Compare commits

..

1 Commits

Author SHA1 Message Date
overtrue 528e22c150 test(e2e): fail incomplete conditional PUT races 2026-08-23 04:47:38 +08:00
15 changed files with 123 additions and 310 deletions
+1 -2
View File
@@ -30,8 +30,7 @@ make build-docker BUILD_OS=ubuntu22.04
- Crate membership: `Cargo.toml` `[workspace].members`
- Architecture, layering, crate map: [ARCHITECTURE.md](ARCHITECTURE.md)
- Migration guardrails & readiness contracts: [docs/architecture/](docs/architecture/README.md)
- CI workflow steps: `.github/workflows/`; event, timeout, and required-status
matrix: [docs/testing/ci-gates.md](docs/testing/ci-gates.md)
- CI gates: `.github/workflows/ci.yml` (source of truth; never copy its steps into docs)
- Test-layer taxonomy, per-layer entry commands, serial/nextest rules, flake
policy: [docs/testing/README.md](docs/testing/README.md)
- Tier/ILM transition debugging (xl.meta inspection, versionId tracing):
-2
View File
@@ -70,8 +70,6 @@ make pre-pr
> For the full test-layer taxonomy (unit / ecstore black-box / e2e / s3s-e2e / S3 compatibility / chaos / fuzz / bench), each layer's entry command, the naming conventions the migration gate depends on, and the serial/nextest rules, see [docs/testing/README.md](docs/testing/README.md).
> For the event, timeout, required-status, and local reproduction matrix, see [docs/testing/ci-gates.md](docs/testing/ci-gates.md).
### 🔒 Automated Pre-commit Hooks
#### What `make pre-commit` and `make pre-pr` actually run
Generated
+27 -22
View File
@@ -1858,9 +1858,9 @@ dependencies = [
[[package]]
name = "cc"
version = "1.4.4"
version = "1.4.3"
source = "registry+https://github.com/rust-lang/crates.io-index"
checksum = "0ad534f4357a5264cce5019c989cf66a4f0dc4e0d1b1d15f8aacec0ff7360273"
checksum = "509591b7bcd67f4ef775afad7662703b4935daaa6ec0e5605cfb1090b32a2b6d"
dependencies = [
"find-msvc-tools",
"jobserver",
@@ -2522,6 +2522,12 @@ dependencies = [
"subtle",
]
[[package]]
name = "cty"
version = "0.2.2"
source = "registry+https://github.com/rust-lang/crates.io-index"
checksum = "b365fabc795046672053e29c954733ec3b05e4be654ab130fe8f1f94d7051f35"
[[package]]
name = "curve25519-dalek"
version = "4.1.3"
@@ -5982,6 +5988,15 @@ version = "0.2.16"
source = "registry+https://github.com/rust-lang/crates.io-index"
checksum = "b6d2cec3eae94f9f509c767b45932f1ada8350c4bdb85af2fcab4a3c14807981"
[[package]]
name = "libmimalloc-sys"
version = "0.1.49"
source = "git+https://github.com/xonatius/mimalloc_rust.git?rev=6d4c41bb10c6d9da1d1b6f07b38c4cc051667f11#6d4c41bb10c6d9da1d1b6f07b38c4cc051667f11"
dependencies = [
"cc",
"cty",
]
[[package]]
name = "libredox"
version = "0.1.20"
@@ -6382,6 +6397,14 @@ dependencies = [
"synstructure 0.13.2",
]
[[package]]
name = "mimalloc"
version = "0.1.52"
source = "git+https://github.com/xonatius/mimalloc_rust.git?rev=6d4c41bb10c6d9da1d1b6f07b38c4cc051667f11#6d4c41bb10c6d9da1d1b6f07b38c4cc051667f11"
dependencies = [
"libmimalloc-sys",
]
[[package]]
name = "mime"
version = "0.3.17"
@@ -9139,11 +9162,13 @@ dependencies = [
"insta",
"jiff",
"libc",
"libmimalloc-sys",
"libsystemd",
"matchit 0.9.2",
"md-5 0.11.0",
"metrics",
"metrics-util",
"mimalloc",
"mime_guess",
"opentelemetry",
"opentelemetry_sdk",
@@ -9179,8 +9204,6 @@ dependencies = [
"rustfs-lock",
"rustfs-log-analyzer",
"rustfs-madmin",
"rustfs-mimalloc",
"rustfs-mimalloc-sys",
"rustfs-notify",
"rustfs-object-capacity",
"rustfs-object-data-cache",
@@ -9852,24 +9875,6 @@ dependencies = [
"tokio",
]
[[package]]
name = "rustfs-mimalloc"
version = "0.5.0"
source = "registry+https://github.com/rust-lang/crates.io-index"
checksum = "a406f4aa07084301d485beec873af6dccc8e3f8762da244743df92038b1db1a6"
dependencies = [
"rustfs-mimalloc-sys",
]
[[package]]
name = "rustfs-mimalloc-sys"
version = "0.5.0"
source = "registry+https://github.com/rust-lang/crates.io-index"
checksum = "c3051b819175f58445d4c369a72f0ab88149f3885ba8bea2aff3be01f53fe7cd"
dependencies = [
"cc",
]
[[package]]
name = "rustfs-notify"
version = "1.0.0-rc.3"
+2 -2
View File
@@ -350,8 +350,8 @@ russh-sftp = "2.4.0"
dav-server = "0.11.0"
# Performance Analysis and Memory Profiling
rustfs-mimalloc = { version = "0.5.0" }
