Hi,
I've added support to Prometheus to use Twitter serversets for service
discovery, and we're now got this deployed in production. A key point is
that you can configure it to look at a tree of ZNodes in Zookeeper, so that
it can automatically pick up any new jobs announced by the Aurora executor
with no additional configuration required.
The scrape config we have in ansible for this is:
- job_name: "aurora_job"
serverset_sd_configs: - servers:{% for host in
groups["zookeeper"] %} - "{{ host }}:2181"{% endfor %}
paths: - "/aurora/serversets" relabel_configs: -
source_labels: [ __meta_serverset_path ] regex:
'^/aurora/serversets/[^/]+/[^/]+/([^/]+)/.*' target_label: job
replacement: "${1}" - source_labels: [ __address__ ]
regex: '(ip-\d+-\d+-\d+-\d+)\..*:(\d+)' target_label:
__address__ replacement: "${1}:${2}"
The relabel of the addresses is as our mesos slaves have the externally
visible host names so they work nicely in the browser, but which don't
actually exist in DNS so need some munging.
Note that you'll need the Prometheus at head to use this, and that this
should also work for non-Aurora created serversets such as if you're using
Finagle.
Relevant docs:
http://aurora.apache.org/documentation/latest/configuration-reference/#announcer-objects
http://prometheus.io/docs/operating/configuration/#zookeeper-serverset-sd-configurations-serverset_sd_config
Brian
