I am testing a Timescale DB for a migration, and there is a query which is very slow depending on the time range, when using m.ts >= '2019-07-03 11:50:00+00' the query takes 1m30s (!), but with m.ts >= '2019-08-03 11:50:00+00' it takes about 250ms.

The query is this one, bear in mind that there are no rows for the requested sensor type 'ca1' so the whole table (hopefully index) has to be scanned.

SELECT ts, s.type, m.val
FROM measure m 
 JOIN node n on node_id = n.id 
 JOIN measure_sensor s on m.sensor_id = s.id 
WHERE 
(n.serial = 'TEST0000003') 
AND 
(s.type IN ('ca1'))
AND 
(m.ts <= '2025-04-01 12:51:02+00'::timestamptz )
AND (m.ts >= '2019-07-03 11:50:00+00'::timestamptz) 
ORDER BY m.ts desc 
LIMIT 4;

The Timescale chunk does not vary in this period.

The table measure has 128M rows and the following indexes (probably redundant, I am not the only one who worked on this, but I am the only one left...)