rustfs-mimalloc-sys = { version = "0.5.0" }
mimalloc = { version = "0.1.52", git = "https://github.com/xonatius/mimalloc_rust.git", rev = "6d4c41bb10c6d9da1d1b6f07b38c4cc051667f11" }
libmimalloc-sys = { version = "0.1.49", git = "https://github.com/xonatius/mimalloc_rust.git", rev = "6d4c41bb10c6d9da1d1b6f07b38c4cc051667f11", features = ["extended"] }
hotpath = { version = "0.23.3", default-features = false }
# Snapshot testing for output format regression detection
insta = { version = "1.48" }
+21 -17
View File
@@ -22,10 +22,9 @@ use tracing::{info, warn};
const BUCKET: &str = "conditional-put-race-bucket";
async fn cleanup_object(client: &Client, key: &str) {
if let Err(e) = client.delete_object().bucket(BUCKET).key(key).send().await {
warn!("Failed to delete object '{}' from bucket '{}' during cleanup: {:?}", key, BUCKET, e);
}
async fn cleanup_object(client: &Client, key: &str) -> Result<(), Box<dyn std::error::Error + Send + Sync>> {
client.delete_object().bucket(BUCKET).key(key).send().await?;
Ok(())
}
async fn conditional_put(
@@ -71,14 +70,13 @@ async fn run_race_iteration(
test_key: &str,
iteration: usize,
) -> Result<usize, Box<dyn std::error::Error + Send + Sync>> {
cleanup_object(&clients[0], test_key).await;
cleanup_object(&clients[0], test_key).await?;
tokio::time::sleep(tokio::time::Duration::from_millis(100)).await;
let head_result = clients[0].head_object().bucket(BUCKET).key(test_key).send().await;
if head_result.is_ok() {
warn!("Warning: Object still exists after cleanup, skipping iteration {}", iteration);
return Ok(0);
match clients[0].head_object().bucket(BUCKET).key(test_key).send().await {
Ok(_) => return Err(format!("object still exists after cleanup in iteration {iteration}").into()),
Err(error) if error.as_service_error().is_some_and(|error| error.is_not_found()) => {}
Err(error) => return Err(format!("failed to verify cleanup in iteration {iteration}: {error:?}").into()),
}
info!("\n=== Iteration {} ===", iteration);
@@ -120,14 +118,16 @@ async fn run_race_iteration(
info!("Result: {} out of {} succeeded", success_count, clients.len());
if had_error {
return Err("one or more conditional PUTs failed unexpectedly".into());
}
if success_count > 1 {
info!(">>> RACE CONDITION DETECTED!");
} else if success_count == 1 {
info!(">>> Correct behavior: exactly 1 writer succeeded.");
} else if had_error {
return Err("all conditional PUTs failed (e.g. cluster/bucket not ready)".into());
} else {
info!(">>> Unexpected: no writers succeeded.");
return Err("no conditional PUT succeeded".into());
}
Ok(success_count)
@@ -167,7 +167,7 @@ async fn test_conditional_put_race_cluster() -> Result<(), Box<dyn std::error::E
}
}
cleanup_object(&clients[0], &test_key).await;
cleanup_object(&clients[0], &test_key).await?;
tokio::time::sleep(tokio::time::Duration::from_millis(50)).await;
}
@@ -177,7 +177,7 @@ async fn test_conditional_put_race_cluster() -> Result<(), Box<dyn std::error::E
info!("Total iterations: {}", iterations);
info!("Correct (1 winner): {}", correct_count);
info!("Race conditions: {}", races_detected);
info!("Errors (skipped): {}", error_count);
info!("Failed iterations: {}", error_count);
assert_eq!(races_detected, 0, "Race conditions detected: {}/{}", races_detected, iterations);
assert_eq!(
@@ -185,6 +185,10 @@ async fn test_conditional_put_race_cluster() -> Result<(), Box<dyn std::error::E
"{} iteration(s) failed due to errors (e.g. cluster not ready)",
error_count
);
assert_eq!(
correct_count, iterations,
"only {correct_count}/{iterations} iterations observed exactly one winner"
);
Ok(())
}
@@ -201,7 +205,7 @@ async fn test_conditional_put_basic_cluster() -> Result<(), Box<dyn std::error::
let client = cluster.create_s3_client(0)?;
let test_key = "basic-conditional-put";
cleanup_object(&client, test_key).await;
cleanup_object(&client, test_key).await?;
let result = client
.put_object()
@@ -233,6 +237,6 @@ async fn test_conditional_put_basic_cluster() -> Result<(), Box<dyn std::error::
assert_eq!(code, "PreconditionFailed");
}
cleanup_object(&client, test_key).await;
cleanup_object(&client, test_key).await?;
Ok(())
}
