Skip to content

Immediate backup from replica cluster fails #169

Description

@fcanovai

An immediate backup (e.g.: creating together with the cluster a scheduled backup with immediate:true) on a freshly-created replica cluster will fail.

send-wal on the designated primary has no prior archive to resume from here, so it falls back to "whatever's current now" as its starting point; but pg_backup_start() on a standby can only report the last already-replayed restartpoint as the first WAL required, which only advances when the primary takes (and ships down) a new checkpoint. On an otherwise-idle cluster send-wal's start point lies ahead of that almost every time, and once it does, the segment the backup needs is permanently outside what tier1 will ever cover: klio backup run --wait-for-wals=true retries forever with no way to recover.

Metadata

Metadata

Assignees

No one assigned

    Labels

    bugSomething isn't working

    Type

    No type

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions