I tried to investigate slow export on my model. I found out, that if I write dehydrate function on ModelResource like this:
def dehydrate_user_profile(self, search_report):
return search_report.user_profile_id
The export rapidly speeds up. I think, that django-import-export has bad method of getting the id values for ForeignKeys.
The bad performance is similar to this (notice the . instead of _)
def dehydrate_user_profile(self, search_report):
return search_report.user_profile.id
Hey @PetrDlouhy ! Thanks for the investigation. How much of an increase would you say? Would you like to submit a PR with some of these changes?
@andrewgy8 In my case the increase was extreme - when I try with just 1300 records it is almost instant vs. 20 seconds.
I investigated a little bit further. The problem is, that on line 90 in fields.py in get_value() function the value of foreign key is fetched by:
value = getattr(value, attr, None)
If the value is not already cached or not fetch by the queryset with select_related, it will trigger new DB query for every row of the export.
That value is further processed by Widget.render() where the value of certain key is fetched from the related object. This is unnecessary, if the key we want is primary key of the ForeignKey relation.
I am not sure, how to fix this properly.
One method might be to detect, that we want to fetch pk/id key and bypass the get_value()->Widget.render() mechanism.
Downside is that it might change behaviour if one of these function was overridden by user.
Second method is to automatically include select_related to the queryset.
Upside is that it can greatly improve performance not only for primary keys, but also for related fields.
Downside is that it might not work for some user modified querysets properly. Also in some cases it can lead to slowdown, because the select_related would increase the query complexity and/or fetch unwanted data (which might be solved by using only).
Any ideas how to improve this?
Most helpful comment
@andrewgy8 In my case the increase was extreme - when I try with just 1300 records it is almost instant vs. 20 seconds.
I investigated a little bit further. The problem is, that on line
90infields.pyinget_value()function the value of foreign key is fetched by:If the value is not already cached or not fetch by the queryset with
select_related, it will trigger new DB query for every row of the export.That value is further processed by
Widget.render()where the value of certain key is fetched from the related object. This is unnecessary, if the key we want is primary key of the ForeignKey relation.I am not sure, how to fix this properly.
One method might be to detect, that we want to fetch
pk/idkey and bypass theget_value()->Widget.render()mechanism.Downside is that it might change behaviour if one of these function was overridden by user.
Second method is to automatically include
select_relatedto the queryset.Upside is that it can greatly improve performance not only for primary keys, but also for related fields.
Downside is that it might not work for some user modified querysets properly. Also in some cases it can lead to slowdown, because the
select_relatedwould increase the query complexity and/or fetch unwanted data (which might be solved by usingonly).