Freshness metrics #275
Freshness metrics #275
Conversation
KludgeKML
force-pushed
the
freshness-metrics
branch
from
7bad3f2 to
1d27155
Compare
| def get_oldest_postcode | ||
| # Cache metric to prevent needless expensive calls to the database | ||
| Rails.cache.fetch("metrics:oldest_postcode", expires_in: 1.hour) do | ||
| (Time.zone.now.to_i - Postcode.os_places.order(updated_at: :asc).first.updated_at.to_i) |
There was a problem hiding this comment.
Using the updated_at timestamp seems a little flakey as any update to the record would influence the timestamp, have we considered adding a separate time field for OS places request?
There was a problem hiding this comment.
I think this is pretty safe because the only way these records get updated is via OS Places API calls (with one exception: bulk CSV import, which is also a reasonable consideration of freshness and is unlikely to happen again anyway).
There was a problem hiding this comment.
For now yes, agreed. However, if the code changes in the future and the records are updated outside of OS places requests then this is metric may no longer be meaningful. I think relying on the standard timestamp values for harvesting data will probably trip someone up in the future. Don't think I'd classify this as a blocker as we're only at risk of breaking a metric but one we'd probably want to consider more if we plan to alert on it 👍
KludgeKML
force-pushed
the
freshness-metrics
branch
from
1d27155 to
b06fe4d
Compare
Adds a freshness metric for OS Places postcodes. Every hour we update the oldest postcode sourced from OS Places APi and allow prometheus to collect its age rounded down to the day. We can then see a graph of the age in seconds of the least-recently updated postcode. Since we're expecting the entire database to be refreshed roughly once a week, we should see this hover around the 7 mark. If it increases significantly above that, we may be seeing problems.
We will evaluate this for a while before deciding if we need to add an alarm state to it.
https://trello.com/c/8P4LRqvl/2171-investigate-how-we-can-handle-os-places-api-being-down