CPU usage polling via API

[tsumaru720]

Hi Guys

I'm looking at making some scripts that give me stats of my containers but I'm having difficulty getting CPU usage for it via the API.

When looking at a container's status via the API, I suspect the "cpu" value should represent the percentage of CPU its using against its allocation?

This container is at 20% usage (looking at its summary in the web UI), but the value is 0.

pve:/> get nodes/pmx01/lxc/100/status/current
200 OK
{
   "cpu" : 0,
   "cpus" : "1",
   "disk" : 24936472576,
   "diskread" : "911796413440",
   "diskwrite" : "2439126016",
   "ha" : {
      "managed" : 0
   },
   "lock" : "",
   "maxdisk" : 52710469632,
   "maxmem" : 2147483648,
   "maxswap" : 2147483648,
   "mem" : 1572667392,
   "name" : "syslog.example.com",
   "netin" : 3827656894,
   "netout" : 2730394259,
   "pid" : "2980",
   "status" : "running",
   "swap" : 626429952,
   "template" : "",
   "type" : "lxc",
   "uptime" : 257654
}

The same seems to apply to QEMU/KVM too - I'd loaded this VM at 100% but "cpu" still 0

pve:/> get nodes/pmx01/qemu/104/status/current
200 OK
{
   "balloon" : 536870912,
   "ballooninfo" : {
      "actual" : 536870912,
      "max_mem" : 536870912
   },
   "blockstat" : {
      "ide2" : {
         "account_failed" : false,
         "account_invalid" : false,
         "failed_flush_operations" : 0,
         "failed_rd_operations" : 0,
         "failed_wr_operations" : 0,
         "flush_operations" : 0,
         "flush_total_time_ns" : 0,
         "idle_time_ns" : 2319738740,
         "invalid_flush_operations" : 0,
         "invalid_rd_operations" : 0,
         "invalid_wr_operations" : 0,
         "rd_bytes" : 14274774,
         "rd_merged" : 0,
         "rd_operations" : 793042,
         "rd_total_time_ns" : 24046696658,
         "timed_stats" : [],
         "wr_bytes" : 0,
         "wr_highest_offset" : 0,
         "wr_merged" : 0,
         "wr_operations" : 0,
         "wr_total_time_ns" : 0
      },
      "virtio0" : {
         "account_failed" : true,
         "account_invalid" : true,
         "failed_flush_operations" : 0,
         "failed_rd_operations" : 0,
         "failed_wr_operations" : 0,
         "flush_operations" : 0,
         "flush_total_time_ns" : 0,
         "idle_time_ns" : 185813278619,
         "invalid_flush_operations" : 0,
         "invalid_rd_operations" : 0,
         "invalid_wr_operations" : 0,
         "rd_bytes" : 196230144,
         "rd_merged" : 4,
         "rd_operations" : 19872,
         "rd_total_time_ns" : 65452421340,
         "timed_stats" : [],
         "wr_bytes" : 1512230912,
         "wr_highest_offset" : 1860009472,
         "wr_merged" : 436,
         "wr_operations" : 76615,
         "wr_total_time_ns" : 1770928176815
      }
   },
   "cpu" : 0,
   "cpus" : 1,
   "disk" : 0,
   "diskread" : 210504918,
   "diskwrite" : 1512230912,
   "ha" : {
      "managed" : 0
   },
   "maxdisk" : 2147483648,
   "maxmem" : 536870912,
   "mem" : 266082277,
   "name" : "vpn1.example.com",
   "netin" : 2130369187,
   "netout" : 1473923150,
   "nics" : {
      "tap104i0" : {
         "netin" : "1148498060",
         "netout" : "31456760"
      },
      "tap104i1" : {
         "netin" : "981871127",
         "netout" : "1442466390"
      }
   },
   "pid" : "8787",
   "qmpstatus" : "running",
   "status" : "running",
   "template" : "",
   "uptime" : 2370124
}

Am I looking in the right place?

 

[wolfgang]

Hi,

what you like to have is here

pvesh get /nodes/ella/lxc/100/rrddata -timeframe 'hour' 

pvesh get /nodes/ella/qemu/104/rrddata -timeframe 'hour'

see
http://pve.proxmox.com/pve-docs/api-viewer/index.html

 

[tsumaru720]

Hi @wolfgang

Thanks, but I was hoping more for somthing I could poll to get usage at the time of polling. That command seems to give me a lot of un-nessecery output and seems to be delayed by 1 minute. Should the command I listed above not output CPU usage at the time of polling?

It seems like it should, but isnt - So is it a bug?

 

[dcsapak]

Hi @wolfgang

Thanks, but I was hoping more for somthing I could poll to get usage at the time of polling. That command seems to give me a lot of un-nessecery output and seems to be delayed by 1 minute. Should the command I listed above not output CPU usage at the time of polling?

It seems like it should, but isnt - So is it a bug?

this is a limitation of how we collect the information.

the linux kernel measures the cpu time since the start, so a momentary information is not really possible, only when you have at least 2 points in time you can
calculate the difference (which makes sense, because at any point in time only one process is active at one core, so you have to calculate over a time period to get cpu usage in %)

since you call pvesh it will only have one point in time -> no sensible cpu usage information

as an alternative you can query via curl and the api -> the pvedaemon collects the information on a regular basis, so it has more datapoints

 

[Logan_Rocks]

Is there any updates for this? i was looking some something similar.

 

[oguz]

Is there any updates for this? i was looking some something similar.

i think not much has changed with regards to this.

you can still use an external software for monitoring purposes though (zabbix or similar)

 

[freebee]

Hi.
The answer is cpu * 100.
Best regards.