Postgres: Upsert mutation¶
Table of contents
Introduction¶
An upsert query will insert an object into the database in case there is no conflict with another row in the table. In case there is a conflict with one or more rows, it will either update the fields of the conflicted rows or ignore the request.
Convert insert mutation to upsert¶
Note
Only tables with update permissions are upsertable. i.e. a table’s update permissions are respected before updating an existing row in case of a conflict.
To convert an insert mutation into an upsert, you need to use the on_conflict
argument to specify:
- a unique or primary key constraint using the
constraint
field, and - the columns to be updated in the case of a violation of that constraint using the
update_columns
field.
The value of the update_columns
field determines the behaviour of the upsert request as shown via the use cases
below.
Fetching Postgres constraint names
You can fetch details of unique or primary key constraints on a table by running the following SQL:
SELECT * FROM "information_schema"."table_constraints" WHERE table_name='<table>' AND table_schema='<schema>';
GraphQL engine will automatically generate constraint names as enum values for the constraint
field (try
autocompleting in GraphiQL). Typically, the constraint is automatically named as <table-name>_<column-name>_key
.
Upsert is not a substitute for update¶
The upsert functionality is sometimes confused with the update functionality. However, they work slightly
differently. An upsert mutation is used in the case when it’s not clear if the respective row is already present
in the database. If it’s known that the row is present in the database, update
is the functionality to use.
For an upsert, all columns that are necessary for an insert are required.
How it works
- Postgres tries to insert a row (hence all the required columns need to be present)
- If this fails because of some constraint, it updates the specified columns
If not all required columns are present, an error like NULL value unexpected for <not-specified-column>
can occur.
Update selected columns on conflict¶
The update_columns
field can be used to specify which columns to update in case a conflict occurs.
Example: Insert a new object in the article
table or, if the unique constraint article_title_key
is
violated, update the content
column of the existing article:
mutation upsert_article {
insert_article (
objects: [
{
title: "Article 1",
content: "Article 1 content",
published_on: "2018-10-12"
}
],
on_conflict: {
constraint: article_title_key,
update_columns: [content]
}
) {
returning {
id
title
content
published_on
}
}
}
Note that the published_on
column is left unchanged as it wasn’t present in update_columns
.
Update selected columns on conflict using a filter¶
A where
condition can be added to the on_conflict
clause to check a condition before making the update in case a
conflict occurs
Example: Insert a new object in the article
table, or if the unique key constraint article_title_key
is
violated, update the published_on
columns specified in update_columns
only if the previous published_on
value is lesser than the new value:
mutation upsert_article {
insert_article (
objects: [
{
title: "Article 2",
published_on: "2018-10-12"
}
],
on_conflict: {
constraint: article_title_key,
update_columns: [published_on],
where: {
published_on: {_lt: "2018-10-12"}
}
}
) {
returning {
id
title
published_on
}
}
}
Ignore request on conflict¶
If update_columns
is an empty array then on conflict the changes are ignored.
Example: Insert a new object into the author table or, if the unique constraint author_name_key
is violated,
ignore the request.
mutation upsert_author {
insert_author(
objects: [
{ name: "John" }
],
on_conflict: {
constraint: author_name_key,
update_columns: []
}
) {
affected_rows
}
}
In this case, the insert mutation is ignored because there is a conflict and update_columns
is empty.
Upsert in nested mutations¶
You can specify the on_conflict
clause while inserting nested objects:
Example:
mutation upsert_author_article {
insert_author(
objects: [
{
name: "John",
articles: {
data: [
{
title: "Article 3",
content: "Article 3 content"
}
],
on_conflict: {
constraint: article_title_key,
update_columns: [content]
}
}
}
]
) {
affected_rows
}
}
Nested upsert caveats¶
Note
The process by which nested inserts/upserts are executed is documented here.
Nested upserts will fail when:
- In case of an array relationship, the parent upsert does not affect any rows (i.e.
update_columns: []
for parent and a conflict occurs), as the array relationship objects are inserted after the parent. - In case of an object relationship, the nested object upsert does not affect any row (i.e.
update_columns: []
for nested object and a conflict occurs), as the object relationship object is inserted before the parent.
To allow upserting in these cases, set update_columns: [<conflict-columns>]
. By doing this, in case of a
conflict, the conflicted column/s will be updated with the new value (which is the same values as they had before and hence
will effectively leave them unchanged) and will allow the upsert to go through.