Files
rustfs/docs/operations
Chris An 2381096448 fix(scanner): preserve restart progress and background admission (#9341)
* fix(scanner): retain unpublished bucket progress across leader changes

* test(scanner): verify remote completion reuse and publication floors

* fix(scanner): persist read-only checkpoint continuity

* test(scanner): avoid cloning the final activity fixture

* fix(scanner): resolve receipt owner before store startup

* fix(scanner): gate read-only handoff behind reader-first rollout

* test(scanner): accept consumed markers after heal startup

* test(scanner): run handoff rollback in the pinned release lane

* fix(scanner): preserve background admission during cache saves

* fix(scanner): admit remote cache publications as background I/O

* fix: handle disabled cgroup controllers in resource sampling

* fix(scanner): wait for namespace RPC preflight admission

* fix(scanner): align metric loops with current clippy

* fix: resolve Rust 1.100 workspace lint diagnostics

* fix(heal): simplify Linux test opt-in check

* chore(scanner): expose admission and allocator diagnostics

* fix(scanner): retry transient resource samples within one bracket

* fix(protos): retain cancelled RPC streams during response grace

* feat(diagnostics): track live Rust allocation bytes

* docs(diagnostics): explain unsafe allocator test scope

* fix(ecstore): reduce multipart recovery stack usage

* chore: refresh PR verification

* fix(protos): retain reset streams for the full cancelled burst

The branch's channel reset tests cancel 80 batches of 64 streams; main's 4096 cap spends the protocol-error budget on that burst, so keep the 8192 capacity under main's constant.

* fix(scanner): complete budget-stopped cycles with carried buckets pending

Align with main's partial-carry semantics: a work budget that stops a bucket whose certified value is carried completes the cycle with that bucket pending and the snapshot unconverged, instead of forcing the cycle incomplete.
2026-11-21 09:12:23 +08:00
..