no user_metadata added in network samples data

[Hang Liu]

For other meter samples such as cpu_util or disk.*, "user_metadata" is included in sample data

{
    "counter_name": "disk.read.bytes",
    "counter_type": "cumulative",
    "counter_unit": "B",

     ... ...

    "resource_metadata": {
        "OS-EXT-AZ:availability_zone": "nova",
        "disk_gb": 20,
        "display_name": "te-ServerGroup-4j577oohiuli-gpbzcdrp7ymx-dd2k2dtivusr",
        "ephemeral_gb": 0,

         ... ...

        "root_gb": 20,
        "status": "active",
        "user_metadata": {
            "stack": "bcb16239-c7ae-4ccc-b686-5f2a35878e28"
        },
        "vcpus": 1
    },
    "source": "openstack",
    "timestamp": "2014-09-15T07:56:11Z",
    "user_id": "311ef8f58df7478fb1cf680f86c51e37"
}

while for network.* meters, "user_metadata" is missing

{
    "counter_name": "network.incoming.bytes.rate",
    "counter_type": "gauge",
    "counter_unit": "B/s",

      ... ...

    "resource_metadata": {
        "fref": null,
        "instance_id": "b0da23a1-e0a8-495b-bf14-2629654162a5",
        "instance_type": "2",
        "mac": "fa:16:3e:f1:ee:10",
        "name": "tap6cb98968-ab",
        "parameters": {}
    },
    "source": "openstack",
    "timestamp": "2014-09-15T07:56:11Z",
    "user_id": "311ef8f58df7478fb1cf680f86c51e37"
}

In the heat template, "use_metadata" is using to match the sample to the stack ID. If it is missing,
alarm would not be raised.

[OpenStack Infra (hudson-openstack)]

Fix proposed to branch: master
Review: https://review.openstack.org/123646

[OpenStack Infra (hudson-openstack)]

Reviewed: https://review.openstack.org/123646
Committed: https://git.openstack.org/cgit/openstack/ceilometer/commit/?id=cca6fcd89e0044f310f6b05c62665302582e14e1
Submitter: Jenkins
Branch: master

commit cca6fcd89e0044f310f6b05c62665302582e14e1
Author: Hang Liu <email address hidden>
Date: Wed Sep 24 01:59:53 2014 -0500

    Add user_metadata to network samples

    Previously 'user_metadata' is not included in ceilometer network samples,
    while it is in other samples. This causes network samples not triggering
    alarms heat created.

    Change-Id: Iafb4c42955fceaaf20fef1f9580ddb1331a0c724
    Closes-Bug: 1369500

[OpenStack Infra (hudson-openstack)]

Fix proposed to branch: stable/juno
Review: https://review.openstack.org/137359

[OpenStack Infra (hudson-openstack)]

Reviewed: https://review.openstack.org/137359
Committed: https://git.openstack.org/cgit/openstack/ceilometer/commit/?id=48a2957c5c06d1fd43c86486345cb0784f5a700b
Submitter: Jenkins
Branch: stable/juno

commit 48a2957c5c06d1fd43c86486345cb0784f5a700b
Author: Hang Liu <email address hidden>
Date: Wed Sep 24 01:59:53 2014 -0500

    Add user_metadata to network samples

    Previously 'user_metadata' is not included in ceilometer network samples,
    while it is in other samples. This causes network samples not triggering
    alarms heat created.

    Change-Id: Iafb4c42955fceaaf20fef1f9580ddb1331a0c724
    Closes-Bug: 1369500
    (cherry picked from commit cca6fcd89e0044f310f6b05c62665302582e14e1)

[Marcus (mga0)]

Sure you want a format like that? Thats a huuuuuuge resource_ID

foo@bar~# ceilometer sample-list -m 'network.outgoing.bytes.rate' | less
+-----------------------------------------------------------------------+-----------------------------+-------+-----------------+------+---------------------+
| Resource ID | Name | Type | Volume | Unit | Timestamp |
+-----------------------------------------------------------------------+-----------------------------+-------+-----------------+------+---------------------+
| instance-00000074-1360f8a6-94a7-4bd7-8667-deca6e33ecad-tapb81a824b-d3 | network.outgoing.bytes.rate | gauge | 182.262295082 | B/s | 2014-11-28T07:21:55 |
| instance-00000073-6a10f983-834f-4175-906c-dbc6d99642a8-tap2aeb5e1c-d1 | network.outgoing.bytes.rate | gauge | 1541.67213115 | B/s | 2014-11-28T07:21:55 |

[Hang Liu (hangliu)]

The resource_ID is formatted as "instance_name - instance.id - vnic.name". The network samples are associated with
a specific port of a VM instance. So the vnic.name should be there. But instance_name might be not necessary if only for a unique
resource ID. I also don't know why it is included, it looks this way long before this bug.

[Eoghan Glynn (eglynn)]

Marcus - the resource ID is orthogonal to the current bug, but I agree that it seems needlessly lenthy. The instance UUID provides sufficient global uniqueness in itself to identify the VM unambiguously, without also prepending the libvirt domain ID (which is only unique per-host)

[Marcus (mga0)]

Thats also my opinion.
The problem was, that I did querying on the CLR for debugging reasons and didn't get any results for
ceilometer sample-list -m 'network.outgoing.bytes.rate' --query resources_id=<InstanceID> while the same command with -m 'cpu_util' works fine.
Would be great to have a 'corporate design' for every meter (like the normal instance id). Otherwise we need an extra screen especially for displaying the resource_id :)

[ZhiQiang Fan (aji-zqfan)]

@Marcus

There is a patch which fix the long resource-id issue for NIC releated sample, it is still reviewing in progress: https://review.openstack.org/132148

[Hang Liu (hangliu)]

Hi, Eoghan Glynn, ZhiQiang

Can network meters also report combining data of vnics, just like disk meters do?

For per vnic port data, use deeper meters to query, refer to this change

https://review.openstack.org/#/c/132040

Maybe those meters looks like these,

 - network.port.incoming.bytes
 - network.port.incoming.packets
 - network.port.outgoing.bytes
 - network.port.outgoing.packets

Then UUID can be used for keys of both the combined and per port network meter samples.
What do you think about it?

[ZhiQiang Fan (aji-zqfan)]

@Hang Liu

Agree, I'm trying to work on this