The Personal Dashboard health metrics dashboard in Superset regularly times out and produces ghost queries which negatively affect Superset and Presto availability and performance (cf. T418542: Ghost queries overload Presto cluster).

One of such queries:

SELECT count(DISTINCT "user_id") AS "distinct visitors"
FROM
  (SELECT CAST(FROM_ISO8601_TIMESTAMP(meta.dt) AS DATE) AS dt,
          mediawiki.database AS wiki,
          agent.client_platform_family AS platform,
          performer.name AS user_id
   FROM event.product_metrics_web_base
   WHERE action = 'pageview'
     AND instrument_name = 'personal-dashboard-health-metrics'
     AND YEAR = 2026
     AND MONTH >= 2
     AND mediawiki.database IN ('enwiki',
                                'trwiki',
                                'thwiki',
                                'simplewiki',
                                'idwiki')
     AND meta.dt >= '2026-04-29T00:00:00'
     AND meta.dt <= '2026-06-11T14:35:48') AS "virtual_table"
WHERE "dt" >= DATE '2026-04-29'
  AND "dt" < DATE '2026-06-11'
LIMIT 50000

The product_metrics_web_base table is so massive now with all the instruments & experiments sending data to it and since it's not timestamp-partitioned (being a Hive table rather than an Iceberg table). This metric's calculation should probably be offloaded into a pipeline.

As an immediately remediation, @MNeisler has unpublished the dashboard and asked the stakeholders not to open it until it's been optimized.

Ideally we would have T417176: Enable Refine event tables to Iceberg but in the meantime let's siphon instrument_name = 'personal-dashboard-health-metrics' events into a timestamp-partitioned Iceberg table wmf_product.personal_dashboard_event_v1 like we're currently siphoning all experiment data out of web base table into wmf_experiments.experiment_event_v1 (which powers GrowthBook's analysis of TK experiments) (work was done in T414105)

A smaller subset of web_base that is timestamp-partitioned will be faster and less demanding to query in Superset. Once the queries powering the dashboard are modified to source from the new table, the dashboard can be re-published and stakeholders will be able to use it again. This solution will also buy time for T428987: Update Personal Dashboard Superset metric calculations to use dbt model.

Acceptance criteria

• An hourly updated wmf_product.personal_dashboard_v1 Iceberg table, partitioned on timestamp, limited to 90 days of data