Nightly build shows no CPU usage


(Rene) #1

I used a nightly build of topbeat from here (ARM architecture) on my Raspberry Pi. It nicely shows the memory usage, however, no CPU usage is being shown. I already played around with toggleing cpu_per_core between true and false, but no luck.
I noticed however that the nightly build is 5.0.0. The other topbeats I'm running on other machines are all version 1.2.0. Is that where the problem is? The template loaded into ES is also of version 1.2.0. I'm cautious to load a 5.0.0 template into ES, worrying it would break the 1.2.0 versions.
Can I make this work by loading a new template into ES? Is it safe? Or should I compile a 1.2.0 topbeat for ARM? (or maybe there's a nightly of the 1.2.x branch?)


(Andrew Kroh) #2

I tried 5.0.0-nightlyc4050f5 on my rpi and it's reporting everything (cpu, cpu/core, mem, and process data).

Here's the config I used to test:

topbeat:
  period: 10
  procs: [".*"]
  stats:
    system: true
    process: true
    filesystem: true
    cpu_per_core: true
output:
  file:
    path: "."
    filename: topbeat.json
  console:
    pretty: true

Are there any errors in your logs (both Topbeat and Elasticsearch)? The 5.x template is not compatible with ES 1.X or 2.X because it uses the new "keyword" type so don't try to load it. But as far as I'm aware there are no breaking changes with regard to the data model that will affect Topbeat 5.x data being sent to ES 2.x.


(Rene) #3

I already see why the graph is not working. Topbeat 5.0.0 is not sending field proc.cpu.user_p which I use for the CPU graph. It is however sending proc.cpu.user, but that's total CPU time. I see a new field proc.cpu.total_p, but I'm scared to refresh the field list in kibana, worried it would break the fields that are working now...


(Andrew Kroh) #4

The rename of proc.cpu.user_p to proc.cpu.total_p is mentioned in the release notes for 5.0.0-alpha1 as a breaking change. The "user_p" field was actual calculated from the system time and user time so it made sense to call it "total_p".


(Rene) #5

Yeah I noticed, I missed this at first. Thanks for the head up.


(system) #6