avoid direct calls to ceph API for volume resize triggers

[balpert89]

Summary

We have seen continuously increasing memory usage for the libvirt provider caused by left open connections to the ceph API (see #444).

The idea is to avoid direct calls to ceph API from libvirt-provider entirely. This provides some advantages, prominently a cleaner implementation and less external dependencies.

This issue is meant to pool ideas how to accomplish this. Current ideas for approaches:

• using QEMU as an event trigger for resizes: QEMU uses an implementation for ceph volumes anyway, it may be possible to leverage this.
• extending the machine runtime interface (IRI): extending the machine-iri protocol to get updates for volumes may serve as a source for resize triggers but has a bigger impact for current implementations.

Please feel free to add other ideas and to discuss the implications, problems, advantages for each approach.

[lukas016]

libvirt is using hypervisor for getting block device information.
Currently qemu doesn't support capacity and it has report current size: https://github.com/qemu/qemu/blob/7872e5fdf38ac0d8d0083aabb98d67da1f530ef4/block/rbd.c#L100
This size is set during connection and resize.

All QMP query-block... commands returned just current size but i wasn't able get full capacity of rbd image.
I couldn't find any suitable command for our purpose: https://qemu-project.gitlab.io/qemu/interop/qemu-qmp-ref.html#qapidoc-304

    {
      "io-status": "ok",
      "device": "",
      "locked": false,
      "removable": false,
      "inserted": {
        "iops_rd": 0,
        "detect_zeroes": "off",
        "image": {
          "virtual-size": 71940702208,
          "filename": "json:{\"driver\": \"raw\", \"file\": {\"pool\": \"ceph\", \"image\": \"...\'}}",
          "cluster-size": 4194304,
          "format": "raw",
          "dirty-flag": false
        },
        ...

Expected size is 75161927680 bytes

[lukas016]

RBD image has to be resized first with libvirt command. After resize of rbd image OS inside VM can automatically detect change of volume size.

[ 1194.769976] virtio_blk virtio1: [vda] new size: 140509184 512-byte logical blocks (71.9 GB/67.0 GiB)
[ 1194.771600] vda: detected capacity change from 136314880 to 140509184
Nov 27 11:56:21 localhost kernel: virtio_blk virtio1: [vda] new size: 140509184 512-byte logical blocks (71.9 GB/67.0 GiB)
Nov 27 11:56:21 localhost kernel: vda: detected capacity change from 136314880 to 140509184