list and between committers at the Build a Cloud Day in Belgi
* package version pattern is updated (since we redo all package names,
we might as well drop the epoch)
The recently added overcommit feature breaks compatibility between older management servers
and 4.2 agents.
This patch fixes that by falling back if needed.
Detail: Removing references to /usr/lib/cloud and /usr/lib64/cloud so that old
systemvm.iso files aren't found by accident. systemvm.iso should exist in
/usr/share/cloudstack-common/vms now.
Signed-off-by: Marcus Sorensen <marcus@betterservers.com> 1360860243 -0700
Detail: If your traffic label points to a bridge that is on a tagged interface
rather than a real physical interface, cloudstack may not parse the physical
interface correctly, bringing up tagged interfaces on the tagged interface.
Signed-off-by: Marcus Sorensen <marcus@betterservers.com> 1360798665 -0700
We used to define domains persistent in libvirt, which caused XML definitions
to stay there after a reboot of the hypervisor.
We however don't do anything with those already defined domains, actually, we wipe
all defined domains when starting the agent.
Some users however reported that libvirt started these domains after a reboot
before the CloudStack agent was started.
By starting domains from the XML description and not defining them we prevent
them from ever being stored in libvirt.
- Fixed new join dao impls as spring components
- Fixed component context xml to load api rate limit checker
- Fixed root pom.xml for duplicate plugin
- Fixed list data centers method
- Fixed following conflicts:
api/src/org/apache/cloudstack/api/command/admin/network/CreateNetworkOfferingCmd.java
api/src/org/apache/cloudstack/api/command/user/offering/ListServiceOfferingsCmd.java
api/src/org/apache/cloudstack/api/command/user/template/DeleteTemplateCmd.java
api/src/org/apache/cloudstack/api/command/user/template/ExtractTemplateCmd.java
plugins/api/discovery/src/org/apache/cloudstack/discovery/ApiDiscoveryServiceImpl.java
server/src/com/cloud/api/ApiDBUtils.java
server/src/com/cloud/api/ApiServer.java
server/src/com/cloud/api/query/QueryManagerImpl.java
server/src/com/cloud/configuration/DefaultComponentLibrary.java
server/src/com/cloud/server/ManagementServerImpl.java
server/src/com/cloud/storage/swift/SwiftManagerImpl.java
Signed-off-by: Rohit Yadav <bhaisaab@apache.org>
Detail: There are several places in the code that do a
"brctl show | grep bridgeName" or similar, which causes all sorts
of problems when you have for example a cloudVirBr50 and a
cloudVirBr5000. This patch attempts to stop relying on the output
of brctl, instead favoring sysfs and /sys/devices/virtual/net.
It cuts a lot of bash out altogether by using java File. It was
tested in my devcloud-kvm against current 4.0, as well as by the
customer reporting initial bug.
BUG-ID: CLOUDSTACK-938
Fix-For: 4.0.1
Signed-off-by: Marcus Sorensen <marcus@betterservers.com>
Description: When selecting a storage adaptor, cleanupDisk assumes that
libvirt is being used. Therefore, we pass a StoragePoolType that maps to
libvirt. This is the only place in LibvirtComputingResource where the
StoragePoolType can't be pulled from somewhere else.
BUG-ID: CLOUDSTACK-1011
Signed-off-by: Marcus Sorensen <marcus@betterservers.com>
Detail: This merges the resizevolume feature branch, which provides the
ability to migrate a disk between disk offerings, thereby changing its
size, or specifying a new size if current disk offering is custom.
BUG-ID: CLOUDSTACK-644
Signed-off-by: Marcus Sorensen <marcus@betterservers.com> 1358358209 -0700
Although I still think the templates aren't well maintained, I just
added 12.04 since this is an LTS and people probably want it in the
list of templates.
This system should be more generic I think though.
Create OvsVifDriver to deal with openvswitch specifics for plugging
intefaces
Create a parameter to set the bridge type to use in
LibvirtComputingResource.
Create several functions to get bridge information from openvswitch
Add a check to detect the libvirt version and throw an exception when
the version is to low ( < 0.9.11 )
Fix classpath loading in Script.findScript to deal with missing path
separators at the end.
Add notification to the BridgeVifDriver that lswitch broadcast type is
not supported.
Create OvsVifDriver to deal with openvswitch specifics for plugging
intefaces
Create a parameter to set the bridge type to use in
LibvirtComputingResource.
Create several functions to get bridge information from openvswitch
Add a check to detect the libvirt version and throw an exception when
the version is to low ( < 0.9.11 )
Fix classpath loading in Script.findScript to deal with missing path
separators at the end.
Add notification to the BridgeVifDriver that lswitch broadcast type is
not supported.
Modify the spec file to package the agent files and the scripts
Some changes to the poms to put the java dependencies in the right place.
