#862 much faster version of aggregate_15_min_mvt #863

gabrielwol · 2024-02-07T22:50:31Z

What this pull request accomplishes:

Significant speed improvements for aggregate_15_min_mvt, especially for long time ranges.
this aggregation (intersection 68 from 2023-01-23 to 2023-06-29) took 6 hours today. I had similarly bad performance last week for other new intersections over long time ranges. It takes about 15s with the new version of the function.

07 Feb 2024 16:07:17 INFO Updated gapsize table and found gaps exceeding allowable size
07 Feb 2024 22:34:48 INFO Aggregated intersections [68] to 15 minute movement bins

Issue(s) this solves:

Miovision: Extremely slow aggregation for single intersection, many days #862

What, in particular, needs to reviewed:

we should still test the performance of the full function with the update part.

What needs to be done by a sysadmin after this PR is merged

replace function in postgres

… now

gabrielwol · 2024-02-08T13:51:24Z

Here is an example just for a single day (only testing select, not update):

ran in 8s:

"count"	"sum"	"source"
349851	3685031	"new"
349851	3685031	"old"

WITH temp AS (
    -- Cross product of dates, intersections, legal movement for cars, bikes, and peds to aggregate
    SELECT
        im.intersection_uid,
        dt.datetime_bin,
        im.classification_uid,
        im.leg,
        im.movement_uid,
        0 AS volume
    FROM miovision_api.intersection_movements AS im
    CROSS JOIN generate_series(
        '2024-02-07'::date,
        '2024-02-08'::date - interval '15 minutes',
        interval '15 minutes'
    ) AS dt(datetime_bin)
    WHERE
        --0 padding for certain modes (padding)
        im.classification_uid IN (1,2,6,10)
        --AND im.intersection_uid = ANY(target_intersections)
        
    UNION ALL
    
    --real volumes
    SELECT
        v.intersection_uid,
        datetime_bin_15(v.datetime_bin) AS datetime_bin,
        v.classification_uid,
        v.leg,
        v.movement_uid,
        SUM(volume)
    FROM miovision_api.volumes AS v
    --only aggregate common movements
    JOIN miovision_api.intersection_movements USING (
        intersection_uid, classification_uid, leg, movement_uid
    )
    WHERE
        v.datetime_bin >= '2024-02-07'::date
        AND v.datetime_bin < '2024-02-08'::date
        --AND v.intersection_uid = ANY(target_intersections)
    GROUP BY
        v.intersection_uid,
        datetime_bin_15(v.datetime_bin),
        v.classification_uid,
        v.leg,
        v.movement_uid
),

aggregate_insert AS (
    /*INSERT INTO miovision_api.volumes_15min_mvt(
        intersection_uid, datetime_bin, classification_uid, leg, movement_uid, volume
    )*/
    SELECT DISTINCT ON (
        v.intersection_uid, v.datetime_bin, v.classification_uid, v.leg, v.movement_uid
    )
        v.intersection_uid,
        v.datetime_bin,
        v.classification_uid,
        v.leg,
        v.movement_uid,
        CASE
            --set unacceptable gaps as nulls
            WHEN un.datetime_bin IS NOT NULL THEN NULL
            --gap fill with zeros (restricted to certain modes in temp CTE)
            ELSE v.volume
        END AS volume
    FROM temp AS v
    JOIN miovision_api.intersections AS i USING (intersection_uid)
    --set unacceptable gaps as null
    LEFT JOIN miovision_api.unacceptable_gaps AS un USING (
        intersection_uid, datetime_bin
    )
    WHERE
        -- Only include dates during which intersection is active 
        -- (excludes entire day it was added/removed)
        v.datetime_bin >= i.date_installed + interval '1 day'
        AND (
            i.date_decommissioned IS NULL
            OR (v.datetime_bin < i.date_decommissioned - interval '1 day')
        )
    ORDER BY 
        v.intersection_uid,
        v.datetime_bin,
        v.classification_uid,
        v.leg,
        v.movement_uid,
        --select real value instead of padding value if available
        v.volume DESC
)

SELECT COUNT(*), SUM(volume), 'new' AS source FROM aggregate_insert
UNION
SELECT COUNT(*), SUM(volume), 'old' AS source FROM miovision_api.volumes_15min_mvt WHERE datetime_bin >= '2024-02-07'::date AND datetime_bin < '2024-02-08'::date

And this select for 1 month, 2 intersections ran in 6s!!!

