I am running a query where I look for a record, and another record a certain amount of time later.
Table definition:
(
id integer primary key,
gpsstatus character(2),
datetime timestamp without time zone,
lat numeric(9,6),
lon numeric(9,6),
alt numeric(9,4),
time integer,
datafileid integer,
shape geometry,
speed double precision,
dist double precision,
shape_utm geometry,
lokalitet character(128),
cowid integer
)
There are indexes on datetime,lokalitet,cowid,gpsstatus, gist-index on shape and shape_utm.
The points should be sampled every 5th seconds, so I tried to do
select <something more>,p1.timestamp
from table p1, table p2
where p1.timestamp + interval '5 secound' = p2.timestamp
That ran reasonably quick, but then I found out that I had lost quite a few points due to jitter in the sampling, so that the points may be from 4 to 6 seconds apart.
Then I tried:
where (p2.timestamp, interval'0 second')
overlaps (p1.timestamp + interval '4 second', interval '2 second')
and that took for ages. I also tried the simpler solution:
WHERE p1.timestamp + interval '4 second' <= p2.timestamp
AND p1.timestamp + interval '6 second' >= p2.timestamp
which also ended up being unusably slow.
The timestamp field has a normal index. Is there a special kind of index are something else that would make this query usable?
The query at the moment:
SELECT
p1.cowid,
p1.datetime,
st_distance(p1.shape_utm, lead(p1.shape_utm)
OVER (ORDER BY p1.datetime)) AS meters_obs,
st_distance(p1.shape_utm, lead(p1.shape_utm, 720)
OVER (ORDER BY p1.datetime)) AS meters_hour,
observation.observation
FROM (gpspoint p1 LEFT JOIN observation
ON (observation.gpspointid = p1.id)),
status
WHERE p1.gpsstatus = status.id
AND status.use = true;
I may also get an an acceptable query time by asking for some specific intervals.
If you just want the previous record, you can do:
If you need a record
4to6seconds before the current, use this:This may return multiple previous records if they all are in the range.