Move the agent script to the dedicated os dir in packaging.
Detail: Instead of using LibvirtStorageAdaptor for everything, you can create
your own storage adaptor and use it. We select storage adaptor based on storage
pool type, thus we needed to adjust LibvirtComputingResource to pass pool type
to everything in KVMStoragePoolManager. This in turn required that we pass the
info necessary to LibvirtComputingResource as well, so a few agent Commands were
modified.
Note this patch in and of itself shouldn't change any existing behavior, just
allow for new storage adaptors to be selected based on storage pool type.
Reviewed-by: Edison Su
Signed-off-by: Marcus Sorensen <marcus@betterservers.com> 1355769696 -0700
Detail: If source image is qcow2, and we want a qcow2 image, then doing a
convert strips off compression and any snapshots the user had in that image. If
a backing file exists, we stick with convert so we can pull in both the backing
file and the COW image, otherwise we just cp the qcow2 file. This is also faster
Signed-off-by: Marcus Sorensen <marcus@betterservers.com> 1354755241 -0700
Detail: This patch deletes any patchdisk found when deleting root volume for
system VM.
BUG-ID: CLOUDSTACK-566
Bugfix-for: 4.0.1
Signed-off-by: Marcus Sorensen <marcus@betterservers.com> 1354222335 -0700
stopped
Detail: This patch fixed an issue with hosts trying to stop system vms that were
already not running and deleting a patch disk for the system vm running on
another host. It got applied to 4.0 but not master.
Signed-off-by: Marcus Sorensen <marcus@betterservers.com> 1354222160 -0700
Detail: Because of the way most other primary storage types work with cloudstack
(i.e. backing stores) CLVM actually copies the template to a local logical
volume on primary storage, then uses that. This causes all of your primary
storage to be littered with a copy of every template used. Since we're not
using these, dump the template direct to the newly created logical volume.
This is faster as well since the template is sparse; we're not creating a fat
template on primary storage and then copying that to a logical volume when we
deploy from template.
BUG-ID: CLOUDSTACK-508
Bugfix-for: 4.1
Signed-off-by: Marcus Sorensen <marcus@betterservers.com> 1353221260 -0700
Detail: In com.cloud.hypervisor.kvm.resource.BridgeVifDriver.java, in 2 places
an if block should have evaluated to true if trafficLabel was null, however it
was causing a NullPointerException instead.
BUG-ID : NONE
Bugfix-for: 4.0
Reviewed-by: Marcus Sorensen
Reported-by: Dave Cahill
Signed-off-by: Marcus Sorensen <marcus@betterservers.com> 1352307750 -0700
Detail: There was a regression in functionality introduced by
915babd970 where the public
bridge could not also be the private bridge. This had several
additional consequences, this patch should revert the behavior
back while keeping the functionality enhancements introduced by that
commit.
BUG-ID : NONE
Reviewed-by: Dave Cahill
Reported-by: Dave Cahill via cloudstack-dev
Signed-off-by: Marcus Sorensen <shadowsor@gmail.com> 1351574006 -0600
called.
VifDriver.unplug must be called in MigrateCommand which hooks VM
migration in source host, because plug will be called in
PrepareForMigration in destination host. But that operation is missing
in current LibvirtComputingResources.
Signed-off-by: Edison Su <sudison@gmail.com>
On kvm computing host, vifdriver.unplug will always fails (throws
LibvirtException) and network cleanup will not be called. This was
because the code first undefine the computing domain, and then tries to
query the destroyed machine definition to fetch NIC information. IMHO,
kvm plugin code rounds LibvirtException too much.
Signed-off-by: Edison Su <sudison@gmail.com>
Since only the cephx user like 'admin' was passed we couldn't define two RBD storage pools
using the cephx user admin, even if they were running on different Ceph clusters.
By adding the monitor hostname and poolname to the secret's usage (which we don't even use) it becomes
unique.
work)
Cloudstack seems to let you create guest traffic types on multiple
physical networks. However, when I try this with KVM I end up always
bridging to whatever device is used for guest.network.device. This pulls
the traffic label (NicTO.getName()) and uses that bridge to ensure that
we get on the correct physical network, rather than just always using
the guest.network.device.
This also changes the bridge naming scheme from cloudVirBr + vlanid to
br + physicalinterface + "-" + vlanid. This is because we should be able
to support the same vlan numbers per physical network, and the previous
bridge name would not support this and collide.
Signed-off-by: Edison Su <sudison@gmail.com>
create
The code is unable to detect an existing pool, because we use a random
UUID each time. New Libvirt doesn't allow multiple pools to be defined
to the same storage. This patch generates a UUID based on the storage
path, so that it can be detected as existing and reused. It also cleans
up no-op code and adjusts the naming of a few things to clean up any
confusion.