"count"	"sum"	"source"
349851	3685031	"new"
349851	3685031	"old"

WITH temp AS (
    -- Cross product of dates, intersections, legal movement for cars, bikes, and peds to aggregate
    SELECT
        im.intersection_uid,
        dt.datetime_bin,
        im.classification_uid,
        im.leg,
        im.movement_uid,
        0 AS volume
    FROM miovision_api.intersection_movements AS im
    CROSS JOIN generate_series(
        '2024-01-07'::date,
        '2024-02-08'::date - interval '15 minutes',
        interval '15 minutes'
    ) AS dt(datetime_bin)
    WHERE
        --0 padding for certain modes (padding)
        im.classification_uid IN (1,2,6,10)
        AND im.intersection_uid = ANY(ARRAY[12,25]::integer [])
        
    UNION ALL
    
    --real volumes
    SELECT
        v.intersection_uid,
        datetime_bin_15(v.datetime_bin) AS datetime_bin,
        v.classification_uid,
        v.leg,
        v.movement_uid,
        SUM(volume)
    FROM miovision_api.volumes AS v
    --only aggregate common movements
    JOIN miovision_api.intersection_movements USING (
        intersection_uid, classification_uid, leg, movement_uid
    )
    WHERE
        v.datetime_bin >= '2024-01-07'::date
        AND v.datetime_bin < '2024-02-08'::date
        AND v.intersection_uid = ANY(ARRAY[12,25]::integer [])
    GROUP BY
        v.intersection_uid,
        datetime_bin_15(v.datetime_bin),
        v.classification_uid,
        v.leg,
        v.movement_uid
),

aggregate_insert AS (
    /*INSERT INTO miovision_api.volumes_15min_mvt(
        intersection_uid, datetime_bin, classification_uid, leg, movement_uid, volume
    )*/
    SELECT DISTINCT ON (
        v.intersection_uid, v.datetime_bin, v.classification_uid, v.leg, v.movement_uid
    )
        v.intersection_uid,
        v.datetime_bin,
        v.classification_uid,
        v.leg,
        v.movement_uid,
        CASE
            --set unacceptable gaps as nulls
            WHEN un.datetime_bin IS NOT NULL THEN NULL
            --gap fill with zeros (restricted to certain modes in temp CTE)
            ELSE v.volume
        END AS volume
    FROM temp AS v
    JOIN miovision_api.intersections AS i USING (intersection_uid)
    --set unacceptable gaps as null
    LEFT JOIN miovision_api.unacceptable_gaps AS un USING (
        intersection_uid, datetime_bin
    )
    WHERE
        -- Only include dates during which intersection is active 
        -- (excludes entire day it was added/removed)
        v.datetime_bin >= i.date_installed + interval '1 day'
        AND (
            i.date_decommissioned IS NULL
            OR (v.datetime_bin < i.date_decommissioned - interval '1 day')
        )
    ORDER BY 
        v.intersection_uid,
        v.datetime_bin,
        v.classification_uid,
        v.leg,
        v.movement_uid,
        --select real value instead of padding value if available
        v.volume DESC
)

SELECT COUNT(*), SUM(volume), 'new' AS source FROM aggregate_insert
UNION
SELECT COUNT(*), SUM(volume), 'old' AS source FROM miovision_api.volumes_15min_mvt
WHERE datetime_bin >= '2024-01-07'::date AND datetime_bin < '2024-02-08'::date 
AND intersection_uid = ANY(ARRAY[12,25]::integer [])

chmnata · 2024-02-09T21:23:21Z

volumes/miovision/sql/function/function-aggregate-volumes_15min_mvt.sql

        )
-        --exclude movements already aggregated
-        AND v.volume_15min_mvt_uid IS NULL


is this excluded cause we have the clearing function?

yes exactly! it is redundant - we always run clear + aggregate together whether through run-api or airflow.

chmnata · 2024-02-09T21:52:55Z

I tested it on a day with unacceptable gaps and the result is different comparing to the data in 15_mvt. I am wondering if the following worked previously with the old function as intended cause I dont see Nulls for unacceptable gaps in the 15_mvt table. I tested with the date where v.datetime_bin >= '2024-01-02'::date AND v.datetime_bin < '2024-01-03'::date, can you investigate?

