Add sequence_overflow metrics

[alphabet5]

This query was based on

https://www.crunchydata.com/blog/the-integer-at-the-end-of-the-universe-integer-overflow-in-postgres

and exposes metrics for % used for the underlying column type.

This requires usage on all sequences, or GRANT pg_read_all_data TO postgres_exporter;

pg_read_all_data is a bit excessive, but granting usage on all sequences won't apply for new sequences that are added, and setting default privileges for schemas won't apply if new schemas are added.

pg_sequence_overflow_column_ratio{column_datatype="bigint", datname="mydb", instance="mydb.example.com", job="postgres-exporter", owned_by="my_table.id", schemaname="public", sequence="my_table_id_seq", sequence_datatype="bigint"}
	8.745220514546834e-10
pg_sequence_overflow_last_value{column_datatype="bigint", datname="mydb", instance="mydb.example.com", job="postgres-exporter", owned_by="my_table.id", region="dal", schemaname="public", sequence="my_table_id_seq", sequence_datatype="bigint"}
	8066042235
pg_sequence_overflow_sequence_ratio{column_datatype="bigint", datname="mydb", instance="mydb.example.com", job="postgres-exporter", owned_by="my_table.id", schemaname="public", sequence="my_table_id_seq", sequence_datatype="bigint"}
	8.745220514546834e-10

[Copilot]

Pull request overview

Adds a new optional sequence_overflow collector to expose “% used” overflow-related metrics for owned sequences, based on the underlying sequence and column integer types.

Changes:

• Introduces a new sequence_overflow collector (default disabled) that queries all non-template databases for owned sequences and exports last-value and ratio metrics.
• Documents the new collector flag and required privileges in the README.
• Adds unit tests validating metric emission for normal and NULL/unused sequence cases.

Reviewed changes

Copilot reviewed 3 out of 3 changed files in this pull request and generated 4 comments.

File	Description
README.md	Documents the new collector.sequence_overflow flag, default state, and required privileges.
collector/pg_sequence_overflow.go	Implements the new collector, including DB discovery, per-DB querying, and metric descriptors.
collector/pg_sequence_overflow_test.go	Adds unit tests for updateDatabase() metric output.

Comments suppressed due to low confidence (1)

collector/pg_sequence_overflow.go:110

• There are no unit tests for dsnForDatabase, but it has tricky behavior around URL vs key=value DSNs and dbname precedence (path vs query param). Adding focused tests for common DSN forms (postgresql://.../db, postgresql://.../?dbname=..., and key=value DSN with existing/absent dbname) would help prevent regressions.

// dsnForDatabase returns a new DSN with the database name replaced.
// It updates both the URL path and the dbname query parameter (if present),
// since libpq gives precedence to the query parameter over the path.
func dsnForDatabase(dsn string, datname string) (string, error) {
	u, err := url.Parse(dsn)
	if err != nil {
		return "", fmt.Errorf("could not parse DSN: %w", err)
	}
	u.Path = "/" + datname
	if q := u.Query(); q.Get("dbname") != "" {
		q.Set("dbname", datname)
		u.RawQuery = q.Encode()
	}
	return u.String(), nil
}

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[sysadmind]

The commit is also missing the DCO which is a requirement for this project. The CI details have instructions for how to fix it.

[sysadmind]

This functionality is unique to this collector. I think the intention is that these need to be run per database, but I think that other functionality would benefit from that too. I don't think that it should be specific to this collector. It's also not obvious that this collector would make new connections.

[sysadmind]

I don't think this exporter needs to calculate these values. That can be done in the Prometheus query. Having the raw value is much more useful for Prometheus. It means you can calculate this value on the fly and you know the absolute value, all with only storing a single time series.

[sysadmind]

Same here as above - we can calculate this at Prometheus query time.

[sysadmind]

I agree with copilot - this creates a lot of new connections which might not be desirable for users.