Add aurora_replica_status collector by megative · Pull Request #1311 · prometheus-community/postgres_exporter

megative · 2026-05-17T10:48:38Z

Summary

First PR in a series splitting #1308 (Aurora support) into one-PR-per-collector as requested by @sysadmind.

This PR adds:

Aurora auto-detection infrastructure on instance struct. Mirrors the existing version field pattern (per @SuperQ feedback). On everysetup(), a one-shot SELECT aurora_version() runs via sync.Once — exactly one probe per process lifetime. The result is cached in instance.isAurora and read by Aurora-specific collectors.
aurora_replica_status collector. Exposes the values from Aurora's aurora_replica_status() built-in: replica lag (per server), replay latency, and pending read IOs. Disabled by default (--collector.aurora_replica_status); on non-Aurora servers it short-circuits to ErrNoData cleanly.

Test plan

go build ./..., go vet ./..., go test ./... clean.
Tested live against an Aurora PostgreSQL 17.7.0 cluster — collector
emits metrics; per-process aurora_version() probe fires exactly once.
Tested against vanilla PostgreSQL 17 — collector returns ErrNoData
silently; no error logs.

Follow-up PRs in the series will add the remaining 8 Aurora collectors on top of this infrastructure.

@sysadmind

Add the first Amazon Aurora PostgreSQL collector, with the supporting Aurora detection infrastructure on the instance struct. - instance.isAurora bool field, set by setup() via a one-shot SELECT aurora_version() (sync.Once, one probe per process) - aurora_replica_status collector exposes replica lag, replay latency, and pending read IOs from aurora_replica_status(). Gated on instance.isAurora — on non-Aurora servers returns ErrNoData. - Disabled by default; opt-in via --collector.aurora_replica_status. First in a series of per-collector PRs splitting prometheus-community#1308 as requested by @sysadmind. Signed-off-by: Pavel K <megativ3@gmail.com>

megative · 2026-05-19T19:07:32Z

hi @SuperQ @sysadmind
Could you guys please take a look? It would be great to deliver Aurora support to upstream.

megative · 2026-05-22T16:18:38Z

@SuperQ @sysadmind guys, sorry for pushing but any luck you have a chance to review this one?

…ction-and-replica-status

ArthurSens

Disclaimer, I'm not a maintainer of postgres_exporter, I'm just trying to unload a bit of their review load :)

ArthurSens · 2026-06-02T20:07:50Z


 * [CHANGE] ...
-* [FEATURE] ...
+* [FEATURE] Add `aurora_replica_status` collector for Amazon Aurora PostgreSQL, with Aurora auto-detection on the `instance` struct.


Suggested change

* [FEATURE] Add `aurora_replica_status` collector for Amazon Aurora PostgreSQL, with Aurora auto-detection on the `instance` struct.

* [FEATURE] Add `aurora_replica_status` collector for Amazon Aurora PostgreSQL.

The second part of the phrase seems like implementation details; the postgres_exporter end-user doesn't care what an "instance struct" is

ArthurSens · 2026-06-02T20:16:59Z

+
+var (
+	auroraReplicaLagDesc = prometheus.NewDesc(
+		prometheus.BuildFQName(namespace, auroraReplicaSubsystem, "lag_milliseconds"),


prometheus recommends using base units for metric units. We have a few examples of other collectors doing this transformation, like pg_setting and pg_stat_statements

agreed. This should be converted.

ArthurSens · 2026-06-02T20:17:10Z

+		[]string{"server_id"}, nil,
+	)
+	auroraReplayLatencyDesc = prometheus.NewDesc(
+		prometheus.BuildFQName(namespace, auroraReplicaSubsystem, "replay_latency_microseconds"),


same applies here

ArthurSens · 2026-06-02T20:23:54Z

 	db, err := sql.Open("postgres", dsn)
 	if err != nil {
 		return nil, err
 	}
 	db.Close()


asking without thinking too much... could the aurora detection be made here? If I understand it correctly, this runs only once and never again. So if we made the detection here, it would make the sync.Once shenanigans unecessary

I'm probably missing context too since I'm not a maintainer of this exporter, but I'm wondering why version detection isn't made here either

We also have the option to do detection where we do PostgreSQL instance version checks.

ArthurSens · 2026-06-02T20:26:34Z

+type auroraProbe struct {
+	once     sync.Once
+	isAurora bool
+}
+
 type instance struct {
 	dsn     string
 	db      *sql.DB
 	version semver.Version
+
+	// isAurora is set by setup() via the shared auroraProbe (one-shot
+	// SELECT aurora_version() per process). Aurora-specific collectors
+	// gate on this field; on non-Aurora servers they short-circuit to
+	// ErrNoData.
+	isAurora    bool
+	auroraProbe *auroraProbe
 }


why isAurora set in both instance and auroraProbe? Seems like instance has access to two fields that have the same meaning/value

I don't think we need both. I think it's complicated and doesn't add much value.

ArthurSens · 2026-06-02T20:32:52Z

+	if i.auroraProbe != nil {
+		i.auroraProbe.once.Do(func() {
+			i.auroraProbe.isAurora = detectAurora(i.db)
+		})
+		i.isAurora = i.auroraProbe.isAurora
+	}


I don't see this sync.Once protection against the query that detects version. I'm not sure if I'm missing something, but looks like an overcomplication....?

sysadmind · 2026-06-10T02:11:17Z

+// auroraReplicaStatusCollector exposes metrics from Aurora PostgreSQL's
+// aurora_replica_status() function. Disabled by default because the function
+// only exists on Aurora.
+const auroraReplicaSubsystem = "aurora_replica"


I think it's unintuitive for the subsystem and the collector name below to be different. Someone runs --collector.aurora_replica_status but the metrics are aurora_replace_*. These should match. Based on the table name, they should both be aurora_replica_status.

sysadmind · 2026-06-10T02:14:58Z

+	}()
+
+	// writer: 1 metric (pending_read_ios), reader: 3 metrics → 4 total
+	expected := 4


This only validates that we had 4 metrics, not what they were. It's incomplete.

sysadmind · 2026-06-10T02:17:35Z

+non-Aurora server they silently emit no data.
+
+* `[no-]collector.aurora_replica_status`
+  Replica lag, replay latency, pending read IOs from `aurora_replica_status()`. (default: disabled)


I think calling out each metric here makes it easy to miss when the collector changes.

megative mentioned this pull request May 17, 2026

Add Amazon Aurora PostgreSQL support #1308

Open

5 tasks

Merge branch 'prometheus-community:master' into feat/aurora-init-dete…

a0324bc

…ction-and-replica-status

ArthurSens reviewed Jun 2, 2026

View reviewed changes

sysadmind requested changes Jun 10, 2026

View reviewed changes

	* [FEATURE] Add `aurora_replica_status` collector for Amazon Aurora PostgreSQL, with Aurora auto-detection on the `instance` struct.
	* [FEATURE] Add `aurora_replica_status` collector for Amazon Aurora PostgreSQL.

Uh oh!

Conversation

megative commented May 17, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Summary

Test plan

Uh oh!

megative commented May 19, 2026

Uh oh!

megative commented May 22, 2026

Uh oh!

ArthurSens left a comment

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

megative commented May 17, 2026 •

edited

Loading