CASE
            --set unacceptable gaps as nulls
            WHEN un.datetime_bin IS NOT NULL THEN NULL
            --gap fill with zeros (restricted to certain modes in temp CTE)
            ELSE v.volume
        END AS volume
        ```

gabrielwol · 2024-02-12T14:50:03Z

Here's the test you ran last week @chmnata which gave suspicious results along with my investigation ✅
The exceptions were:

intersections 67, 68 which were only added to intersection_movements last week, and so were excluded from the initial run of volumes_15min_mvt. In my validation above I had purposely picked dates/intersections where this wouldn't apply, but I should have specified this.
intersection_uid = 60 AND datetime_bin = '2024-01-02 23:45:00' has null volumes in new results (correct, due to presence of anomalous_range) and zero in old results (incorrect). When I run the SELECT part of the old function, I correctly get nulls for this case, which indicates that unacceptable_gaps table has changed since the initial run. I'm not able to see the Airflow logs for this date - it's likely this date was also re-run via run-api while adding new intersections. I think since the new result is correct we are OK to merge this change.

--create a temp table to store and compare testing results
create table gwolofs.miovision_Test_null as
WITH temp AS (
    -- Cross product of dates, intersections, legal movement for cars, bikes, and peds to aggregate
    SELECT
        im.intersection_uid,
        dt.datetime_bin,
        im.classification_uid,
        im.leg,
        im.movement_uid,
        0 AS volume
    FROM miovision_api.intersection_movements AS im
    CROSS JOIN generate_series(
        '2024-01-02'::date,
        '2024-01-03'::date - interval '15 minutes',
        interval '15 minutes'
    ) AS dt(datetime_bin)
    WHERE
        --0 padding for certain modes (padding)
        im.classification_uid IN (1,2,6,10)
        --AND im.intersection_uid = ANY(target_intersections)
        
    UNION ALL
    
    --real volumes
    SELECT
        v.intersection_uid,
        datetime_bin_15(v.datetime_bin) AS datetime_bin,
        v.classification_uid,
        v.leg,
        v.movement_uid,
        SUM(volume)
    FROM miovision_api.volumes AS v
    --only aggregate common movements
    JOIN miovision_api.intersection_movements USING (
        intersection_uid, classification_uid, leg, movement_uid
    )
    WHERE
 v.datetime_bin >= '2024-01-02'::date
        AND v.datetime_bin < '2024-01-03'::date
        --AND v.intersection_uid = ANY(target_intersections)
    GROUP BY
        v.intersection_uid,
        datetime_bin_15(v.datetime_bin),
        v.classification_uid,
        v.leg,
        v.movement_uid
)


    /*INSERT INTO miovision_api.volumes_15min_mvt(
        intersection_uid, datetime_bin, classification_uid, leg, movement_uid, volume
    )*/
    SELECT DISTINCT ON (
        v.intersection_uid, v.datetime_bin, v.classification_uid, v.leg, v.movement_uid
    )
        v.intersection_uid,
        v.datetime_bin,
        v.classification_uid,
        v.leg,
        v.movement_uid,
        CASE
            --set unacceptable gaps as nulls
            WHEN un.datetime_bin IS NOT NULL THEN NULL
            --gap fill with zeros (restricted to certain modes in temp CTE)
            ELSE v.volume
        END AS volume
    FROM temp AS v
    JOIN miovision_api.intersections AS i USING (intersection_uid)
    --set unacceptable gaps as null
    LEFT JOIN miovision_api.unacceptable_gaps AS un USING (
        intersection_uid, datetime_bin
    )
    WHERE
        -- Only include dates during which intersection is active 
        -- (excludes entire day it was added/removed)
        v.datetime_bin >= i.date_installed + interval '1 day'
        AND (
            i.date_decommissioned IS NULL
            OR (v.datetime_bin < i.date_decommissioned - interval '1 day')
        )
    ORDER BY 
        v.intersection_uid,
        v.datetime_bin,
        v.classification_uid,
        v.leg,
        v.movement_uid,
        --select real value instead of padding value if available
        v.volume DESC NULLS LAST

