How Max Partition Size Checks Work

Definition

Asserts the maximum number of records that should be loaded from each file or table partition.

Overview

The Max Partition Size rule sets an upper bound on the number of records the platform loads from a single partition. After each partition is loaded, its record count is compared against the configured Maximum partition size, and a partition that exceeds it is reported.

Because the rule looks at the load rather than the data, it takes no field, no filter clause, and no coverage threshold. A partition either fits under the ceiling or it does not.

Typical use cases:

Detect partition skew in distributed data.
Monitor daily batch sizes.
Identify runaway data growth in partitions.

Field Scope

None: The rule takes no field. It counts the records in each loaded partition, so the check is attached to the container itself and the form shows no field picker.

Anomaly Types

Type	Supported
Record Flag inconsistencies at the row level
Shape Flag inconsistencies in the overall patterns and distributions of a field

Evaluation Flow

Every Max Partition Size check follows the same three-step evaluation flow:

Load the partition. The Scan reads one partition of the container: a file, or the slice of a table the read strategy selected.
Count its records. The platform uses the measured record count for that partition. When no measured count is available, it falls back to the number of rows in the loaded data.
Compare it against the threshold. The partition passes when its count is at or below Maximum partition size. A partition above the threshold produces a Shape Anomaly naming the partition.

What Counts as a Partition

A partition is one unit of data the platform loads during a Scan. What that means depends on the container:

For a file container, each file loaded from the folder is a partition.
For a table, the partition is the slice the Scan loads: the whole table on a Full scan, or the incremental slice on an Incremental scan.

The check runs once per loaded partition, so a container that loads several partitions in one Scan is evaluated several times, and each failing partition produces its own anomaly.

The Rule Evaluates the Load, Not the Values

Max Partition Size never reads a column. It compares a record count against a number, which is why the form shows no field picker, no filter clause, and no coverage slider: there are no rows to scope or to tolerate.

That also means the rule cannot tell you which records are missing or extra. It tells you the partition is the wrong size. Pair it with a value-level check when you also need to know what is wrong inside it.

The Threshold Applies Per Partition, Not Per Container

A container that loads ten partitions in one Scan is evaluated ten times against the same threshold. The number you set is the size of one partition, not the total for the Scan. If you want to bound the whole container instead, use a Volumetric check, which tracks the container's row count over time.

No Filter Clause

Max Partition Size takes no filter clause. The rule counts the records in a loaded partition, and a row filter would change the count it is meant to measure, so the form shows no Filter Clause field and the API ignores the filter value.

Coverage

Max Partition Size does not use a coverage threshold. The coverage field on the API payload is ignored.