Promtail: Remove promtail_log_entries_bytes_bucket histogram

[slim-bean]

What this PR does / why we need it:

This histogram reports a distribution of the size (bytes) of every log line processed by promtail, and does so with a label for each file being tailed. There are 8 buckets in the histogram which can quickly lead to a huge amount of exported metrics.

In our experiences, we have found little use for this metric, not enough to justify the cost of the number of series it creates so this PR removes it.

Checklist

• Documentation added
• Tests updated
• Add an entry in the CHANGELOG.md about the changes.

[dannykopping]

Overall LGTM, but I wonder if we shouldn't follow a "first-deprecate-then-remove" approach - since some folks might be relying on this metric.
What do you think?

[kavirajk]

LGTM

[slim-bean]

Overall LGTM, but I wonder if we shouldn't follow a "first-deprecate-then-remove" approach - since some folks might be relying on this metric. What do you think?

This is a great question to ask, but in this case I'd prefer to just stop the bleeding. We've had user and customer complaints about the significant number of series this generates. Internally in our ops infra it's responsible for 750,000 active series, putting it at the 14th highest cardinality metric in our infra.

Combine this with myself having never actually used this histogram, and I don't know of any dashboards we have using it either. I think this is a pretty good example of something we could trim that would be a great gain for everyone with really minimal impact.

I am sorry for anyone using this and if that's you, please open an issue and let's see if there are alternatives. But in the meantime I'd like to have our next release just drop this metric and save everyone some active series.