cpu_util meter should always be non-negative

[Kurt]

In VMWare environment, when a VM is shutdown, vsphere will return -1 when ceilometer query for performance, and as a result the cpu_util is set to -0.01%, like below:

[root@cc vmware]# ceilometer sample-list -m cpu_util -l 1
+--------------------------------------+----------+-------+-----------------+------+---------------------+
| Resource ID | Name | Type | Volume | Unit | Timestamp |
+--------------------------------------+----------+-------+-----------------+------+---------------------+
| f8596226-94b7-4a8b-9569-52672bbf5865 | cpu_util | gauge | -0.01 | % | 2014-12-19T10:19:37 |
+--------------------------------------+----------+-------+-----------------+------+---------------------+

Expected results:

There are two ways to handle this:
1, the inspector should know "-1" is an error code rather than an actual number, and shouldn't generate a new sample.
2, cpu_util volume should be set to "0" if it is negative.

And what's more, the same policy should be applied to memory usage, disk IO, network IO, etc.

[ZhiQiang Fan (aji-zqfan)]

or we can raise instance shut off exception when instance is shut off....

[Kurt (dingyuan-rao)]

Agree with Zhiqiang

[Kurt (dingyuan-rao)]

The fix should handle the following three cases:

1, when instance is shutdown (or any other *stopped-like* status), all performance query will return -1. In this case, an exception should be raised indicating the VM status is invalid for monitoring. No samples will be generated at all in this case.

2, when instance is active, but one disk is broken or one NIC is error, some samples in the performance query will return -1 and others are OK. In that case, as the disk io is calculated as a summary of all disk IOs, the -1s should be replaced by 0s, and others remain unchanged. Samples are still generated in this case, and the -1s are replaced by 0s.

3, when instance is active, but cpu_util or memory usage somehow returns -1. In this case, an exception should be raised indicating the volume is invalid. The specific sample (cpu_util or memory usage) will not be generated, other correct samples will be generated.

[OpenStack Infra (hudson-openstack)]

Fix proposed to branch: master
Review: https://review.openstack.org/144192

[Haifeng.Yan (yanheven)]

I think we should reuse InstanceShutOffException & NoDataException.so no more work to do with codes in dir pollster/

[Luo Gangyi (luogangyi)]

Hi guys, I have another question.

Should we guarantee the cpu_util value smaller than 100%？

In current implementation, cpu_util value sometime are large than 100%, does such value reasonable?

[gordon chung (chungg)]

@Luo Gangyi, that issue can be tracked here: https://bugs.launchpad.net/ceilometer/+bug/1421584

[Luo Gangyi (luogangyi)]

 Thanks gordon，

Still a bit difference from my issue. But the issue you pointed is very useful.

[gordon chung (chungg)]

this can be done by adding support for growth_only like delta transformer

[OpenStack Infra (hudson-openstack)]

Change abandoned by gordon chung (<email address hidden>) on branch: master
Review: https://review.openstack.org/144192
Reason: clean up