GRANT ALL ON gwolofs.miovision_Test_null TO natalie, dbadmin;

--the 44 records that do not match are all for intersection_uid = 60 AND datetime_bin = '2024-01-02 23:45:00'.
--These records are zero in the old result and null in the new result. The record is in unacceptable_ranges, so it should be null. 
SELECT intersection_uid, datetime_bin, classification_uid, leg, movement_uid, volume
FROM miovision_api.volumes_15min_mvt as v
WHERE v.datetime_bin >= '2024-01-02'::date
AND v.datetime_bin < '2024-01-03'::date
EXCEPT
SELECT * from gwolofs.miovision_Test_null
WHERE intersection_uid NOT IN (67, 68) --these intersections were not aggregated in the old dag run from 01-02 because they were only added to intersection_movements later

--the 44 records that do not match are all for intersection_uid = 60 AND datetime_bin = '2024-01-02 23:45:00'.
--These records are null in the new result and zero in the old result. The record is in unacceptable_ranges, so it should be null. 
SELECT * from gwolofs.miovision_Test_null
WHERE intersection_uid NOT IN (67, 68) --these intersections were not aggregated in the old dag run from 01-02 because they were only added to intersection_movements later
EXCEPT
SELECT intersection_uid, datetime_bin, classification_uid, leg, movement_uid, 
volume 
FROM miovision_api.volumes_15min_mvt as v
WHERE v.datetime_bin >= '2024-01-02'::date AND v.datetime_bin < '2024-01-03'::date

--correctly get nulls for this case when re-aggregating using old function:
    SELECT
        im.intersection_uid,
        dt.datetime_bin,
        im.classification_uid,
        im.leg,
        im.movement_uid,
        CASE
            --set unacceptable gaps as nulls
            WHEN un.datetime_bin IS NOT NULL THEN NULL
            --gap fill with zeros (restricted to certain modes in having clause)
            ELSE (COALESCE(SUM(v.volume), 0))
        END AS volume
    -- Cross product of dates, intersections, legal movement for cars, bikes, and peds to aggregate
    FROM miovision_api.intersection_movements AS im
    CROSS JOIN generate_series(
        '2024-01-02 23:45:00'::timestamp,
        '2024-01-03'::timestamp - interval '15 minutes',
        interval '15 minutes'
    ) AS dt(datetime_bin)
    JOIN miovision_api.intersections AS mai USING (intersection_uid)     
    --To avoid aggregating unacceptable gaps
    LEFT JOIN miovision_api.unacceptable_gaps AS un ON
        un.intersection_uid = im.intersection_uid
        --remove the 15 minute bin containing any unacceptable gaps
        AND dt.datetime_bin = un.datetime_bin
    --To get 1min bins
    LEFT JOIN miovision_api.volumes AS v ON
        --help query choose correct partition
        v.datetime_bin = '2024-01-02 23:45:00'::timestamp
        AND v.datetime_bin >= dt.datetime_bin
        AND v.datetime_bin < dt.datetime_bin + interval '15 minutes'
        AND v.intersection_uid = im.intersection_uid
        AND v.classification_uid = im.classification_uid
        AND v.leg = im.leg
        AND v.movement_uid = im.movement_uid
    WHERE
        -- Only include dates during which intersection is active 
        -- (excludes entire day it was added/removed)
        dt.datetime_bin > mai.date_installed + interval '1 day'
        AND (
            mai.date_decommissioned IS NULL
            OR (dt.datetime_bin < mai.date_decommissioned - interval '1 day')
        )
        --exclude movements already aggregated
        AND v.volume_15min_mvt_uid IS NULL
        AND im.intersection_uid = ANY(ARRAY[46]::integer[])
    GROUP BY
        im.intersection_uid,
        dt.datetime_bin,
        im.classification_uid,
        im.leg,
        im.movement_uid, 
        un.datetime_bin
    HAVING
        --retain 0s for certain modes (padding)
        im.classification_uid IN (1,2,6,10)
        OR SUM(v.volume) > 0