+14 -1
View File
@@ -2124,13 +2124,26 @@ impl SetDisks {
let put_object_size = known_put_object_storage_size(data.size());
let shard_file_size_raw = erasure.shard_file_size(put_object_size);
let is_inline_buffer = storage_class_config.should_inline(shard_file_size_raw, erasure.data_shards, opts.versioned);
let is_inline_buffer =
storage_class_config.should_inline(shard_file_size_raw, erasure.data_shards, opts.versioned);
let collect_stage_timing = rustfs_io_metrics::put_stage_metrics_enabled() || issue3031_diag_enabled();
let shard_file_size = shard_file_size_raw;
let shard_size = erasure.shard_size();
let write_path = classify_put_write_path(is_inline_buffer, put_object_size, fi.erasure.block_size);
let direct_inline_commit = matches!(write_path, SmallWritePath::Inline);
{
use std::io::Write;
let msg = format!(
"INLINE_DEBUG: bucket={} obj={} size={} shard_fs={} ds={} bs={} inline={} direct={} path={} iblock={} ver={}\n",
bucket, object, put_object_size, shard_file_size_raw, erasure.data_shards, fi.erasure.block_size,
is_inline_buffer, direct_inline_commit, write_path.metric_label(), storage_class_config.inline_block(), opts.versioned
);
if let Ok(mut f) = std::fs::OpenOptions::new().create(true).append(true).open("/tmp/rustfs_inline_debug.log") {
let _ = f.write_all(msg.as_bytes());
}
let _ = std::io::stderr().write_all(msg.as_bytes());
}
rustfs_io_metrics::record_put_object_path(write_path.metric_label());
let writer_setup_stage_start = collect_stage_timing.then(Instant::now);
let (mut writers, errors) = if direct_inline_commit {
+1 -1
View File
@@ -3194,7 +3194,7 @@ impl ECStore {
// Default return value
let mut del_objects = vec![DeletedObject::default(); objects.len()];
let accounting = vec![None; objects.len()];
let mut accounting = vec![None; objects.len()];
let mut del_errs = Vec::with_capacity(objects.len());
for _ in 0..objects.len() {
@@ -271,7 +271,7 @@ pub(super) fn resolve_latest_object_info_candidates(
.filter(|candidate| latest_candidate_mod_time(candidate) == Some(latest_mod_time))
.collect::<Vec<_>>();
latest_candidates.sort_by_key(|candidate| std::cmp::Reverse(candidate.idx));
latest_candidates.sort_by(|left, right| right.idx.cmp(&left.idx));
let Some(winner) = latest_candidates.first() else {
return Err(Error::ErasureReadQuorum);
+3
View File
@@ -43,6 +43,9 @@ allow-git = [
# RustFS fork carrying presigned expiry and constant-time authentication fixes.
# owner: rustfs-maintainers review: 2026-10
"https://github.com/rustfs/s3s.git",
# MiMalloc fork pinned for hotpath allocation counting support.
# owner: houseme review: 2026-10
"https://github.com/xonatius/mimalloc_rust.git",
]
[bans]
-149
View File
@@ -1,149 +0,0 @@
# CI gate matrix
This file is the source of truth for which validation runs on each event, its
configured wall-clock budget, and whether it can block a merge. Test taxonomy,
naming, and nextest serialization rules remain in [README.md](README.md); e2e
membership and counts remain in
[e2e-suite-inventory.md](e2e-suite-inventory.md).
The distinction between **required** and **report-only** is load-bearing:
a failing job blocks a merge only when its exact check name is present in the
live `main` ruleset. A workflow name, a `merge_group` trigger, or a red PR check
does not make a job required by itself.
## Required merge checks
The live `main` ruleset (`6436880`) currently requires exactly these contexts:
| Required context | Producer | Validation |
|---|---|---|
| `CLA Check` | `.github/workflows/cla.yml` | Contributor agreement |
| `Quick Checks` | `.github/workflows/ci.yml` | Formatting and repository guard scripts |
| `Test and Lint` | `.github/workflows/ci.yml` | Clippy, workspace nextest excluding `e2e_test`, doctests, and migration proofs |
For pull requests limited to the paths excluded by the main CI workflow,
`.github/workflows/ci-docs-only.yml` reports `Quick Checks` and
`Test and Lint` under the same names. It runs the real quick checks and the
planning-document guard; it does not claim that Rust compilation or runtime
tests ran. Despite the workflow name, these paths also include selected deploy,
workflow, and lock files.
