Drop redundant `memory_bytes` metrics from ruby_sampler
The following discussion from !30374 (merged) should be addressed:
-
@oswaldo started a discussion: (+7 comments) It was here before, but would you know the reason we're setting the same RSS to two different Gauge metrics?
🤔 Having a look at our metrics catalog, it seems like we just make use of
process_resident_memory_bytes
. @andrewn would you have some background on that?
It appears that all metrics reporting /proc
readings in ruby_sampler.rb
report them under the ruby_process_
prefix. However, RSS is currently reported under two different metrics, ruby_process_resident_memory_bytes
and ruby_memory_bytes
, but they originate from the same system metric.
We should drop ruby_memory_bytes
and keep ruby_process_resident_memory_bytes
for consistency and reducing confusion.
Tasks:
-
Go through existing dashboards and alerts and ensure that ruby_memory_bytes
is unused or update them to useruby_process_*
instead -
Drop the ruby_memory_bytes
gauge
Places where the metric might be referenced:
-
https://gitlab.com/gitlab-com/runbooks for alerting rules, esp: rules/
thanos-rules/
dashboards/
-
https://gitlab.com/gitlab-org/omnibus-gitlab/ -
https://gitlab.com/gitlab-org/grafana-dashboards