Filling Out & Filtering Irregular Time Series Data - sql

Question

Using Postgresql 9.4, I am trying to craft a query on time series log data that logs new values whenever the value updates (not on a schedule). The log can update anywhere from several times a minute to once a day.
I need the query to accomplish the following:
Filter too much data by just selecting the first entry for the timestamp range
Fill in sparse data by using the last reading for the log value. For example, if I am grouping the data by hour and there was an entry at 8am with a log value of 10. Then the next entry isn't until 11am with a log value of 15, I would want the query to return something like this:
Timestamp | Value
2015-07-01 08:00 | 10
2015-07-01 09:00 | 10
2015-07-01 10:00 | 10
2015-07-01 11:00 | 15
I have got a query that accomplishes the first of these goals:
with time_range as (
select hour
from generate_series('2015-07-01 00:00'::timestamp, '2015-07-02 00:00'::timestamp, '1 hour') as hour
),
ranked_logs as (
select
date_trunc('hour', time_stamp) as log_hour,
log_val,
rank() over (partition by date_trunc('hour', time_stamp) order by time_stamp asc)
from time_series
)
select
time_range.hour,
ranked_logs.log_val
from time_range
left outer join ranked_logs on ranked_logs.log_hour = time_range.hour and ranked_logs.rank = 1;
But I can't figure out how to fill in the nulls where there is no value. I tried using the lag() feature of Postgresql's Window functions, but it didn't work when there were multiple nulls in a row.
Here's a SQLFiddle that demonstrates the issue:
http://sqlfiddle.com/#!15/f4d13/5/0

Answer 1

your columns are log_hour and first_vlue
with time_range as (
select hour
from generate_series('2015-07-01 00:00'::timestamp, '2015-07-02 00:00'::timestamp, '1 hour') as hour
),
ranked_logs as (
select
date_trunc('hour', time_stamp) as log_hour,
log_val,
rank() over (partition by date_trunc('hour', time_stamp) order by time_stamp asc)
from time_series
),
base as (
select
time_range.hour lh,
ranked_logs.log_val
from time_range
left outer join ranked_logs on ranked_logs.log_hour = time_range.hour and ranked_logs.rank = 1)
SELECT
log_hour, log_val, value_partition, first_value(log_val) over (partition by value_partition order by log_hour)
FROM (
SELECT
date_trunc('hour', base.lh) as log_hour,
log_val,
sum(case when log_val is null then 0 else 1 end) over (order by base.lh) as value_partition
FROM base) as q
UPDATE
this is what your query return
Timestamp | Value
2015-07-01 01:00 | 10
2015-07-01 02:00 | null
2015-07-01 03:00 | null
2015-07-01 04:00 | 15
2015-07-01 05:00 | nul
2015-07-01 06:00 | 19
2015-07-01 08:00 | 13
I want this result set to be split in groups like this
2015-07-01 01:00 | 10
2015-07-01 02:00 | null
2015-07-01 03:00 | null
2015-07-01 04:00 | 15
2015-07-01 05:00 | nul
2015-07-01 06:00 | 19
2015-07-01 08:00 | 13
and to assign to every row in a group the value of first row from that group (done by last select)
In this case, a method for obtaining the grouping is to create a column which holds the number of
not null values counted until current row and split by this value. (use of sum(case))
value | sum(case)
| 10 | 1 |
| null | 1 |
| null | 1 |
| 15 | 2 | <-- new not null, increment
| nul | 2 |
| 19 | 3 | <-- new not null, increment
| 13 | 4 | <-- new not null, increment
and now I can partion by sum(case)