The backups line on Resilience (#8867)
/resilience and the terminal's Resilience dialog (palette: Resilience,
admins) both open with one line on backups. Both read it from
src/snapshot/backups_line.ts, so the web page and the terminal use the same
words.
- With a store:
Snapshots go to <where>.Here<where>is the store's
own name for itself: s3://… for a bucket, the path for a directory. It
never includes a key. On the web, the line links to Backups, which lists
the snapshots.
- With none: `Backups are off here: this forge was given no bucket to
keep snapshots in.` This is muted text, not an alarm: a preview and a forge on a desk have no bucket on purpose. On the web, the line links to Backups, which says how to give the forge one.
- An empty name counts as none. A store whose
whereis blank reads as
off. Without this rule the line would end in a bare "Snapshots go to .".
Giving a forge a bucket
The first of these that is set wins (src/snapshot/destination.ts):
- Sealed in the forge. On the web,
/settings/backupstakes an
endpoint, bucket, region and two keys. In the terminal, the Resilience
dialog's d asks for the same five fields. The keys are sealed and are
never shown again. t tests the destination.
- A directory:
FORGE_SNAPSHOT_DIR=/pathin the forge's environment. - The host's bucket:
S3_ENDPOINT,S3_BUCKET,S3_ACCESS_KEY,
S3_SECRET_KEY and optionally S3_REGION, which a host-provisioned forge is
given.
The line describes the store that the running process holds. After you save a destination, the next use resolves it again and the line names it.
When a restore was last proven (#13052)
A backup is only as good as the last restore that worked. The forge's restore drill restores the newest snapshot into a scratch directory and checks it; this line says when that last passed, so an owner sees it without running one.
Where to read it
/system(admins): the Restore section is one sentence.- The terminal's System screen: the same sentence, under the forge's
health line.
What it says
Restore never proven.until a drill has passed.Restore proven 2 days ago, 45 seconds, snapshot 3 hours old.after one:
when it passed, how long it took (seconds or minutes, never milliseconds), and how old the restored snapshot was then.
- A drill that failed after the last pass is added:
…; the last drill failed 2 hours ago at restore.
Prove one now
Open /settings/backups and run a drill, or call the MCP tool
restore_drill. Reload /system: the line reads Restore proven just now.