chmnata · 2024-02-12T14:52:53Z

Here's the test you ran last week @chmnata which gave suspicious results along with my investigation ✅ The exceptions were:

intersections 67, 68 which were only added to intersection_movements last week, and so were excluded from the initial run of volumes_15min_mvt. In my validation above I had purposely picked dates/intersections where this wouldn't apply, but I should have specified this.
intersection_uid = 60 AND datetime_bin = '2024-01-02 23:45:00' has null volumes in new results (correct, due to presence of anomalous_range) and zero in old results (incorrect). When I run the SELECT part of the old function, I correctly get nulls for this case, which indicates that unacceptable_gaps table has changed since the initial run. I'm not able to see the Airflow logs for this date - it's likely this date was also re-run via run-api while adding new intersections. I think since the new result is correct we are OK to merge this change.

--create a temp table to store and compare testing results
create table gwolofs.miovision_Test_null as
WITH temp AS (
    -- Cross product of dates, intersections, legal movement for cars, bikes, and peds to aggregate
    SELECT
        im.intersection_uid,
        dt.datetime_bin,
        im.classification_uid,
        im.leg,
        im.movement_uid,
        0 AS volume
    FROM miovision_api.intersection_movements AS im
    CROSS JOIN generate_series(
        '2024-01-02'::date,
        '2024-01-03'::date - interval '15 minutes',
        interval '15 minutes'
    ) AS dt(datetime_bin)
    WHERE
        --0 padding for certain modes (padding)
        im.classification_uid IN (1,2,6,10)
        --AND im.intersection_uid = ANY(target_intersections)
        
    UNION ALL
    
    --real volumes
    SELECT
        v.intersection_uid,
        datetime_bin_15(v.datetime_bin) AS datetime_bin,
        v.classification_uid,
        v.leg,
        v.movement_uid,
        SUM(volume)
    FROM miovision_api.volumes AS v
    --only aggregate common movements
    JOIN miovision_api.intersection_movements USING (
        intersection_uid, classification_uid, leg, movement_uid
    )
    WHERE
 v.datetime_bin >= '2024-01-02'::date
        AND v.datetime_bin < '2024-01-03'::date
        --AND v.intersection_uid = ANY(target_intersections)
    GROUP BY
        v.intersection_uid,
        datetime_bin_15(v.datetime_bin),
        v.classification_uid,
        v.leg,
        v.movement_uid
)


    /*INSERT INTO miovision_api.volumes_15min_mvt(
        intersection_uid, datetime_bin, classification_uid, leg, movement_uid, volume
    )*/
    SELECT DISTINCT ON (
        v.intersection_uid, v.datetime_bin, v.classification_uid, v.leg, v.movement_uid
    )
        v.intersection_uid,
        v.datetime_bin,
        v.classification_uid,
        v.leg,
        v.movement_uid,
        CASE
            --set unacceptable gaps as nulls
            WHEN un.datetime_bin IS NOT NULL THEN NULL
            --gap fill with zeros (restricted to certain modes in temp CTE)
            ELSE v.volume
        END AS volume
    FROM temp AS v
    JOIN miovision_api.intersections AS i USING (intersection_uid)
    --set unacceptable gaps as null
    LEFT JOIN miovision_api.unacceptable_gaps AS un USING (
        intersection_uid, datetime_bin
    )
    WHERE
        -- Only include dates during which intersection is active 
        -- (excludes entire day it was added/removed)
        v.datetime_bin >= i.date_installed + interval '1 day'
        AND (
            i.date_decommissioned IS NULL
            OR (v.datetime_bin < i.date_decommissioned - interval '1 day')
        )
    ORDER BY 
        v.intersection_uid,
        v.datetime_bin,
        v.classification_uid,
        v.leg,
        v.movement_uid,
        --select real value instead of padding value if available
        v.volume DESC NULLS LAST

GRANT ALL ON gwolofs.miovision_Test_null TO natalie, dbadmin;