Verify the live rule rather than trusting this snapshot before changing merge
policy:
```bash
gh api repos/rustfs/rustfs/rulesets/6436880 \
--jq '.rules[] | select(.type == "required_status_checks") | .parameters'
```
The ruleset currently has `strict_required_status_checks_policy=false`.
`Continuous Integration` accepts `merge_group` events and runs `e2e-full` for
them, but `End-to-End Tests (full merge gate)` is not currently a required
context. Therefore the repository is prepared to test a merge-queue SHA, but
the workflow alone does not prove that every merge passed that lane.
## Pull request and merge matrix
Budgets below are job `timeout-minutes`, not typical runtimes. “Report-only”
means the result is visible and actionable but is not in the live required
context list.
| Event | Validation | Budget | Merge status | Reproduction |
|---|---|---:|---|---|
| PR, non-doc change | `Quick Checks` | 10 min | Required | `make pre-commit` (broader local umbrella) |
| PR, non-doc change | `Test and Lint` | 90 min | Required | `cargo nextest run --profile ci --all --exclude e2e_test` |
| PR, non-doc change | `Typos` | 10 min | Report-only | `typos` |
| PR, non-doc change | `ILM Integration (serial)` | 90 min | Report-only | Use the exact command in `.github/workflows/ci.yml` |
| PR, non-doc change | rio-v2 / swift / sftp test-and-lint variants | 90 min each | Report-only | `cargo nextest run` with the workflow's feature set |
| PR, non-doc change | `Build RustFS Debug Binary` | 30 min | Report-only; prerequisite for black-box lanes | `cargo build -p rustfs --bins` |
| PR, non-doc change | `io_uring Integration (real)` | 30 min | Report-only | `cargo test -p rustfs-ecstore --lib uring_ -- --test-threads=1 --nocapture` |
| PR, non-doc change | `End-to-End Tests` (`e2e-smoke` plus `s3s-e2e`) | 30 min | Report-only | `cargo nextest run --profile e2e-smoke -p e2e_test`; then `./scripts/e2e-run.sh ./target/debug/rustfs <data-dir>` |
| PR, non-doc change | `S3 Implemented Tests` | 60 min | Report-only | Build `rustfs`, then run `scripts/s3-tests/run.sh` with `DEPLOY_MODE=binary`, `TEST_MODE=single`, and `MAXFAIL=0` |
| PR, non-doc change | `S3 Lifecycle Behavior Tests` | 30 min | Report-only | Use the accelerated scanner environment in `.github/workflows/ci.yml` with `scripts/s3-tests/run.sh` |
| PR touching dependency or workflow inputs | Cargo Deny / Workflow Pin Report / Dependency Review | 20 / 5 / 30 min | Report-only | `cargo deny check`; `scripts/security/check_workflow_pins.sh` |
| PR touching architecture rules or architecture docs | `Architecture Migration Rules` | 10 min | Report-only | `scripts/check_architecture_migration_rules.sh` |
| PR touching Nix or workspace manifests | `Nix Build & Check` | 60 min | Report-only | `nix flake check` |
| PR limited to main-CI-excluded paths | companion `Quick Checks` and `Test and Lint` | 10 min each | Required | `git diff --check`; `make doc-paths-check` when documentation paths changed |
| `merge_group` | Standard CI plus `e2e-full` | 55 min for `e2e-full` | Standard required contexts only; `e2e-full` report-only | `cargo nextest run --profile e2e-full -p e2e_test` |
| Push to `main` | Standard CI plus `e2e-full` | 55 min for `e2e-full` | Post-merge detection | Same as `merge_group` |
| PR touching fuzz inputs or harness paths | Build plus five 60-second fuzz smoke targets | 60 min build; 30 min per target | Report-only | `MAX_TOTAL_TIME=60 ./scripts/fuzz/run.sh` |
| PR touching selected ecstore disk/format paths | `Rename Safety` on Windows | 60 min | Report-only | Run the four `cargo test -p rustfs-ecstore --lib <filter>` commands in `windows-filesystem.yml` on Windows |
The authoritative e2e filters live in `.config/nextest.toml`; extend a profile
instead of adding a second ad-hoc selector. Before a profile runs,
`scripts/check_test_wiring.py` compares its exact membership to the committed
digest so a silent test drop fails closed.
## Scheduled and manual validation
Scheduled lanes are independent fault domains. They do not block a pull
request, but their workflow-local gate can fail the run and scheduled failures
are routed to the shared failure-issue action. The scheduled-validation
watchdog and freshness workflow separately detect incomplete runs and missing
schedules.
