PHP Most efficient way of checking if rows exist - php

Question

Consider the following table
+-------------+---------+------+-----+---------+----------------+
| Field | Type | Null | Key | Default | Extra |
+-------------+---------+------+-----+---------+----------------+
| id | int(11) | NO | PRI | NULL | auto_increment |
| date | date | NO | | NULL | |
| sku | varchar(10) | | | NULL |
| impressions | int(11) | NO | | NULL | |
| sales | int(11) | NO | | NULL | |
+-------------+---------+------+-----+---------+----------------+
The table gets populated daily from a bulk download of the previous days sales records.
Each days download not only contains the previous days sales data but also all data from the last 90 days (possible 50k+ records).
However the data for previous days may change since the original insert due to matters outside our control, e.g.
Day 1.
Date: 2015-01-01
SKU: ABCD
Impressions: 100
Sales: 0
Day 2.
Date: 2015-01-01
SKU: ABCD
Impressions: 100
Sales: 3
Date: 2015-01-02
SKU: ABCD
Impressions: 105
Sales: 0
So for any given record from the data download it could be
a) Already seen and the same as before - ignore
b) New - add to database
c) Already seen but new data - Update
Arguably this could be trivially solved by checking each row as so
while (!$file->eof()) {
$row = $file->fgets();
$data = explode("\t", $row);
$sku = $data[0];
$date = $data[1];
$impressions = $data[2];
$sales = $data[3];
$order = $em->getRepository('Orders')->findOneBy(['sku' => $sku, 'date' => $date]);
if($order && $order->getImpressions() != $impressions && $order->getSales() != $sales) {
$order->setImpressions($impressions);
$order->setSales($sales);
} else {
... create new model
}
$em->persist($order);
}
However the rows which will have updated data will be minimal and doing a select for each and every row would mean this job would be incredibly slow due to sheer number of rows.
So my question is what patterns could be used to solve this problem as efficiently as possible?
Any ideas welcome

Answer 1

I would suggest you completely replace the previous 90 days' data with the newly downloaded data.
The reasoning is simple:
The processing time to do this will be trivial. 50,000 rows is tiny in database terms. I would probably do this even if it were a million rows.
Trying to replace only the changed rows is complicated and could introduce errors.

Answer 2

When you say "same as before" it seems like the keys are date and sku (combined) and sales and impressions are the fields that could be updated. If that's correct, then the most efficient way to do this in MySQL is to use INSERT ... ON DUPLICATE KEY UPDATE ... query:
Create a unique key on date and sku columns.
In your php script pre-parse all data from file (or do it in batches if you'd like).
Run a query similar to this (substitute actual data from parsed values in step 1):
INSERT INTO
mytable (`date`, sku, impressions, sales)
VALUES
('2015-01-01', 'ABCD', 100, 3),
('2015-01-02','ABCD', 100, 3),
...
ON DUPLICATE KEY UPDATE
impressions = VALUES(impressions),
sales = VALUES(sales)
A couple of notes:
check out the documentation for this syntax
if the next day's data update containing previous date record was supplementary, you could do sales = sales + VALUES(sales) but I don't think that's the case for you