Race condition in PATCH on newly created documents with clustered mongo

[ehiggs]

When using a replicated set of mongo instances, it's normal to have the default client be configured to PREFER_SECONDARY. When performing a patch (or patch_internal), the code will call get_document, which then calls find_one which can be to the secondary. If the write concern is not total (also common) then the retrieved document state can be stale. This is a problem when one wants to insert data immediately, do some quick processing, and then patch the document to say the processing is done: the patch request might find that 'there is no document'.

Expected Behavior

PATCH on newly created documents should work.

post_internal(...)
patch_internal(...)

Actual Behavior

'No such document'

Environment

• Python version: 3.7
• Eve version: 1.1.1

[nicolaiarocci]

Fixed with #1412