Jooq fails with "on conflict ... where" for a partical index, but sql works directly?-CodePudding

Using JOOQ 3.17.4, pgjdbc 42.5.0, postgres 14.3

I am aware of this answer here: https://stackoverflow.com/a/67292162/924597 - I've tried using the unqualified field name, but it makes no difference.

I'm trying to issue SQL that does "on conflict .. do update ... where" using a partial index, but I'm getting the error:

 ERROR: there is no unique or exclusion constraint matching the ON CONFLICT specification

The strange thing is, Postgres gives the failure, only when issuing the SQL through JOOQ. If I copy the SQL out of my console (as printed by P6Spy) and paste into IntelliJ IDEA's SQL editory - the exact same SQL executes properly.

Schema definition:

create table user_authz_request (
  id               bigint generated always as identity
    (start with 30000000)
    primary key                                not null,
  status           auth_request_status         not null,
  service_point_id bigint                      not null references service_point,
  email            varchar(256)                not null,
  client_id        varchar(256)                not null,
  id_provider      id_provider                 not null,
  subject          varchar(256)                not null,
  responding_user  bigint references app_user  null,
  description      varchar(1024)               not null,
  date_requested   timestamp without time zone default transaction_timestamp(),
  date_responded   timestamp without time zone null
);

create unique index user_authz_request_once_active_key
  on user_authz_request(service_point_id, client_id, subject)
  where status = 'REQUESTED';

JOOQ Code:

    db.insertInto(USER_AUTHZ_REQUEST).
      set(USER_AUTHZ_REQUEST.SERVICE_POINT_ID, req.getServicePointId()).
      set(USER_AUTHZ_REQUEST.STATUS, REQUESTED).
      set(USER_AUTHZ_REQUEST.EMAIL, email).
      set(USER_AUTHZ_REQUEST.CLIENT_ID, user.getClientId()).
      set(USER_AUTHZ_REQUEST.ID_PROVIDER, idProvider).
      set(USER_AUTHZ_REQUEST.SUBJECT, user.getSubject()).
      set(USER_AUTHZ_REQUEST.DESCRIPTION, req.getComments()).
      onConflict(
        USER_AUTHZ_REQUEST.SERVICE_POINT_ID,
        USER_AUTHZ_REQUEST.CLIENT_ID,
        USER_AUTHZ_REQUEST.SUBJECT).
        where(USER_AUTHZ_REQUEST.STATUS.eq(REQUESTED)).
      doUpdate().
        set(USER_AUTHZ_REQUEST.DESCRIPTION, req.getComments()).
        set(USER_AUTHZ_REQUEST.DATE_REQUESTED, LocalDateTime.now()).
      execute();

Generated SQL:

insert into api_svc.user_authz_request (service_point_id, status, email,
                                        client_id, id_provider, subject,
                                        description)
values (20000001, cast('REQUESTED' as api_svc.auth_request_status),
        'email', 'client id',
        cast('AAF' as api_svc.id_provider),
        'subject', 'first')
on conflict (service_point_id, client_id, subject)
where status = cast('REQUESTED' as api_svc.auth_request_status) do
update
set description  = 'first',
  date_requested = cast('2022-09-20T05:35:35.927 0000' as timestamp(6))

The above is the SQL that fails when it runs in my server, but works fine when I execute it through IntelliJ.

What am I doing wrong?

CodePudding user response：

Starting from jOOQ 3.18 and #12531, jOOQ will auto-inline all the bind variables in that WHERE clause, because it hardly ever makes sense to use bind values. It's a big exception in jOOQ's usual behaviour, because there are only very few cases where:

Bind values are syntactically correct
But at the same time, completely useless

In most other cases where bind values are auto-inlined, they are also not syntactically correct, so auto-inlining may be less controversial.

Until 3.18 and #12531 ships, you can simply inline your bind value manually, instead:

where(USER_AUTHZ_REQUEST.STATUS.eq(inline(REQUESTED, USER_AUTHZ_REQUEST.STATUS))).