How to optimize a sort based on “latest” related model

问题

So say we have two models

class Product(models.Model):
    """ A model representing a product in a website. Has new datapoints referencing this as a foreign key daily """
    name = models.CharField(null=False, max_length=1024, default="To be Scraped")
    url = models.URLField(null=False, blank=False, max_length=10000)


class DataPoint(models.Model):
    """ A model representing a datapoint in a Product's timeline. A new one is created for every product daily """
    product = models.ForeignKey(Product, null=False)
    price = models.FloatField(null=False, default=0.0)
    inventory_left = models.BigIntegerField(null=False, default=0)
    inventory_sold = models.BigIntegerField(null=False, default=0)
    date_created = models.DateField(auto_now_add=True)

    def __unicode__(self):
        return "%s - %s" % (self.product.name, self.inventory_sold)

The goal is to sort a QuerySet of products based on the inventory_sold value of the latest datapoint attached to the product. Here's what I have so far:

products = Product.objects.all()
datapoints = DataPoint.objects.filter(product__in=products)

datapoints = list(datapoints.values("product__id", "inventory_sold", "date_created"))
products_d = {}
# Loop over the datapoints values array
for i in datapoints:
    # If a datapoint for the product doesn't exist in the products_d, add the datapoint
    if str(i["product__id"]) not in products_d.keys():
        products_d[str(i["product__id"])] = {"inventory_sold": i["inventory_sold"], "date_created": i["date_created"]}
    # Otherwise, if the current datapoint was created after the existing datapoint, overwrite the datapoint in products_d
    else:
        if products_d[str(i["product__id"])]["date_created"] < i["date_created"]:
            products_d[str(i["product__id"])] = {"inventory_sold": i["inventory_sold"], "date_created": i["date_created"]}
# Sort the products queryset based on the value of inventory_sold in the products_d dictionary
products = sorted(products, key=lambda x: products_d.get(str(x.id), {}).get("inventory_sold", 0), reverse=True)

This works alright, but it's extremely slow with a high (500,000~) number of products and datapoints. Is there any better way to do this?

And on a side note (unimportant), since I haven't able to find anything about this, it seems like the unicode method of the DataPoint model is also making needless SQL queries. Is this something that's a default characteristic of Django models once they get passed to templates?

回答1:

I think you can use a subquery here to annotate the value of the most recent datapoint, then sort on that.

Based on the example in those docs, it would be something like:

from django.db.models import OuterRef, Subquery
newest = DataPoint.objects.filter(product=OuterRef('pk')).order_by('-date_created')
products = Product.objects.annotate(
    newest_inventory_sold=Subquery(newest.values('inventory_sold')[:1])
).order_by('newest_inventory_sold')

For your side point, to avoid the extra queries when outputting DataPoints you will need to use select_related in the original query:

datapoints = DatePoint.objects.filter(...).select_related('product')

This will do a JOIN so that getting the product name doesn't cause a new db lookup.

来源：https://stackoverflow.com/questions/51113494/how-to-optimize-a-sort-based-on-latest-related-model