| Cadence (UTC unless noted) | Workflow / validation | Budget | Verdict and artifacts | Reproduction |
|---|---|---:|---|---|
| Daily 02:17 | Fuzz: five nightly corpus targets | 60 min build; 60 min per target | Gate; corpus/crash artifacts, scheduled failure alert | `MAX_TOTAL_TIME=<seconds> ./scripts/fuzz/run.sh` |
| Daily 03:17 | MinIO interop (EC + SSE read parity) | 40 min | Gate; scheduled failure alert | Dispatch `minio-interop.yml` or follow its pinned Docker fixture steps |
| Daily 04:29 | Replication / cluster-fault / protocol e2e | 45 / 90 / 90 min | Three independent gates; JUnit, membership, and server logs | `cargo nextest run --profile e2e-repl-nightly -p e2e_test`; `--profile e2e-nightly`; `-j 1 --profile e2e-protocols` |
| Daily 06:31 | Warp performance A/B | 180 min | Regression budget gate; A/B summaries and server logs | `bash scripts/run_hotpath_warp_abba.sh --help` |
| Daily 00:07 Asia/Shanghai (16:07 UTC previous day) | Nightly GNU build and Vault lanes | 150 / 90 / 60 min | Build, live Vault, and HA failover gates | Use the commands and pinned Vault images in `nightly-gnu.yml` |
| Daily 03:23 | Security Audit | 20 / 5 min, plus 30 min on PR dependency review | Cargo Deny and workflow-pin gates; scheduled failure alert | `cargo deny check`; `scripts/security/check_workflow_pins.sh` |
| Daily 23:47 | Scheduled Validation Freshness | 10 min | Fails when a critical schedule was never created or is stale | Dispatch `scheduled-validation-freshness.yml` |
| Sunday 00:11 | Full `Continuous Integration` matrix | Per-job budgets above | Weekly variant coverage, including dormant rio-v2 binary/e2e lanes | Dispatch `ci.yml` |
| Sunday 01:13 | Seven-platform build matrix | 150 min per platform | Build/package integrity; scheduled failure alert | Dispatch `build.yml` with an exact platform set |
| Sunday 02:19 | Ceph s3-tests full sweep: single and real four-node, four shards each | 180 min per shard | Compatibility gate; report, JUnit, exact node IDs, and server logs | `scripts/s3-tests/run.sh` against an existing single or distributed target |
| Sunday 06:41 | Mint | 120 min | **Report-only by design**; per-suite PASS/FAIL/NA and raw `log.json` | Reproduce the pinned Docker sequence in `mint.yml` or dispatch it |
| Sunday 07:43 | Workspace line coverage | 120 min | Report-only trend; lcov and JSON retained 90 days | `make coverage` |
| Monthly, day 1 06:37 | Runner Hygiene | 15 min | Validates runner ephemerality; scheduled failure alert | Dispatch `runner-hygiene.yml` |
Manual `workflow_dispatch` exists for the scheduled workflows above. Manual
runs are debugging evidence and intentionally do not open scheduled-failure
issues. A manual performance run may explicitly allow a known regression; that
override must not be treated as an ordinary passing baseline.
## Release validation
Release validation is post-merge and tag-driven; it does not substitute for a
pull-request gate.
| Event | Validation | Budget | Result |
|---|---|---:|---|
| Push to `main` or weekly schedule | `Build and Release` platform matrix | 150 min per platform | Build artifacts for all selected targets; no release publication on a main push |
| Valid release or preview tag | `Build and Release` plus asset checks | 150 min per platform | Draft release, checksummed assets, and publish step |
| Successful non-preview release-tag build | Docker image build and image scan | 60 min build; 30 min scan | Multi-architecture images plus vulnerability report |
| Successful release-tag build | DEB/RPM packaging | 30 min per architecture | Packages and checksum files uploaded to the release |
| Successful non-preview release-tag build | Helm template test and package | 30 min build; 30 min publish | Versioned chart and repository index |
Use an exact preview tag for end-to-end release rehearsal. Manual dispatches
are backfill/debug paths and do not prove the automatic `workflow_run` chain.
## Evidence requirements
A green check is useful only when it proves the intended behavior ran:
- Record the exact commit SHA and run URL.
- Separate product failure from runner prerequisites, service readiness, and
cancellation. Repair the precondition, then rerun the exact workload.
- Preserve membership manifests, JUnit, raw compatibility logs, seeds, and
server logs where the workflow provides them.
- For a bug fix or a new fault checker, provide sensitivity evidence: the old
behavior or an intentional mutation must fail the new oracle, and the fixed
behavior must pass it.
- Never promote a report-only lane to required from one green run. Require at
least 14 days and 30 representative pull requests with at least 99% complete
execution, then update the ruleset and this table together.
## Change checklist
Update this file in the same pull request when any of these change:
- workflow triggers, job names, timeouts, or nextest profile ownership;
- required status contexts or strict/merge-queue policy;
- scheduled cadence, alert routing, artifact contract, or local reproduction;
- report-only versus gating semantics.
Do not copy per-module test counts here. Update
[e2e-suite-inventory.md](e2e-suite-inventory.md) and its enforced membership
digest instead.