--the 44 records that do not match are all for intersection_uid = 60 AND datetime_bin = '2024-01-02 23:45:00'.
--These records are zero in the old result and null in the new result. The record is in unacceptable_ranges, so it should be null. 
SELECT intersection_uid, datetime_bin, classification_uid, leg, movement_uid, volume
FROM miovision_api.volumes_15min_mvt as v
WHERE v.datetime_bin >= '2024-01-02'::date
AND v.datetime_bin < '2024-01-03'::date
EXCEPT
SELECT * from gwolofs.miovision_Test_null
WHERE intersection_uid NOT IN (67, 68) --these intersections were not aggregated in the old dag run from 01-02 because they were only added to intersection_movements later

--the 44 records that do not match are all for intersection_uid = 60 AND datetime_bin = '2024-01-02 23:45:00'.
--These records are null in the new result and zero in the old result. The record is in unacceptable_ranges, so it should be null. 
SELECT * from gwolofs.miovision_Test_null
WHERE intersection_uid NOT IN (67, 68) --these intersections were not aggregated in the old dag run from 01-02 because they were only added to intersection_movements later
EXCEPT
SELECT intersection_uid, datetime_bin, classification_uid, leg, movement_uid, 
volume 
FROM miovision_api.volumes_15min_mvt as v
WHERE v.datetime_bin >= '2024-01-02'::date AND v.datetime_bin < '2024-01-03'::date

--correctly get nulls for this case when re-aggregating using old function:
    SELECT
        im.intersection_uid,
        dt.datetime_bin,
        im.classification_uid,
        im.leg,
        im.movement_uid,
        CASE
            --set unacceptable gaps as nulls
            WHEN un.datetime_bin IS NOT NULL THEN NULL
            --gap fill with zeros (restricted to certain modes in having clause)
            ELSE (COALESCE(SUM(v.volume), 0))
        END AS volume
    -- Cross product of dates, intersections, legal movement for cars, bikes, and peds to aggregate
    FROM miovision_api.intersection_movements AS im
    CROSS JOIN generate_series(
        '2024-01-02 23:45:00'::timestamp,
        '2024-01-03'::timestamp - interval '15 minutes',
        interval '15 minutes'
    ) AS dt(datetime_bin)
    JOIN miovision_api.intersections AS mai USING (intersection_uid)     
    --To avoid aggregating unacceptable gaps
    LEFT JOIN miovision_api.unacceptable_gaps AS un ON
        un.intersection_uid = im.intersection_uid
        --remove the 15 minute bin containing any unacceptable gaps
        AND dt.datetime_bin = un.datetime_bin
    --To get 1min bins
    LEFT JOIN miovision_api.volumes AS v ON
        --help query choose correct partition
        v.datetime_bin = '2024-01-02 23:45:00'::timestamp
        AND v.datetime_bin >= dt.datetime_bin
        AND v.datetime_bin < dt.datetime_bin + interval '15 minutes'
        AND v.intersection_uid = im.intersection_uid
        AND v.classification_uid = im.classification_uid
        AND v.leg = im.leg
        AND v.movement_uid = im.movement_uid
    WHERE
        -- Only include dates during which intersection is active 
        -- (excludes entire day it was added/removed)
        dt.datetime_bin > mai.date_installed + interval '1 day'
        AND (
            mai.date_decommissioned IS NULL
            OR (dt.datetime_bin < mai.date_decommissioned - interval '1 day')
        )
        --exclude movements already aggregated
        AND v.volume_15min_mvt_uid IS NULL
        AND im.intersection_uid = ANY(ARRAY[46]::integer[])
    GROUP BY
        im.intersection_uid,
        dt.datetime_bin,
        im.classification_uid,
        im.leg,
        im.movement_uid, 
        un.datetime_bin
    HAVING
        --retain 0s for certain modes (padding)
        im.classification_uid IN (1,2,6,10)
        OR SUM(v.volume) > 0

Thanks for the investigation! Do you know how much of the data currently in 15_mvt will need to be updated due to changed unacceptable gap?

gabrielwol · 2024-02-12T15:47:59Z

Here are 12 dates with records that should be nulls but are not. You were lucky to pick one of them at random! I will create a new issue for this investigation.

