disk device stats via api call / cmdline

[RolandK]

hi,

i searched for quite a while how to get to more detailed VM stats via api/cmdline and was lead to multiple wrong paths/dead ends.

eventually, i found that default text-based output of "pvesh get" is severely crippled, but you get to all of the vm performance/statistics data when you use json based output instead. it looks this isn't documented anywhere!? i would add an RFE to proxmox bugzilla then.

here is an example to get the bytes read and written to/from scsi0 disk:

# pvesh get nodes/pve-node6/qemu/223/status/current --output-format=json | jq -jr '"scsi0.rd_bytes="
,.blockstat.scsi0.rd_bytes ,"," , "scsi0.wr_bytes=",.blockstat.scsi0.wr_bytes,"\n"'

scsi0.rd_bytes:190018602496,scsi0.wr_bytes:93583524352

# pvesh get nodes/pve-node6/qemu/223/status/current --output-format=text

┌─────────────────┬────────────────────┐
│ key             │ value              │
╞═════════════════╪════════════════════╡
│ agent           │ 1                  │
├─────────────────┼────────────────────┤
│ cpus            │ 4                  │
├─────────────────┼────────────────────┤
│ ha              │ {"managed":0}      │
├─────────────────┼────────────────────┤
│ lock            │ backup             │
├─────────────────┼────────────────────┤
│ maxdisk         │ 40.00 GiB          │
├─────────────────┼────────────────────┤
│ maxmem          │ 8.00 GiB           │
├─────────────────┼────────────────────┤
│ name            │ gitlab │
├─────────────────┼────────────────────┤
│ pid             │ 27590              │
├─────────────────┼────────────────────┤
│ qmpstatus       │ running            │
├─────────────────┼────────────────────┤
│ running-machine │ pc-i440fx-5.2+pve0 │
├─────────────────┼────────────────────┤
│ running-qemu    │ 5.2.0              │
├─────────────────┼────────────────────┤
│ status          │ running            │
├─────────────────┼────────────────────┤
│ uptime          │ 17w 1d 12h 51m     │
├─────────────────┼────────────────────┤
│ vmid            │ 223                │
└─────────────────┴────────────────────┘

# pvesh get nodes/pve-node6/qemu/223/status/current --output-format=json|jq "."|head -n30

{
  "agent": 1,
  "balloon": 8589934592,
  "ballooninfo": {
    "actual": 8589934592,
    "free_mem": 572932096,
    "last_update": 1677352844,
    "major_page_faults": 3596724,
    "max_mem": 8589934592,
    "mem_swapped_in": 457097216,
    "mem_swapped_out": 882458624,
    "minor_page_faults": 23990786471,
    "total_mem": 8200802304
  },
  "blockstat": {
    "ide2": {
      "account_failed": false,
      "account_invalid": false,
      "failed_flush_operations": 0,
      "failed_rd_operations": 0,
      "failed_unmap_operations": 0,
      "failed_wr_operations": 0,
      "flush_operations": 0,
      "flush_total_time_ns": 0,
      "idle_time_ns": 10414516232180416,
      "invalid_flush_operations": 0,
      "invalid_rd_operations": 0,
      "invalid_unmap_operations": 0,
      "invalid_wr_operations": 0,
      "rd_bytes": 152,

 

[bbgeek17]

I would just put it in the "lesson learned" box. Text output is for humans, json - for automated tools. Always use json or some other structured output for processing.

As for documentation - "man pvesh" :

FORMAT_OPTIONS
       It is possible to specify the output format using the --output-format parameter. The default format text uses ASCII-art to draw nice borders around tables. It additionally transforms some values into human-readable
       text, for example:

Pretty ascii-art tables can fit so much data before they stop being pretty.

---

Blockbridge : Ultra low latency all-NVME shared storage for Proxmox - https://www.blockbridge.com/proxmox

 

[RolandK]

yes, but following the unix paradigm, it's not obvious that output modifiers which switch appearance of the output also switch the contents of the output. that should at least be documented.

 

[RolandK]

https://bugzilla.proxmox.com/show_bug.cgi?id=4554