+2 -2
View File
@@ -336,13 +336,13 @@ opentelemetry = { workspace = true }
tracing-opentelemetry = { workspace = true }
# Data structures
hashbrown = { workspace = true, features = ["serde", "rayon"] }
rustfs-mimalloc = { workspace = true }
mimalloc = { workspace = true }
[target.'cfg(target_os = "linux")'.dependencies]
libsystemd.workspace = true
[target.'cfg(not(target_os = "windows"))'.dependencies]
rustfs-mimalloc-sys.workspace = true
libmimalloc-sys.workspace = true
[dev-dependencies]
uuid = { workspace = true, features = ["v4", "v5", "fast-rng", "macro-diagnostics"] }
-81
View File
@@ -70,12 +70,6 @@ const SITE_REPLICATION_EDIT_ROUTE: &str = "/rustfs/admin/v3/site-replication/edi
const SITE_REPLICATION_RESYNC_ROUTE: &str = "/rustfs/admin/v3/site-replication/resync/op";
const SITE_REPLICATION_REPAIR_ROUTE: &str = "/rustfs/admin/v3/site-replication/repair";
const SITE_REPLICATION_REPAIR_STATUS_ROUTE: &str = "/rustfs/admin/v3/site-replication/repair/status";
const IAM_POLICY_ATTACH_ROUTE: &str = "/rustfs/admin/v3/idp/builtin/policy/attach";
const IAM_POLICY_DETACH_ROUTE: &str = "/rustfs/admin/v3/idp/builtin/policy/detach";
const IAM_POLICY_ENTITIES_ROUTE: &str = "/rustfs/admin/v3/idp/builtin/policy-entities";
const IAM_ACCESS_KEYS_BULK_ROUTE: &str = "/rustfs/admin/v3/list-access-keys-bulk";
const IAM_ACCESS_KEYS_BULK_LDAP_ROUTE: &str = "/rustfs/admin/v3/idp/ldap/list-access-keys-bulk";
const IAM_ACCESS_KEYS_BULK_OPENID_ROUTE: &str = "/rustfs/admin/v3/idp/openid/list-access-keys-bulk";
macro_rules! log_system_request_rejected {
($operation:expr, $reason:expr) => {
@@ -667,24 +661,9 @@ pub struct RuntimeCapabilitiesSummary {
pub manual_transition_jobs: CapabilityStatus,
}
/// One named admin capability advertised to management clients
/// (rustfs/backlog#1900). `name` is a cross-repo wire contract: the rc
/// client gates commands on these exact strings (see rustfs/cli
/// `IAM_POLICY_DETACH_CAPABILITY` etc.), so entries may be added but
/// existing names must never be renamed or removed.
#[derive(Debug, Clone, PartialEq, Eq, Serialize)]
pub struct AdvertisedAdminCapability {
pub name: &'static str,
pub status: CapabilityStatus,
}
#[derive(Debug, Clone, PartialEq, Eq, Serialize)]
pub struct RuntimeCapabilitiesResponse {
pub summary: RuntimeCapabilitiesSummary,
/// Additive field: absent in responses from older servers, so clients
/// must treat a missing list as "no dynamic advertisement" and fall
/// back to their pinned per-version contract.
pub advertised: Vec<AdvertisedAdminCapability>,
pub replication: ReplicationCapabilities,
pub manual_transition_jobs: ManualTransitionJobCapabilities,
pub diagnostic_probes: DiagnosticProbeCapabilities,
@@ -1007,7 +986,6 @@ pub(crate) async fn build_runtime_capabilities_response()
Ok(RuntimeCapabilitiesResponse {
summary,
advertised: advertised_admin_capabilities(),
replication: ReplicationCapabilities::current(),
manual_transition_jobs: ManualTransitionJobCapabilities::current(),
diagnostic_probes: DiagnosticProbeCapabilities::current(),
@@ -1099,23 +1077,6 @@ fn admin_route_capability(method: HttpMethod, path: &str) -> CapabilityStatus {
admin_route_capability_from_inventory(method, path, ADMIN_ROUTE_POLICY_SPECS, DEFERRED_ADMIN_ROUTE_POLICIES)
}
fn advertised_admin_capabilities() -> Vec<AdvertisedAdminCapability> {
[
("admin.iam.policy-attach", HttpMethod::Post, IAM_POLICY_ATTACH_ROUTE),
("admin.iam.policy-detach", HttpMethod::Post, IAM_POLICY_DETACH_ROUTE),
("admin.iam.policy-entities", HttpMethod::Get, IAM_POLICY_ENTITIES_ROUTE),
("admin.iam.access-keys-bulk", HttpMethod::Get, IAM_ACCESS_KEYS_BULK_ROUTE),
("admin.iam.access-keys-bulk.ldap", HttpMethod::Get, IAM_ACCESS_KEYS_BULK_LDAP_ROUTE),
("admin.iam.access-keys-bulk.openid", HttpMethod::Get, IAM_ACCESS_KEYS_BULK_OPENID_ROUTE),
]
.into_iter()
.map(|(name, method, route)| AdvertisedAdminCapability {
name,
status: admin_route_capability(method, route),
})
.collect()
}
fn admin_route_capability_from_inventory(
method: HttpMethod,
path: &str,
@@ -1278,48 +1239,6 @@ mod tests {
);
}
/// Wire-contract pin (rustfs/backlog#1900): the rc client keys its
/// command gates on these exact capability names, and parses each
/// entry as `{name, status: {state, reason?}}`. Renaming or dropping
/// a name silently disables the corresponding rc command.
