Finding local maximum between zero-values using SQL - mysql

Question

I have a table with minute-by-minute data from an IOT device. Every minute there is a new row with a timestamp and a value that represents a metric. The metric starts at 0 and increments for a while before it resets and starts over.
When I plot it, it looks like the picture. I want to find the local maximum value of each run, as the blue circles indicate.
Is it possible to find and group the consecutive rows where the metric is > 0 and then find the maximum of each group?
Update
Table structure:
+-------------+------------------+
| Field | Type |
+-------------+------------------+
| id | int(10) unsigned |
| timestamp | timestamp |
| metric_name | varchar(32) |
| value | int(10) |
+-------------+------------------+

Answer 1

This is based on the following assumptions:
Id is a perfectly sequential integer (with no gaps)
You want to get the value logged directly before the 0 value
Code:
SELECT *
FROM metrics m1
WHERE m.id IN (
SELECT m2.id - 1
FROM metrics m2
WHERE m1.value = 0)

Answer 2

I join everything that isnt zero before a timestamp where it is zero, then I find the ones with no values inbetween that 0 and the last one..
SELECT
value,
timestamp
FROM
metrics
LEFT JOIN metrics zeros
on metrics.time < zeros.time
and zeros.value = 0
LEFT JOIN metrics betweenZero
on metrics.time < betweenZero.time
and betweenZero.time < zeros.time
INNER JOIN metrics noBetweens
on table.id = noBetweens.id
and betweenZero.id IS NULL
If you need it for a paritulcar metric_name, WHERE metric_name = the_metric_nameon the end.

Answer 3

This should give you the max value per group along with start time and end time of each window with only 1 pass over the data.
select metric_name, max(value) value, max(start_group) start_time, max(end_group) end_time from(
select metric_name, value,
case when #prev_ts is not null then #prev_ts end prev_ts,
case when value = 0 then #ts := timestamp end as start_group,
#ts as grouping,
#prev_ts := timestamp end_group
from metric join (select #prev_ts := null as p) prev
order by timestamp
) q
group by metric_name, grouping;
This will create a sample data set of 1000 rows, that resets every minute.
insert into metric(timestamp, metric_name, value)
select now() - interval rn second, 'pressure', v
from(
select #rn := #rn + 1 rn, mod(1000 - #rn,60) * pow(1000 - mod(#rn,121),1) v
from table_with_at_least_1000_rows
join (select #rn := 0) rn
limit 1000
) q
;

Answer 4

Try this:
SELECT
T.min_id
,T.max_id
,MAX(M.value) as local_max
FROM
metrics M
JOIN (
SELECT
id as min_id
,(
SELECT MIN(id) FROM Metrics MI
WHERE
MI.id > MO.id
AND MI.value = 0) as max_id
FROM Metrics MO
WHERE
value = 0
)T ON M.id BETWEEN T.min_id AND T.max_id
GROUP BY
T.min_id, T.max_id

Answer 5

My solution doesn't care about gaps but I am assuming that the sequence of ids is monotonic, that is they increase along the series by time. (You could probably substitute id for timestamp in the query even.) I had made a few minor syntax-type errors that I have since corrected since my first attempt and I have tested it with a simple Fiddle. I think it works.
select t0.*
from
T t0 inner join
(
select max_z, max(id) as max_id, max(value) as local_max
from
(
select
id, value,
(
select max(t2.id) as max_id from T t2
where t2.id < t.id and t2.value = 0
) as max_z
from T t
where t.value <> 0
) p /* partitions */
group by p.max_z
) x /* extrema */
on t0.id between max_z and max_id and t0.value = x.local_max
Btw it returns all the rows when there's a tie for the local maximum.
http://sqlfiddle.com/#!9/de832/2