Kubelet's cAdvisor metrics endpoint does not reliably return all metrics

Question

I am having an issue with cAdvisor where not all metrics are being reliably returned when I query its metrics endpoint. Specifically, querying container_fs_limit_bytes{device=~"^/dev/.*$",id="/",kubernetes_io_hostname=~"^.*"} through Prometheus often displays results for only a fraction of the nodes in my Kubernetes cluster. This happens when the corresponding metrics are not scraped for over 5mins (due to the metrics becoming stale), but I'm not sure why all metrics are not being displayed every time the endpoint is queried successfully.

Curling the endpoint over and over again shows that some metrics are only returned at particular times and so the above Prometheus query will return data for all nodes only if it happens to scrape them once within the last 5mins, but more often than not this is not the case.

One workaround is to take the average of the metric over a longer period than 5mins, but this is not ideal.

kubectl version:

Client Version: version.Info{Major:"1", Minor:"7", GitVersion:"v1.7.4", GitCommit:"793658f2d7ca7f064d2bdf606519f9fe1229c381", GitTreeState:"clean", BuildDate:"2017-08-17T08:48:23Z", GoVersion:"go1.8.3", Compiler:"gc", Platform:"darwin/amd64"}
Server Version: version.Info{Major:"1", Minor:"7", GitVersion:"v1.7.3+coreos.0", GitCommit:"42de91f04e456f7625941a6c4aaedaa69708be1b", GitTreeState:"clean", BuildDate:"2017-08-07T19:44:31Z", GoVersion:"go1.8.3", Compiler:"gc", Platform:"linux/amd64"}

Prometheus version: 1.7.1

Prometheus configuration:

global:
  scrape_interval: 15s
  scrape_timeout: 10s
  evaluation_interval: 1m
alerting:
  alertmanagers:
  - static_configs:
    - targets:
      - alertmanager:9093
    scheme: http
    timeout: 10s
rule_files:
- /etc/prometheus-rules/alert.rules
scrape_configs:
- job_name: kubernetes-nodes
  scrape_interval: 15s
  scrape_timeout: 10s
  metrics_path: /metrics
  scheme: https
  kubernetes_sd_configs:
  - api_server: null
    role: node
    namespaces:
      names: []
  bearer_token_file: /var/run/secrets/kubernetes.io/serviceaccount/token
  tls_config:
    ca_file: /var/run/secrets/kubernetes.io/serviceaccount/ca.crt
    insecure_skip_verify: false
  relabel_configs:
  - source_labels: []
    separator: ;
    regex: __meta_kubernetes_node_label_(.+)
    replacement: $1
    action: labelmap
  - source_labels: []
    separator: ;
    regex: (.*)
    target_label: __address__
    replacement: kubernetes.default.svc:443
    action: replace
  - source_labels: [__meta_kubernetes_node_name]
    separator: ;
    regex: (.+)
    target_label: __metrics_path__
    replacement: /api/v1/nodes/${1}:4194/proxy/metrics
    action: replace
  metric_relabel_configs:
  - source_labels: [id]
    separator: ;
    regex: ^/machine\.slice/machine-rkt\\x2d([^\\]+)\\.+/([^/]+)\.service$
    target_label: rkt_container_name
    replacement: ${2}-${1}
    action: replace
  - source_labels: [id]
    separator: ;
    regex: ^/system\.slice/(.+)\.service$
    target_label: systemd_service_name
    replacement: ${1}
    action: replace

Answer 1

This is a known bug in how cAdvisor is using the Prometheus client libraries.