Braindump some knowledge about that:

• In the past, I've been able to look for global requests made to the maps server through the webrequest_sampled_128 at https://w.wiki/4N7w
• You can also take a look at our live stats in grafana https://grafana.wikimedia.org/d/000000305/maps-performances?orgId=1

Typical superset-presto-hive query:

select
  count(*) as sample_size,
  count(api_error_codes) as error_count,
--  histogram(coalesce(api_error_codes)) as errors_by_code,
--  count_if(
--    element_at(http.request_headers, 'user-agent') = 'kartotherian (yurik @ wikimedia)'
--  ) as is_kartotherian
--  count_if(is_wmf_domain) as is_wmf_domain,
--  histogram(database) as database
--  histogram(element_at(params, 'titles')) as page_title,
--  approx_percentile(backend_time_ms, 0.50) as backend_time_ms_p50,
--  approx_percentile(backend_time_ms, 0.70) as backend_time_ms_p70,
--  approx_percentile(backend_time_ms, 0.95) as backend_time_ms_p95
-- TODO: params.mpdgroups underscore vs wikivoyage group names
from event.mediawiki_api_request
  where year=2021
  and month=11
  and day=1
--  and hour=1
  and element_at(params, 'action') = 'query'
  and element_at(params, 'prop') = 'mapdata'
--  and element_at(params, 'titles') = 'Verrazzano-Narrows Bridge'
;

Sample of distribution across wikis, from Nov 1, from 1-2am UTC (N=131997): P17738

This shows that roughly half of requests are for enwiki. 112 distinct wikis appear in the sample.

Docs for this version of Presto: https://prestodb.io/docs/0.246/

Trying to query the page titles but superset sqllab is crashing even on the simple and limited query:

with titles as (
select
  element_at(params, 'titles') as title
from event.mediawiki_api_request
  where year=2021
  and month=11
  and day=1
  and hour=1
  and element_at(params, 'action') = 'query'
  and element_at(params, 'prop') = 'mapdata'
  limit 100
)

SELECT 
  count(distinct title)
from titles;

Sample of distribution across wikis, from Nov 1, from 1-2am UTC (N=131997): P17738

@awight What is this showing? Is this already enough to point to wikis with frequent Kartographer views? For example ja.wiki seems to really stand out as a high number, though de wikivoyage is really low. So maybe I'm misinterpreting this or shouldn't be looking at this number at all? Sorry if these questions are premature.

In T295363#7500443, @ECohen_WMDE wrote:

Sample of distribution across wikis, from Nov 1, from 1-2am UTC (N=131997): P17738

@awight What is this showing? Is this already enough to point to wikis with frequent Kartographer views? For example ja.wiki seems to really stand out as a high number, though de wikivoyage is really low. So maybe I'm misinterpreting this or shouldn't be looking at this number at all? Sorry if these questions are premature.

I probably shouldn't have dumped the raw data here without more explanation! I would say there are some open questions about how we want to normalize the numbers, by pageview, by wiki article count, etc. This particular set of data is especially tricky because it's measuring backend API requests coming *from* Kartotherian, so it's only very loosely related to how many humans are requesting pages. For example, I looked at one particular page and there were 8 API requests, but roughly 800 pageviews in the same time span. The data so far only really shows us where maps are *not*, and gives us a surprising insight into how the caching works at the moment.

I'm slowly realizing that the only user-relevant information we can get from the API logs is to count unique titles where maps appeared. So we can say * certain wikis have no maps activity, and * for the wikis with activity, N titles of M titles viewed contained a map.

Remaining questions are moved to T295675.