The session watchdog now selects its detection method per
platform. It previously probed kernel TCP state through a
Linux-only procfs path, so on macOS every healthy idle session
was killed within a minute, and on Windows the watchdog never
spawned at all, leaving wedged sessions with no cleanup. Linux
keeps its fast-kill watchdog unchanged. Other platforms get a
silence-only backstop that kills a session only at the
documented 30-minute idle ceiling.
The host-key loader now has a Windows arm. It loads OpenSSH
format host keys from the configured directory and logs a
one-time warning to restrict NTFS ACLs on the key directory,
the same operator-managed approach FTPS, WebDAV, KMS, and IAM
already use on Windows. Startup previously aborted with
UnsupportedPlatform because the Unix mode-bit permission check
has no Windows equivalent. tokio's io-uring feature is now
enabled only in Linux builds. io-uring is a Linux kernel
interface and enabling it unconditionally broke the Windows
build.
Co-authored-by: houseme <housemecn@gmail.com>
Gate scanner-triggered deep heal behind a short new-object cooldown and retry transient bitrot shard size mismatches during local verification.
Add targeted tests for cooldown mode selection and size-mismatch classification, and log when deep scans are downgraded or size-mismatch retries occur.
* fix(lifecycle): harden scanner ILM expiry accounting
* fix(scanner): gate ILM action accounting on enqueue
* fix(metrics): avoid scanner source work argument list
* fix(scanner): gate local ILM accounting on enqueue
---------
Co-authored-by: Henry Guo <marshawcoco@users.noreply.github.com>
Co-authored-by: 安正超 <anzhengchao@gmail.com>
Co-authored-by: houseme <housemecn@gmail.com>
* feat(scanner): expose checkpoint and source work status
* fix(scanner): count ignored checkpoints once per scan
---------
Co-authored-by: Henry Guo <marshawcoco@users.noreply.github.com>
* fix(scanner,data-usage): fix add() logic inversion and usize underflow in reduce_children_of
Bug #1: add() had inverted logic — returned early when children were present,
making the recursive child traversal dead code. Leaf compaction path was
completely non-functional. Fixed by recursing into children when present and
collecting leaves only at leaf nodes.
Bug #2: reduce_children_of used which underflows in debug
(panic) or release (wraps to usize::MAX, compacts entire cache). Fixed with
saturating_sub.
Both bugs existed identically in crates/scanner and crates/data-usage.
Added 5 tests covering add() subtree traversal, edge cases, and
reduce_children_of compaction behavior.
Closesrustfs/backlog#652
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
* fix: collect internal nodes as compaction candidates, not leaves
The previous fix for add() collected leaf nodes (children empty) as
compaction candidates. This is incorrect: total_children_rec returns 0
for leaves, so compacting them never decrements — the
compaction loop makes no progress.
Collect internal nodes (children non-empty) instead. Compacting an
internal node removes its subtree via delete_recursive, which actually
reduces the total child count. Leaf nodes are skipped since they have
no children to remove.
Updated tests to match the corrected semantics.
* refactor: rename leaves→candidates, add data-usage regression tests
- Rename misleading variable/parameter to in
add() and reduce_children_of() for both crates
- Add 4 regression tests in crates/data-usage covering add() candidate
selection and reduce_children_of() compaction + saturating_sub
* fix: address data usage compaction review feedback
---------
Co-authored-by: Claude Opus 4.8 <noreply@anthropic.com>
Co-authored-by: houseme <housemecn@gmail.com>
Clear persisted background heal active state after a deep scanner cycle completes so root heal status does not remain in progress between cycles.
Treat path-based stop for an already missing heal task as stopped while keeping client-token cancellation strict.
Co-authored-by: Henry Guo <marshawcoco@users.noreply.github.com>
* chore(deps): refresh workspace deps and linux fs_type gating
- refresh workspace dependency pins and lockfile updates
- remove now-unused crate dependency entries in multiple Cargo.toml files
- enable profiling export defaults in config and scripts/run.sh
- gate os::fs_type module/function/tests to Linux to avoid non-Linux dead_code warnings
* fix(utils): simplify fs_type linux gating
- keep fs_type module-level linux cfg in os::mod
- remove redundant linux cfg on get_fs_type and test module
* chore(deps): bump s3s git revision
- update workspace s3s dependency to rev 507e1312b211c3ddc214b03875d6fabd15d22ed5
- refresh Cargo.lock source entry for s3s
* chore(dev): allow mysql_async git source and env overrides
- allow mysql_async git source in deny.toml allow-git list
- make scripts/run.sh core env vars overrideable via existing shell env
* fix(utils): import get_fs_type in fs_type tests
- add explicit super::get_fs_type import in fs_type test module
- fix Linux E0425 unresolved function errors in unit tests
* chore(dev): tune run script observability defaults
- make profiling export env overrideable in scripts/run.sh
- set RUSTFS_OBS_SAMPLE_RATIO default from 2.0 to 1.0
- update allow-git review window comments in deny.toml
* test(obs): stabilize profiling env alias tests
安正超