gsadmin/rustfs
1
0
Fork 0
mirror of https://github.com/rustfs/rustfs.git synced 2026-08-16 09:58:21 +00:00
Code Issues Packages Projects Releases Wiki Activity

fix(rebalance): converge multipart data movement retries (#6057)

Browse Source 
* fix(rebalance): converge multipart data movement retries

* fix(rebalance): harden multipart retry replacement

* fix(rebalance): isolate internal multipart uploads

* test(ecstore): adapt metadata mutation fixtures

* fix(rebalance): preserve transition metadata semantics

* refactor(ecstore): reuse internal metadata matcher

* Revert "refactor(ecstore): reuse internal metadata matcher"

This reverts commit c87ca0328f.

* refactor(rebalance): reuse data movement log constants

* fix(rebalance): isolate migration-owned state

* fix(rebalance): preserve pre-gate retry compatibility
This commit is contained in:
cxymds
2026-08-13 14:12:26 +08:00
committed by GitHub
parent 11eecdc888
commit e11fcfbd08
36 changed files with 6725 additions and 845 deletions
Download Patch File Download Diff File Expand all files Collapse all files
docs/operations/rolling-restart.md
+15 -3
View File
@@ -5,9 +5,10 @@ availability, what to expect when several nodes are down at once (sequential
cold start), and how to read the degraded-mode signals. Written for the
failure pattern reported in rustfs/rustfs#4304.
> Upgrading the binary or container image never changes the on-disk data
> format. Replacing the executable and restarting is safe; no migration step
> runs on startup.
> Upgrading the binary or container image does not change the on-disk data
> format unless an explicitly enabled feature documents a version floor.
> Replacing the executable and restarting does not run a migration step on
> startup.
> [!WARNING]
> The release that switches local SSE wrapped DEKs from the legacy
@@ -19,6 +20,17 @@ failure pattern reported in rustfs/rustfs#4304.
> and then resume traffic. Downgrading or rolling back after new encrypted
> objects are written is not supported.
> [!WARNING]
> `RUSTFS_DATA_MOVEMENT_PART_CHECKSUMS_WRITE` remains inactive unless
> `RUSTFS_DATA_MOVEMENT_PART_CHECKSUMS_FLEET_CONFIRMED` is also `true`. Enable
> both only after every node that can read or write object metadata supports
> the `part-checksums` sidecar and the fleet has adopted that version as its
> rollback floor. Leave either setting disabled throughout a mixed-version
> rolling upgrade. Once rebalance or decommission has migrated a legacy
> checksummed multipart object with both settings enabled, rolling back to an
> older build is not supported: older readers ignore the sidecar and can
> report an object checksum in place of the requested part checksum.
## TL;DR
- **Rolling restart (no downtime):** restart **one node at a time**, and wait