REAL Data Type in PostgreSQL

The PostgreSQL REAL type (also known as FLOAT4) stores single-precision floating-point numbers using 4 bytes. It is an inexact type — like DOUBLE PRECISION — but trades precision for storage efficiency. Use it for sensor readings, physical measurements, and high-volume time-series data where approximately 6 significant digits is sufficient.

Syntax

column_name REAL
column_name FLOAT4   -- alias, identical behavior

Key characteristics:

Practical Example

Store product weight measurements from a factory scale that only reports 5-digit precision:

CREATE TABLE shipments (
  id             BIGSERIAL PRIMARY KEY,
  order_id       INTEGER     NOT NULL,
  dispatched_at  TIMESTAMPTZ NOT NULL DEFAULT now(),
  weight_kg      REAL        NOT NULL,
  volume_liters  REAL        NOT NULL
);

INSERT INTO shipments (order_id, weight_kg, volume_liters)
VALUES
  (1001, 1.25, 3.50),
  (1002, 0.87, 2.10),
  (1003, 4.62, 8.00),
  (1004, 2.30, 5.25);

SELECT
  order_id,
  weight_kg,
  volume_liters,
  weight_kg / volume_liters AS density_kg_per_liter
FROM shipments
ORDER BY order_id;

Observe the characteristic single-precision rounding in the output:

 order_id | weight_kg | volume_liters | density_kg_per_liter
----------+-----------+---------------+----------------------
     1001 |      1.25 |           3.5 |    0.357142865657806
     1002 |      0.87 |           2.1 |    0.414285719394684
     1003 |      4.62 |           8.0 |             0.577500
     1004 |      2.30 |          5.25 |    0.438095241785049

The trailing digits (e.g. 0.357142865657806) are the visible manifestation of single-precision arithmetic.

REAL vs. DOUBLE PRECISION vs. NUMERIC

Avoiding Equality Comparisons

Single-precision arithmetic makes equality comparisons unreliable. Use range queries:

-- Avoid: may never match due to rounding
WHERE weight_kg = 1.25

-- Prefer: tolerant range
WHERE weight_kg BETWEEN 1.249 AND 1.251

Testing with Vela

When aggregating large volumes of REAL measurements (cumulative weight totals, average densities), accumulated rounding errors can become visible at scale. Before deploying a reporting query to production, test it on a database branch loaded with production-scale data to quantify the precision loss and decide whether to promote the column to DOUBLE PRECISION.

Production Tips

• Use REAL when storing large time-series datasets where storage reduction is meaningful and 6 digits of precision is sufficient.
• Avoid REAL for any financial, billing, or precise scientific computation — use NUMERIC for exact values.
• Avoid equality comparisons on REAL columns — always use range comparisons to account for floating-point imprecision.
• Consider DOUBLE PRECISION if you need to aggregate (SUM, AVG) REAL values frequently — accumulated rounding errors are more pronounced in single precision.
• Inspect the column definition after creation with \d+ table_name to confirm the type is stored as real and not silently promoted.