DevCloud is a XCP Kronos based xen. For this we use the XcpOssResource
whose memory limits were not set explicitly. Instead the base
CitrixResourceBase would set the limits. static_min was set to 128MB
failing the cpvm and ssvm start fails whose offerings have 100MB set to
the max limits.
Signed-off-by: Prasanna Santhanam <tsp@apache.org>
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
platform
Use the Gson adapters to serialize/deserialize the NatRules
Switch the NiciraNvpApi to a single gson Object with the proper adapters
Fix missing order setting for static nat rules and portforwarding rules
Return an error when a port range is passed in a portforwarding rule
The serializer is not required
Fix a bug where an ip address could be released even if it was still in
use for SourceNat
Throw a json parse exception when the type is unknown to the adapter
Each rule created in VNMC under a policy object needs to have an unique order value. Rules are evaluated based on this value.
Eariler order was computed based on the rule count under a policy object. This resulted in duplicate order value when rules get
deleted and recreated. Changed the logic to compute order based on the CS db id of the rule which is unique.
When dmc is enabled allow cloudstack to scale the VM between dynamic
ranges. This requires a further commit where there is enough memory
available between dynamic-max and static-max for the VM to scale under
memory pressure. See the TODO in Xenserver56FP1Resource.java
Signed-off-by: Prasanna Santhanam <tsp@apache.org>
23e54bb0 introduced multiple hypervisors support for cpu and memory
overcommit. Here the HypervisorGuru base which determines the min, max
range for the memory for all hypervisors computes the minCpu using the
MemoryOverCommit ratio and minMemory using the CpuOverCommit ratio.
Minor typo/logic issue but massive damage across all HV if enabled ;)
Signed-off-by: Prasanna Santhanam <tsp@apache.org>
When setting memory constraints on Xen guests we should honor:
static-min <= dynamic-min <= dynamic-max <= static-max
Our VmSpec while allows the guests to like between dynamic-min and
dynamic-max the memory set by the resource set the static min to be
equal to the dynamic max. This restricts the hypervisor from ensuring
optimized memory handling of guests.
see: http://wiki.xen.org/wiki/XCP_FAQ_Dynamic_Memory_Control#How_does_XCP_choose_targets_for_guests_in_dynamic_range_mode.3F
Another fix was related the restrict_dmc option. when enabled (true)
this option disallows scaling a vm. The logic was reverse handled
allowing scaling when restrict_dmc was on. This control flow is now
fixed.
Signed-off-by: Prasanna Santhanam <tsp@apache.org>
Missing default constructor fails the agent manager reloading the XCP
resource on reboot of management server. This is fixed by using the
default constructor as do other Xen resources and include a new resource
ala XenServers for XCP1.6.
Signed-off-by: Prasanna Santhanam <tsp@apache.org>
1. A new implicit planner which extends the functionality provided by FirstFitPlanner.
2. Implicit planner can be used in either strict or preferred mode. In strict mode it tries to deploy a vm of a given account on a host on which vms of the account are already running. If no such host is found it'll search for an empty host to service the request. Otherwise the deploy vm request fails.
3. In preferred mode, if a host which is running vms of the account or an empty host isn't found, the planner then tries to deploy on any other host provided it isn't running implicitly dedicated strict vms of any other account.
4. Updated the createServiceOffering api to configure the details for the planner that the service offering is using.
5. Made db changes to store the service offering details for the planner.
6. Unit tests for testing the implicit planner functionality.
7. Marvin test for validating the functionality.
- Changes merged from planner_reserve branch
- Exposing deploymentplanner as an optional parameter while creating a service offering
- changes to DeploymentPlanningManagerImpl to make sure host reserve-release happens between conflicting planner usages.
No longer creating firewall rule as part of PF/Static NAT rule creation. Now firewall rule needs to be configured separately.
Also made some changes to exception handling.
* Updated SQL upgrade scripts to include midonet configs.
* Fixed bug where default ICMP allow rule was missing
on static NAT creation, keeping VMs from being able
to ping the gateway.
* Changed the filter in the MidoNetElement callbacks to allow
calls when Midonet is configured.
Signed-off-by: Hugo Trippaers <htrippaers@schubergphilis.com>
