BigQuery - removing duplicate records sometimes ta

We implemented following ETL process in Cloud: run a query in our local database hourly => save the result as csv and load it into the cloud storage => load the file from cloud storage into BigQuery table => remove duplicate records using the following query.

SELECT 
  * EXCEPT (row_number)
FROM (
  SELECT 
    *,
    ROW_NUMBER() OVER (PARTITION BY id ORDER BY timestamp DESC) row_number 
  FROM rawData.stock_movement
)
WHERE row_number = 1

Since 8 am (local time in Berlin) this morning the process of removing duplicate records takes much longer than it usual does, even the amount of data is not much different than it usual is: it takes usually 10s to remove duplicate records whereas this morning sometimes half an hour.

Is it the performance to remove duplicate record not stable?

It could be that you have many duplicate values for a particular id, so computing row numbers takes a long time. If you want to check for whether this is the case, you can try:

#standardSQL
SELECT id, COUNT(*) AS id_count
FROM rawData.stock_movement
GROUP BY id
ORDER BY id_count DESC LIMIT 5;

With that said, it may be faster to remove duplicates with this query instead:

#standardSQL
SELECT latest_row.*
FROM (
  SELECT ARRAY_AGG(t ORDER BY timestamp DESC LIMIT 1)[OFFSET(0)] AS latest_row
  FROM rawData.stock_movement AS t
  GROUP BY t.id
);

Here is an example:

#standardSQL
WITH T AS (
  SELECT 1 AS id, 'foo' AS x, TIMESTAMP '2017-04-01' AS timestamp UNION ALL
  SELECT 2, 'bar', TIMESTAMP '2017-04-02' UNION ALL
  SELECT 1, 'baz', TIMESTAMP '2017-04-03')
SELECT latest_row.*
FROM (
  SELECT ARRAY_AGG(t ORDER BY timestamp DESC LIMIT 1)[OFFSET(0)] AS latest_row
  FROM rawData.stock_movement AS t
  GROUP BY t.id
);

The reason that this may be faster is that BigQuery will only keep the row with the largest timestamp in memory at any given point in time.