fix(backup): skip stanza-create on standby to enable prefer-standby backups

[robrigo]

Problem

The Backup RPC (internal/cnpgi/instance/backup.go) calls CreatePgbackrestStanza unconditionally before every backup. pgbackrest stanza-create requires a primary PostgreSQL connection — it reads the live cluster's control data to initialize the stanza metadata in the repo. On a standby it fails with pgbackrest exit code 56 (unable to find primary cluster), which aborts the whole backup.

The practical effect: ScheduledBackup/Backup with target: prefer-standby (or standby) is unusable. Any cluster that wants to offload the read-heavy backup I/O from its primary is forced back to target: primary.

This is the upstream issue we filed: operasoftware#83.

Fix

Skip stanza-create when the instance running the backup is a standby.

This is safe: the stanza is initialized by the primary — both its own backups and WAL archiving's CheckWalArchiveDestination run against the same repo — so by the time a standby backup runs, the stanza already exists. stanza-create is only genuinely required on the primary.

Role detection uses the standby.signal file in PGDATA — the same signal PostgreSQL itself uses to start in standby mode, and the same mechanism CloudNativePG uses to distinguish a replica from a primary (pkg/utils / instance manager). This means no database connection or credentials are needed at backup time. A missing/odd PGDATA falls back to the primary path (stanza-create runs), preserving today's behavior.

isStandby, err := instanceIsStandby(b.PGDataPath)
// ...
if isStandby {
    // skip — stanza is initialized by the primary
} else {
    backupCmd.CreatePgbackrestStanza(...)
}

Tests

• Adds the internal/cnpgi/instance package's first unit-test suite (Ginkgo/Gomega).
• instanceIsStandby covered for: primary (no signal → false), standby (signal present → true), and missing-PGDATA (→ false, falls back to primary path).
• go build ./..., go vet, gofmt -l, and the non-infra unit suites all pass. (The internal/controller envtest suite and test/e2e require a live cluster / envtest binaries and are unaffected by this change.)

Motivation

pgBackRest backups are the single largest recurring read-IO event on our primary data volumes (they pin the volume to ~100% utilization twice daily). Offloading them to a replica via prefer-standby needs this fix.

Upstream

We'll offer this (or an equivalent) to upstream against operasoftware#83. This fork PR lets us build a FACINGS image and validate on staging in the meantime.

[Copilot]

Pull request overview

This PR updates the instance-side Backup RPC to avoid running pgbackrest stanza-create on standby instances, enabling target: prefer-standby / standby backups to succeed by relying on stanza initialization performed on the primary.

Changes:

• Detect standby instances via PGDATA/standby.signal and skip stanza-create when running on a standby.
• Add a new instanceIsStandby helper to encapsulate role detection.
• Introduce the first unit test suite for internal/cnpgi/instance (Ginkgo/Gomega) and cover instanceIsStandby.

Reviewed changes

Copilot reviewed 3 out of 3 changed files in this pull request and generated 2 comments.

File	Description
internal/cnpgi/instance/backup.go	Skips stanza-create on standby; adds instanceIsStandby helper for role detection.
internal/cnpgi/instance/backup_test.go	Adds unit tests for instanceIsStandby behavior across primary/standby/missing PGDATA.
internal/cnpgi/instance/suite_test.go	Adds Ginkgo suite bootstrap for the internal/cnpgi/instance package.

---

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