Signed-off-by: Edison Su <sudison@gmail.com>
Since /root is r-x permissions, Java fails to mkdir /root/.ssh (even
though the agent is running as root) because it looks for the writable
permission. This patch modifies the 'chmod 700 /root/.ssh' shell command
that we already use into 'mkdir -m 700 /root/.ssh', to be able to create
the directory as root even though write permissions are not set on
/root. This seemed cleaner/safer than adding writable to /root.
Signed-off-by: Edison Su <sudison@gmail.com>
The default value for local.storage.path does not exists by
default in CentOS 6. By default, this results in NullPointerException
silently. Without this log message, administrator can't figure out
the reason at all.
Signed-off-by: Edison Su <sudison@gmail.com>
/root/.ssh is created with perms '600' if it doesn't already exist. This causes
a problem in that it can't write out id_rsa.cloud:
2012-08-27 16:35:40,227 DEBUG [cloud.agent.Agent] (agentRequest-Handler-4:null)
Processing command: com.cloud.agent.api.ModifySshKeysCommand
2012-08-27 16:35:40,228 DEBUG [kvm.resource.LibvirtComputingResource]
(agentRequest-Handler-4:null) Failed to create file: java.io.IOException:
Permission denied
Doing 'chmod u+x /root/.ssh' fixed the above, so it seems that even though the
agent is running as root it cares about being able to chdir into /root.ssh
Signed-off-by: Sheng Yang <sheng.yang@citrix.com>
Implements
SetupGuestNetworkCommand,SetNetworkACLCommand,SetSourceNatCommand,IpAssocVpcCommand,SetPortForwardingRulesVpcCommand.
Passes basic functionality, though I'm sure there may be some honing to
do.
Also fixes a few minor things found along the way:
vpc_guestnw.sh wasn't successfully setting up apache due to default
listen IP of 10.1.1.1
vpc_guestnw.sh was referencing a 'logger_it' function, replaced with
'logger -t cloud'
system vms were running with OS type "Debian GNU/Linux 5.0(32-bit)",
which was not found in the KVMGuestOsMapper
the Xen implementation of SetupGuestNetworkCommand had apparently
copied its catch message from UnPlug Nic, fixed string
Send-by: Marcus Sorensen
RB: https://reviews.apache.org/r/6883
This is part 1 in enabling VPC for KVM. The various commands needing
implementation will be submitted individually unless I'm told to do
otherwise, in case I don't complete all of the commands, such that
someone else can take over or build on my work.
RB: https://reviews.apache.org/r/6859
Send-by: shadowsor@gmail.com
Add BridgeVifDriver and move current vif implementation to it.
- remove dependency on VirtualRoutingResource.
- factor out some of the networking code in LibvirtComputingResource
to BridgeVifDriver.
Add base class for KVM VifDriver.
Add VifDriver Interface for KVM.
RB: https://reviews.apache.org/r/6285
Send-by: Tomoe Sugihara <tomoe@midokura.com>
Add BridgeVifDriver and move current vif implementation to it.
- remove dependency on VirtualRoutingResource.
- factor out some of the networking code in LibvirtComputingResource
to BridgeVifDriver.
Add base class for KVM VifDriver.
Add VifDriver Interface for KVM.
RB: https://reviews.apache.org/r/6285
Send-by: Tomoe Sugihara <tomoe@midokura.com>
We used to generate a UUID when this wasn't set, but since we aren't writing to
agent.properties anymore we have to make sure the UUID is persistent across restarts.
Libvirt can also return a bunch of emulators for eg ARM and S390
We filter those out since we do not support these architectures.
This way we don't try to start a x86_64 instance with a S390 emulator
Since we are using libvirt for handling our storage pools we should rely on that information as well.
Before fetching the capacity we refresh the pool so libvirt has the most up-to-date information.
This is not needed with newly created pools since libvirt does a refresh on creation.
JijunLiu
Wei Zhou
Edison Su
Marcus Sorensen
Min Chen
Wido den Hollander
Wido den Hollander
Chip Childers
Toshiaki Hatano
Alex Huang
Jayapal
Kishan Kavala
Vijayendra Bhamidipati
Prasanna Santhanam
Mike Tutkowski
Laszlo Hornyak
Hugo Trippaers
Hiroaki KAWAI
Fang Wang
Edison Su
Sheng Yang
Jayapal Reddy
Nitin Mehta
Alex Huang
Dave Cahill
Phong Nguyen
Bharat Kumar
Noa Resare
Rohit Yadav
Kelven Yang
Hugo Trippaers
anthony
marcus
Kelven Yang
Marcus Sorensen
Edison Su
Jie Feng
Manikanta Kattamuri
olivier lamy
Darren Shepherd
Darren Shepherd
Mice Xia
Alena Prokharchyk
Chip Childers