ORM Speed Up In Frappe

2023-07-18

ORMs can be slow. However, they are also very useful and central to Frappe's entire metadata and document centric model.

Frappe's ORM can trigger validations and all kinds of side-effects via document hooks. Like it, or not, these hooks are how one writes Frappe code. So, the ORM is critical on the write-path, if (far) less efficient than raw SQL writes. We have previously explored this topic for bulk writes.

Some of our apps have read heavy APIs where a number of documents are read during a single API transaction. Until recently we had no choice but to choose one of the following:

• Use frappe.get_doc and accept the performance penalty,
• Write raw SQL queries and combine multiple reads into a single DB transaction,
• Roll our own caching and carefully manage invalidations.

Brave new world

However, I recently discovered frappe.get_cached_doc and it's essentially a free performance upgrade. The documentation is pretty straightforward:

• the return value of get_cached_doc is equivalent to the return value of get_doc
• the cache is updated any time the ORM is used to update the document (doc.save or frappe.db.set_value)
  • Raw DB updates and doc.db_set do not update the cache. This does mean our bulk_insert method does not update the cache. Thankfully, there's frappe.clear_document_cache(doctype, name)

Performance upgrades

What does this mean in terms of performance? A cool 10000x+ increase in read throughput. See the code below for my simplistic benchmark.

from timeit import timeit

def load_doc():
    return frappe.get_doc(doctype, docname)

def load_cached_doc():
    return frappe.get_cached_doc(doctype, docname)

print(timeit(load_doc, number=1000))
# 17.63


print(timeit(load_cached_doc, number=10000000))
# 13.00

Sure, caching documents needs more RAM. But RAM is dirt cheap these days, so go ahead and replace frappe.get_doc with frappe.get_cached_doc in your performance critical paths and enjoy an immediate boost to API responsiveness.