#[tokio::test]
async fn runtime_capabilities_response_advertises_iam_capabilities() {
let response = build_runtime_capabilities_response()
.await
.expect("runtime capabilities response should build");
let expected_supported = [
"admin.iam.policy-attach",
"admin.iam.policy-detach",
"admin.iam.policy-entities",
"admin.iam.access-keys-bulk",
"admin.iam.access-keys-bulk.ldap",
"admin.iam.access-keys-bulk.openid",
];
for name in expected_supported {
let entry = response
.advertised
.iter()
.find(|capability| capability.name == name)
.unwrap_or_else(|| panic!("{name} must be advertised"));
assert_eq!(entry.status.state, CapabilityState::Supported, "{name} must be supported");
}
let mut names: Vec<&str> = response.advertised.iter().map(|capability| capability.name).collect();
let total = names.len();
names.sort_unstable();
names.dedup();
assert_eq!(names.len(), total, "advertised capability names must be unique");
let serialized = serde_json::to_value(&response).expect("response should serialize");
let advertised = serialized["advertised"].as_array().expect("advertised must be an array");
let detach = advertised
.iter()
.find(|entry| entry["name"] == "admin.iam.policy-detach")
.expect("serialized detach entry must exist");
assert_eq!(detach["status"]["state"], "supported");
}
#[tokio::test]
async fn runtime_capabilities_response_reports_missing_topology_before_storage_init() {
let response = build_runtime_capabilities_response()
+7 -1
View File
@@ -369,8 +369,14 @@ pub fn allocator_reclaim_controller_snapshot(ctx: &CancellationToken) -> Allocat
}
#[cfg(not(target_os = "windows"))]
#[allow(unsafe_code)]
fn collect_allocator_memory(force: bool) -> Result<(), String> {
rustfs_mimalloc::MiMalloc::collect(force);
// SAFETY: `mi_collect` is provided by the active global allocator backend
// on this target family. It is explicitly intended to reclaim retained
// pages/segments and does not require additional invariants from the caller.
unsafe {
libmimalloc_sys::mi_collect(force);
}
Ok(())
}
+8 -10
View File
@@ -26,22 +26,22 @@ struct MiMallocAllocator;
unsafe impl GlobalAlloc for MiMallocAllocator {
unsafe fn alloc(&self, layout: Layout) -> *mut u8 {
// SAFETY: the caller upholds GlobalAlloc's contract for layout.
unsafe { rustfs_mimalloc::MiMalloc.alloc(layout) }
unsafe { mimalloc::MiMalloc.alloc(layout) }
}
unsafe fn alloc_zeroed(&self, layout: Layout) -> *mut u8 {
// SAFETY: the caller upholds GlobalAlloc's contract for layout.
unsafe { rustfs_mimalloc::MiMalloc.alloc_zeroed(layout) }
unsafe { mimalloc::MiMalloc.alloc_zeroed(layout) }
}
unsafe fn dealloc(&self, ptr: *mut u8, layout: Layout) {
// SAFETY: ptr and layout came from this allocator and are forwarded unchanged.
unsafe { rustfs_mimalloc::MiMalloc.dealloc(ptr, layout) }
unsafe { mimalloc::MiMalloc.dealloc(ptr, layout) }
}
unsafe fn realloc(&self, ptr: *mut u8, layout: Layout, new_size: usize) -> *mut u8 {
// SAFETY: ptr and layout came from this allocator and are forwarded unchanged.
unsafe { rustfs_mimalloc::MiMalloc.realloc(ptr, layout, new_size) }
unsafe { mimalloc::MiMalloc.realloc(ptr, layout, new_size) }
}
}
@@ -51,7 +51,7 @@ static GLOBAL: hotpath::CountingAllocator<MiMallocAllocator> = hotpath::Counting
#[cfg(not(all(feature = "hotpath", feature = "hotpath-alloc")))]
#[global_allocator]
static GLOBAL: rustfs_mimalloc::MiMalloc = rustfs_mimalloc::MiMalloc;
static GLOBAL: mimalloc::MiMalloc = mimalloc::MiMalloc;
fn main() {
let _hotpath_guard = hotpath::HotpathGuardBuilder::new("main").build();
@@ -71,9 +71,8 @@ mod tests {
allocation.extend_from_slice(&[7_u8; 64]);
assert_eq!(allocation.len(), 64);
let heap = rustfs_mimalloc::heap::Heap::main();
// SAFETY: the live Vec pointer is valid to inspect for heap ownership.
