XML/SOAP support for RPC
RPC now understands the
text/xml media type, allowing SQL functions to send XML output(
Accept: text/xml) and receive XML input(
Content-Type: text/xml). This makes SOAP endpoints possible, check the Create a SOAP endpoint how-to and the Scalar Function Response Format reference for more details.
GeoJSON is supported across the board(reads, writes, RPC) with the
Accept: application/geo+json header, this depends on PostGIS from the versions 3.0.0 and up. The working with PostGIS section has an example to get you started.
The execution plan of a request is now obtainable with the
Accept: application/vnd.pgrst.plan header. The result can be in
json formats and is compatible with EXPLAIN vizualizers like explain.depesz.com or explain.dalibo.com.
A one-to-one relationship is now detected when a foreign key is unique.
Using Computed relationships, you can add custom relationships or override automatically detected ones. This makes Resource Embedding possible on Foreign Data Wrappers and complex SQL views.
Accessing fields of a Composite type or elements of an Array type is now possible with the arrow operators(
->>) in the same way you would access a JSON type fields.
Pattern Matching operators for POSIX regular expressions are now available:
imatch, equivalent in PostgreSQL to
limitcan now affect the number of updated/deleted rows. See Limited Update/Delete.
You can now activate the “Authorize” button in SwaggerUI by enabling the openapi-security-active configuration. Add your JWT token prepending
Bearer to it and you’ll be able to request protected resources.
Two health check endpoints are now exposed in a secondary port.
Logging now shows the database user.
It is now possible to execute PostgREST without specifying any configuration variable. The three that were mandatory on the previous versions, are no longer so.
If db-uri is not set, PostgREST will use the libpq environment variables for the database connection.
If db-schemas is not set, it will use the database
If db-anon-role is not set, it will not allow anonymous requests.
To increase consistency, all the errors messages are now normalized. The
messagefields will always be present in the body, each one defaulting to a
nullvalue. In the same way, the errors that were raised with
SQLSTATEnow include the
codein the body.
To further clarify the source of an error, we now add a
PGRSTprefix to the error code of all the errors that are PostgREST-specific and don’t come from the database. These errors have unique codes that identify them and are documented in the PostgREST Error Codes section.
Added a Working with PostgreSQL data types how-to, which contains explanations and examples on how to work with different PostgreSQL data types such as timestamps, ranges or PostGIS types, among others.
Added in-database and environment variable settings for each configuration variable.
Added the File Descriptors subsection.
Added a reference page for Errors.
Moved the Errors and the HTTP Status Codes sections to the errors reference page.
Moved the Casting type to custom JSON how-to to the Casting a Range to a JSON Object subsection.
Removed direct links for PostgREST versions older than 8.0 from the versions menu.
Removed the Embedding table from another schema how-to.
Restructured the Resource Embedding section:
Added a One-to-many relationships and Many-to-one relationships subsections.
Renamed the Embedding through join tables subsection to Many-to-many relationships.
Split up the Insertions/Updates section into Insert and Update.
Many-to-many relationships now require that foreign key columns be part of the join table composite key
This was needed to reduce Embedding Disambiguation errors in complex schemas(#2070).
For migrating to this version, the less invasive method is to use Computed relationships to replace the previous many-to-many relationships.
Otherwise you can change your join table primary key. For example with
alter table permission_user drop constraint permission_user_pkey, add primary key (id, user_id, permission_id);
Views now are not detected when embedding using Target Disambiguation.
This embedding form was easily made ambiguous whenever a new view was added(#2277).
For migrating to this version, you can use Computed relationships to replace the previous view relationships.
Hint Disambiguation works as usual on views.
limit/offsetnow limits the affected rows on
offsetonly limited the returned rows but not the actual updated rows(#2156)
max-rowsis no longer applied on
This was misleading because the affected rows were not really affected by
max-rows, only the returned rows were limited(#2155)
204 No Contentwithout
Content-Typefor RPCs returning
Previously, those RPCs would return
nullas a body with
Prefer: return=representationno longer returns a
This reduces unnecessary computing for all insertions (#2312)
204 No Contentwithout
Clarify error for failed schema cache load. (#2107)
Database connection lost. Retrying the connectionto
Could not query the database for the schema cache. Retrying.
Fix silently ignoring filter on a non-existent embedded resource (#1771)
Remove functions, which are not callable due to unnamed arguments, from schema cache and OpenAPI output. (#2152)
Fix accessing JSON array fields with
Fix inferring a foreign key column as a primary key column on views (#2254)
Restrict generated many-to-many relationships (#2070)
Only adds many-to-many relationships when a table has foreign keys to two other tables and these foreign key columns are part of the table’s primary key columns.
Allow casting to types with underscores and numbers (e.g.
Prevent views from breaking one-to-many/many-to-one embeds when using column or foreign key as target (#2277, #2238, #1643)
When using a column or foreign key as target for embedding (
/tbl?select=*,col-or-fk(*)), only tables are now detected and views are not.
You can still use a column or an inferred foreign key on a view to embed a table (
db-pool-timeoutto 1 hour to prevent frequent high connection latency (#2317)
The search path now correctly identifies schemas with uppercase and special characters in their names (regression) (#2341)
“404 Not Found” on nested routes and “405 Method Not Allowed” errors no longer start an empty database transaction (#2364)
Fix inaccurate result count when an inner embed was selected after a normal embed in the query string (#2342)
OPTIONSrequests no longer start an empty database transaction (#2376)
Allow using columns with dollar sign ($) without double quoting in filters and
Fix loop crash error on startup in PostgreSQL 15 beta 3.
Log: "UNION types \"char\" and text cannot be matched."(#2410)
Fix race conditions managing database connection helper (#2397)
limit=0in the request query to return an empty array (#2269)
Big thanks from the PostgREST team to our sponsors!
If you like to join them please consider supporting PostgREST development.