"datetime_bin"	"count"
"2024-02-05"	46
"2024-02-01"	43
"2024-01-20"	3990
"2024-01-19"	4032
"2024-01-18"	4032
"2024-01-17"	4032
"2024-01-16"	4032
"2024-01-15"	4032
"2024-01-14"	4032
"2024-01-13"	4032
"2024-01-02"	44
"2023-01-22"	42

SELECT datetime_bin::date, COUNT(*)
FROM miovision_api.volumes_15min_mvt AS v
JOIN miovision_api.unacceptable_gaps AS un USING (intersection_uid, datetime_bin)
WHERE v.volume IS NOT NULL
GROUP BY 1
ORDER BY 1 DESC

chmnata · 2024-02-12T15:54:53Z

This new year is giving me the luck we needed!!!!! 🧧
Probs some changes needed on volumes_daily as well eh?

gabrielwol · 2024-02-13T14:25:51Z

Probs some changes needed on volumes_daily as well eh?

We'll fix all the downstream tables as well by running -agg in run-api, but I think we can do all that as part of the new PR I opened #869.

chmnata · 2024-02-13T14:36:30Z

updated in the database

gabrielwol · 2024-02-13T15:10:22Z

This aggregation today took 5 minutes instead of 6 hours like above! 🥳 only 70x faster in the end, not 3000.
--intersection 67 --intersection 68 --start_date 2023-06-29 --end_date 2024-02-07

#862 much faster version of aggregate_15_min_mvt

7484744

gabrielwol linked an issue Feb 7, 2024 that may be closed by this pull request

Miovision: Extremely slow aggregation for single intersection, many days #862

Closed

gabrielwol requested a review from chmnata February 7, 2024 22:50

gabrielwol self-assigned this Feb 7, 2024

gabrielwol added enhancement Miovision labels Feb 7, 2024

gabrielwol added 2 commits February 7, 2024 23:04

#862 fix small gt to gte bug

ea0d519

#862 remove already aggregated filter - we use clear function instead…

8089230

… now

gabrielwol marked this pull request as ready for review February 8, 2024 13:51

chmnata reviewed Feb 9, 2024

View reviewed changes

This was referenced Feb 12, 2024

Miovision: discrepencies between volumes_15min_mvt and unacceptable_gaps #868

Open

868 miovision find gaps bugs #869

Draft

chmnata approved these changes Feb 13, 2024

View reviewed changes

chmnata merged commit 085a020 into master Feb 13, 2024
5 checks passed

chmnata deleted the 862-miovision-volumes_15min_mvt_extremely-slow branch February 13, 2024 14:36

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

#862 much faster version of aggregate_15_min_mvt #863

#862 much faster version of aggregate_15_min_mvt #863

gabrielwol commented Feb 7, 2024 •

edited

Loading

gabrielwol commented Feb 8, 2024

chmnata Feb 9, 2024

gabrielwol Feb 9, 2024

chmnata commented Feb 9, 2024

gabrielwol commented Feb 12, 2024 •

edited

Loading

chmnata commented Feb 12, 2024

gabrielwol commented Feb 12, 2024

chmnata commented Feb 12, 2024

gabrielwol commented Feb 13, 2024

chmnata commented Feb 13, 2024

gabrielwol commented Feb 13, 2024

#862 much faster version of aggregate_15_min_mvt #863

#862 much faster version of aggregate_15_min_mvt #863

Conversation

gabrielwol commented Feb 7, 2024 • edited Loading

What this pull request accomplishes:

Issue(s) this solves:

What, in particular, needs to reviewed:

What needs to be done by a sysadmin after this PR is merged

gabrielwol commented Feb 8, 2024

chmnata Feb 9, 2024

Choose a reason for hiding this comment

gabrielwol Feb 9, 2024

Choose a reason for hiding this comment

chmnata commented Feb 9, 2024

gabrielwol commented Feb 12, 2024 • edited Loading

chmnata commented Feb 12, 2024

gabrielwol commented Feb 12, 2024

chmnata commented Feb 12, 2024

gabrielwol commented Feb 13, 2024

chmnata commented Feb 13, 2024

gabrielwol commented Feb 13, 2024

gabrielwol commented Feb 7, 2024 •

edited

Loading

gabrielwol commented Feb 12, 2024 •

edited

Loading