Skip to content

docs(redis): PVC backup on node-local storage; refresh CRDs (cherry-pick of #57) - #58

Merged
chideat merged 2 commits into
masterfrom
cherry-pick/backup-node-local-storage-to-master
Sep 7, 2026
Merged

docs(redis): PVC backup on node-local storage; refresh CRDs (cherry-pick of #57)#58
chideat merged 2 commits into
masterfrom
cherry-pick/backup-node-local-storage-to-master

Conversation

@chideat

@chideat chideat commented Sep 7, 2026

Copy link
Copy Markdown
Contributor

Cherry-pick of #57 (release-5.1master), applied clean — master and release-5.1 shared the same tip (eb68068) at the branch point, so the resulting tree is byte-identical to #57's.

What

PVC backup no longer requires a ReadWriteMany storage provider, and the docs said it did. This documents the change and refreshes the CRD reference.

Backup on node-local storage

The backup PVC is now created ReadWriteOnce by default (spec.accessModes on RedisBackup / RedisClusterBackup), and restore no longer mounts that PVC into every Redis pod — a transient in-cluster source serves it read-only and each pod pulls its own data over the network. Verified against redis-operator 4f13d6c1 and f1eff1e6, both carried by tag v5.1 and absent from every v5.0.x tag.

  • docs/en/release_notes.mdx — new v5.1.0 entry, with the two limits that actually apply: the restore path is chosen by the instance's bundle version rather than by the backup's access mode, and selecting a node-local StorageClass in the console needs Data Services Essentials v4.4.1 or later (chart-middleware 320c9cc, first tagged in v4.4.1).
  • docs/en/functions/70-backup-restore.mdx — drops the stale "requires node migration and the ReadWriteMany access mode, such as NFS or Ceph" claim and the "network-based StorageClass" prerequisite. Adds the provisioner allow-list the console filters on (middleware-sc-cm-base union middleware-sc-cm) and the restore caveat for instances held on an older bundle version.
  • TERMINOLOGY_CANDIDATES.md — new unconfirmed terms with code provenance.

CRD refresh

docs/shared/crds re-synced from redis-operator release-5.1 (9b8d1fbc), which carries two breaking API changes made on 2026-09-03:

CRD Change Commit
Redis spec.activeRedis.redisUserNameredisUserRef 9630abfa
ActiveRedis spec.redisUserNameredisUserRef 9630abfa
ActiveRedisInspection spec.secretNamespec.redisUserRef 77851bdc
ActiveRedisConnection spec.secretName removed; spec.peerPort schema default dropped 77851bdc

Anchor fixes

doom lint resolves in-page anchors against explicit ids rather than slugs derived from heading text, and four links resolved to nothing. Three headings get their id; the fourth link is repointed at 30-operations.mdx#tuning, the page that owns the section.

Follow-ups, deliberately not in this PR

  1. Prose still uses the old field names. 12 occurrences of redisUserName across release_notes.mdx and four pages under functions/95-disaster-recovery/, plus ActiveRedisConnection YAML examples that set the now-removed secretName (20-disaster-recovery/20-setup.mdx:295,328, 30-failover.mdx:80) and peerPort guidance that still cites the removed 7379 default (20-setup.mdx:144,153). 77851bdc changed the Redis 6.0 link-auth behavior — it binds its default account now — so this is a rewrite of that flow, not a rename pass.
  2. RedisBackup / RedisClusterBackup are not in docs/shared/crds. This PR references their spec.accessModes; a reader cannot find them in the API section.

Notes

  • The documented bundle-version threshold is 5.1.0, chosen deliberately over the code boundary of 4.1.2 (the frozen v5.0 actor tier serves [4.1.2, 5.1.0) and also takes the network path). Recorded in TERMINOLOGY_CANDIDATES.md.
  • Two working-tree edits that predate this work ride along in release_notes.mdx: the "Redesigned" prefix dropped from the Cross-Datacenter Replication heading, and &project=Middleware removed from the three v5.1.0 release-notes-for-bugs templates.
  • yarn build passes; every in-repo anchor link resolves. yarn lint reports 37 pre-existing cspell errors (Alauda, Oplog, Jedis, Redisson — the shared terms list is not reachable from this sandbox); 0 in the files touched here.

🤖 Generated with Claude Code

Backup no longer needs a ReadWriteMany provider. The backup PVC is created
ReadWriteOnce by default (spec.accessModes on RedisBackup /
RedisClusterBackup), and restore no longer mounts that PVC into every Redis
pod: a transient in-cluster source serves it read-only and each pod pulls
its own data over the network. Verified against redis-operator 4f13d6c1 and
f1eff1e6, both carried by tag v5.1.

- release_notes: new 5.1.0 entry, with the two limits that apply — the path
  is chosen by bundle version, not by the backup's access mode, and picking
  a node-local StorageClass in the console needs Data Services Essentials
  v4.4.1 or later (chart-middleware 320c9cc, first tagged in v4.4.1)
- 70-backup-restore: drop the stale "requires node migration and
  ReadWriteMany, such as NFS or Ceph" claim and the "network-based
  StorageClass" prerequisite; document the provisioner allow-list the
  console filters on (middleware-sc-cm-base merged with middleware-sc-cm)
  and the restore caveat for instances held on an older bundle version

Refresh docs/shared/crds from redis-operator release-5.1 (9b8d1fbc), which
carries two breaking API changes made after the docs were written
(9630abfa, 77851bdc): redisUserName -> redisUserRef on Redis and
ActiveRedis, secretName -> redisUserRef on ActiveRedisInspection,
ActiveRedisConnection.secretName removed, and peerPort's schema default
dropped. The prose still uses the old names in five pages under
functions/95-disaster-recovery — a separate pass, since the 6.0 link-auth
flow changed behavior and not just field names.

Also carries two working-tree edits that predate this work: the
"Redesigned" prefix dropped from the Cross-Datacenter Replication heading,
and the &project=Middleware parameter removed from the three v5.1.0
release-notes-for-bugs templates.
`doom lint` resolves in-page anchors against explicit ids, not against
slugs derived from heading text, so four links resolved to nothing:

- architecture.mdx `## Sentinel Mode` had no id, breaking the link from
  the Disaster Recovery architecture page
- the Active-Active architecture page linked to two of its own headings
  that carried no id
- the Active-Active setup page pointed `tuning` at a same-page `#tuning`
  section that lives in 30-operations.mdx

Add the three ids and repoint the fourth link at the page that owns the
section. Every remaining in-repo anchor link resolves.
@cloudflare-workers-and-pages

Copy link
Copy Markdown

Deploying alauda-redis with  Cloudflare Pages  Cloudflare Pages

Latest commit: 9bb292e
Status: ✅  Deploy successful!
Preview URL: https://f794d92d.alauda-redis.pages.dev
Branch Preview URL: https://cherry-pick-backup-node-loca.alauda-redis.pages.dev

View logs

@chideat
chideat merged commit d924188 into master Sep 7, 2026
3 checks passed
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant