Postgresql – Unexpected unique constraint violation during update

postgresqlsubqueryunique-constraintupdate

In table foo with two columns (id and seq), I'd like to add +1 to seq for all records with an arbitrary seq > 4738. The plan is to insert a new record at seq=4739 once all the seq > 4738 records are shifted by +1.

This is the table.

CREATE TABLE foo
(
  id uuid NOT NULL,
  seq integer NOT NULL,
  CONSTRAINT seq_key UNIQUE (seq)
)

CREATE UNIQUE INDEX idx_id
  ON foo
  USING btree
  (id);

CREATE UNIQUE INDEX idx_seq
  ON foo
  USING btree
  (seq);

I try to achieve the +1 shift with the following query. Note, that I use a subquery in an attempt to update the > 4738 records in descending order (i.e. assuming max seq=10000 then the last record is updated first (10000->10001), then second last (seq=10000 doesn't exist at this point, seq=9999 -> seq=10000 (no constraint violation), then 9998 -> 9999, … to avoid at any one time that a unique constraint violation occurs). However, this assumes sequential execution of the update query which isn't what's seemingly happening.

When running

UPDATE foo SET seq=anon_1.new_seq FROM (
  SELECT foo.id AS id, foo.seq + 1 AS new_seq
  FROM foo
  WHERE foo.seq > 4738 ORDER BY foo.seq DESC
) AS anon_1 WHERE foo.id = anon_1.id

I get the following error.

duplicate key value violates unique constraint "seq_key"
DETAIL: Key (seq)=(7334) already exists.

Obviously, that's unexpected (as the constraint was satisfied before the UPDATE). Is there anything I could try to resolve the problem (different index type, only constraint, only index)? I noticed that this error very much depends on the number of updated records. If fewer records are updated, then this problem doesn't seem to occur (this may hint to some parallel execution for when there are too many records to update which may interleave and cause some intermediate state in which the constraint/index is not unique any more). Some thoughts and ideas would be very appreciated.

Best Answer

You need to create a deferrable constraint rather than an index.

CREATE TABLE foo
(
  id uuid NOT NULL primary key
  seq integer NOT NULL,
  CONSTRAINT seq_key UNIQUE (seq)
    deferrable initially immediate --<< this
);

By declaring the constraint as deferrable it will be checked at the end of the UPDATE statement (to be more precise: at the end of the transaction) rather than row-by-row.

Related Solutions

Sql-server – Unique Constraint with multiple null columns

I think your database is suffering from a bad data model. You might want to rethink it a little, judging from the tables you've mentioned the model isn't well normalized and you're trying to enforce a foreign key reference to a table column which isn't really a key to anything. One way to go is something like this :

CREATE TABLE EMP (ID INT PRIMARY KEY IDENTITY
                 ,EmpName VARCHAR(100) NULL)

CREATE TABLE PCTypes (id INT PRIMARY KEY IDENTITY
                     ,PcTypeName VARCHAR(100) NULL)

CREATE TABLE PC (Id INT PRIMARY KEY IDENTITY
                ,EmpId INT NULL
                ,PCName VARCHAR(100) NULL
                ,HostName VARCHAR(100) NULL
                ,PhysicalLocation VARCHAR(100) NULL
                ,PCType INT NULL)

ALTER TABLE PC ADD CONSTRAINT FK_PC_EmpId FOREIGN KEY (empId) REFERENCES Emp(id)
ALTER TABLE PC ADD CONSTRAINT FK_PC_PCType FOREIGN KEY (PCType) REFERENCES PCTypes(id)

CREATE TABLE PCNetwork (Id INT PRIMARY KEY IDENTITY
                       ,NetworkName VARCHAR(100) NULL
                       ,EmpId INT NULL
                       ,PhysicalLocation VARCHAR(100) NULL)

ALTER TABLE PCNetwork ADD CONSTRAINT FK_PCNetwork_EmpId FOREIGN KEY (empId) REFERENCES Emp(id)

CREATE TABLE PCNetwork_PCs (PCNetworkId INT NOT NULL
                           ,PCId INT NOT NULL
                           ,IPAddr NVARCHAR(20) NULL)

ALTER TABLE PCNetwork_PCs ADD CONSTRAINT FK_PCNetwork_PCs_Network FOREIGN KEY (PCNetworkId) REFERENCES PCNetwork(id)
ALTER TABLE PCNetwork_PCs ADD CONSTRAINT FK_PCNetwork_PCs_PC FOREIGN KEY (PCId) REFERENCES PC(id)
CREATE UNIQUE INDEX uq_pcnetworks ON PCNetwork_PCs(PCNetworkId, PCId)
CREATE UNIQUE INDEX uq_pcnetworks_ip ON PCNetwork_PCs(IPAddr)

Postgresql – Creating unique constraint to be validated from input

This can be achieved by using a unique index on the combination of counter and the month of date_booked. Something like this:

create unique index idx_tbl on tbl (counter, to_char(date_booked, 'yyyymm'));

However the above will fail because to_char() is not an "immutable" function and thus cannot be used in an index (even if we know that the above is fine). The workaround for this is to create a wrapper function that is marked as immutable:

create function my_month(some_date date)
  returns varchar
  immutable 
  language sql
as  
$$
  select to_char(some_date, 'yyyymm')
$$
;

Then you can create an index using that function:

create unique index idx_tbl on tbl (counter, my_month(date_booked));

If you now insert the fourth row, you get the error:

duplicate key value violates unique constraint "idx_tbl"

If the table is big, you might want to use a number instead of a varchar in order to reduce the size of the index as well e.g. by using to_number(to_char(some_date, 'yyyymm'))

Best Answer

Related Solutions

Sql-server – Unique Constraint with multiple null columns

Postgresql – Creating unique constraint to be validated from input

Related Question