Merge pull request #228 from yupoet/docs/single-node-upgrade-from-archived-minio

Preserve the single-node migration report contributed by @yupoet.

Signed-off-by: Feng Ruohang <rh@vonng.com>

docs/upgrade/single-node-from-archived-minio.md

@@ -0,0 +1,80 @@
+# Upgrading a Single-Node Deployment from the Archived MinIO Community Edition
+
+The MinIO community edition repository was archived on 2026-04-25 and its binary
+distribution channels (`dl.min.io`, the Docker Hub tags, quay.io) were withdrawn
+afterwards. Operators still running the final community release need a documented,
+validated path forward. This guide records a real-world, single-node upgrade from
+`RELEASE.2025-10-15T17-29-55Z` (the last community release) to Silo
+`RELEASE.2026-09-16T00-00-00Z`, performed in production on 2026-09-26.
+
+The [compatibility audit](https://silo.pgsty.com/compatibility/server/) covers
+the server behavior changes but does not specifically address the
+single-node/single-drive (FS mode) case. This document fills that gap with
+field data.
+
+## Validated environment
+
+- Single node, single drive: `server /data` in FS mode, running under Docker Compose
+- Root credentials only — no IAM sub-users, no OIDC, no notification targets, no reverse proxy
+- ~5,900 objects / ~34,000 stored chunks accumulated over 15 months on the source version
+- Source: `RELEASE.2025-10-15T17-29-55Z` → target: `RELEASE.2026-09-16T00-00-00Z`
+
+## Why this is a low-risk swap for standalone deployments
+
+- **Zero data migration.** `.minio.sys`, erasure metadata, bucket/object layout and
+  encryption formats keep their names and schemas; the existing data volume is opened
+  in place, with no copy and no metadata rewrite.
+- **Single nodes are unaffected by the removed private `ReadMultiple` storage REST
+  call** — that only constrains mixed-version multi-node clusters.
+- **The IAM/password-policy tightening does not apply** to plain root-credential
+  standalone setups (no sub-users, no OIDC, no bucket policies in use).
+
+## Procedure
+
+1. **Back up the data volume first** (tar the volume or snapshot the disk). Even with
+   a compatible format, an upgrade window is the worst time to discover your only
+   copy is damaged.
+2. **Pin the exact target image tag** in your deployment tooling. Floating tags make
+   both upgrades and rollbacks ambiguous.
+3. **Stop dependent services before swapping the storage container**, so no writes
+   race the restart.
+4. **Recreate the storage container with the Silo image, keeping the data volume
+   unchanged.** Compatibility notes:
+   - `minio server …` invocations translate transparently; the official container
+     entrypoint handles the rename.
+   - Healthchecks that run `curl -f /minio/health/live` inside the container work
+     out of the box with the official images (curl is bundled). If you build custom
+     images, remember to include curl — the upstream release images shipped a static
+     curl, minimal bases may not have one.
+5. **Verify in layers**:
+   - Startup log: expect zero errors. Watch specifically for hardening rejections —
+     Silo refuses malformed historical objects (poisoned metadata, invalid erasure
+     geometry) that older code tolerated; such objects need manual remediation
+     before they will serve again.
+   - `GET /minio/health/live` and `/minio/health/ready` → 200.
+   - Application-level S3 smoke: upload → read back → delete, through the real client.
+
+## Observed results (2026-09-26 field run)
+
+- In-place upgrade with zero migration steps; the 15-month-old data volume opened
+  with no errors and **no objects were rejected by the new hardening**.
+- A full application regression against the storage (multipart upload, object read
+  by a parser service, listing, deletion) passed end to end.
+- **Rollback path is image-swap only** (same on-disk schema), but note we did not
+  exercise a reverse downgrade in production. Our evidence is a staged forward
+  migration — the same data volume was opened read-write by three engine builds
+  in sequence (`2025-09` → `2025-10` → `2026-09`) with no format intervention —
+  plus the documented schema retention. Keep the previous image available locally:
+  registry tags you depend on may disappear, as the community tags already have.
+
+## Post-upgrade behavior changes to check
+
+These did not affect the validated standalone setup, but review them against your
+own clients before upgrading:
+
+- **TLS defaults changed** across 20260903/20260916 — verify clients if you serve TLS
+  directly from the storage node or front it with older proxies/identity providers.
+- **`If-Match` on `DeleteObject`** was ignored in 20260903 and is enforced in 20260916.
+- **Multipart listing is capped at 1,000 entries** per request in newer releases.
+- `MINIO_API_TRUSTED_PROXIES` still defaults to trusting every proxy header; set it
+  explicitly if the node sits behind a proxy.