assert!(unsafe { heap.contains(allocation.as_ptr()) });
assert!(unsafe { libmimalloc_sys::mi_is_in_heap_region(allocation.as_ptr().cast()) });
}
#[test]
@@ -86,13 +85,12 @@ mod tests {
let layout = Layout::from_size_align(32, 8).expect("valid test allocation layout");
let grown_layout = Layout::from_size_align(64, 8).expect("valid grown test allocation layout");
let allocator = super::MiMallocAllocator;
let heap = rustfs_mimalloc::heap::Heap::main();
// SAFETY: The pointer is checked for null before use and later released
// through the same allocator with the corresponding layout.
let ptr = unsafe { allocator.alloc_zeroed(layout) };
assert!(!ptr.is_null());
assert!(unsafe { heap.contains(ptr) });
assert!(unsafe { libmimalloc_sys::mi_is_in_heap_region(ptr.cast()) });
assert!(unsafe { std::slice::from_raw_parts(ptr, 32).iter().all(|byte| *byte == 0) });
// SAFETY: `ptr` was allocated by `allocator` with `layout`; on failure
@@ -104,7 +102,7 @@ mod tests {
panic!("mimalloc realloc failed in allocator smoke test");
}
assert!(unsafe { heap.contains(grown_ptr) });
assert!(unsafe { libmimalloc_sys::mi_is_in_heap_region(grown_ptr.cast()) });
// SAFETY: `grown_ptr` was reallocated by `allocator` and is released
// with the matching grown layout.
unsafe { allocator.dealloc(grown_ptr, grown_layout) };
+36 -19
View File
@@ -17,7 +17,10 @@ use rustfs_io_metrics::{
record_cpu_usage, record_memory_usage, record_process_memory_split,
};
use serde::Serialize;
#[cfg(any(test, not(target_os = "windows")))]
use serde_json::Value;
#[cfg(not(target_os = "windows"))]
use std::ffi::CStr;
use std::path::Path;
use std::sync::{Arc, Mutex, OnceLock};
use std::time::Duration;
@@ -228,18 +231,7 @@ fn read_cgroup_memory_snapshot() -> Option<CgroupMemorySnapshot> {
read_cgroup_v2().or_else(read_cgroup_v1)
}
fn read_allocator_memory_snapshot() -> Option<AllocatorMemorySnapshot> {
let json = rustfs_mimalloc::MiMalloc::stats_json();
if json.is_empty() {
return None;
}
let observation = parse_mimalloc_stats_json(&json)?;
Some(AllocatorMemorySnapshot {
backend: crate::allocator_reclaim::allocator_backend(),
observation,
})
}
#[cfg(any(test, not(target_os = "windows")))]
fn numeric_json_value(value: &Value) -> Option<u64> {
match value {
Value::Number(number) => number
@@ -250,6 +242,7 @@ fn numeric_json_value(value: &Value) -> Option<u64> {
}
}
#[cfg(any(test, not(target_os = "windows")))]
fn numeric_json_field(value: &Value, field: &str) -> Option<u64> {
match value {
Value::Object(fields) => fields
@@ -261,6 +254,7 @@ fn numeric_json_field(value: &Value, field: &str) -> Option<u64> {
}
}
#[cfg(any(test, not(target_os = "windows")))]
fn mimalloc_stat_field(value: &Value, metric: &str, field: &str) -> Option<u64> {
match value {
Value::Object(fields) => {
@@ -277,10 +271,12 @@ fn mimalloc_stat_field(value: &Value, metric: &str, field: &str) -> Option<u64>
}
}
#[cfg(any(test, not(target_os = "windows")))]
fn mimalloc_stat_current(value: &Value, metric: &str) -> Option<u64> {
mimalloc_stat_field(value, metric, "current")
}
#[cfg(any(test, not(target_os = "windows")))]
fn mimalloc_stat_sum(value: &Value, metrics: &[&str], field: &str) -> Option<u64> {
metrics
.iter()
@@ -289,6 +285,7 @@ fn mimalloc_stat_sum(value: &Value, metrics: &[&str], field: &str) -> Option<u64
.filter(|value| *value > 0)
}
#[cfg(any(test, not(target_os = "windows")))]
fn parse_mimalloc_stats_json(stats_json: &str) -> Option<AllocatorMemoryObservation> {
let value = serde_json::from_str::<Value>(stats_json).ok()?;
let malloc_metrics = ["malloc_normal", "malloc_huge"];
@@ -315,6 +312,33 @@ fn parse_mimalloc_stats_json(stats_json: &str) -> Option<AllocatorMemoryObservat
}
}
#[cfg(not(target_os = "windows"))]
#[allow(unsafe_code)]
fn read_allocator_memory_snapshot() -> Option<AllocatorMemorySnapshot> {
// SAFETY: `mi_stats_get_json` returns a null-terminated JSON buffer owned by
// mimalloc when called with a null input buffer. The mimalloc API requires
// freeing that buffer with `mi_free`; parsing finishes before the buffer is freed.
let observation = unsafe {
let stats_ptr = libmimalloc_sys::mi_stats_get_json(0, std::ptr::null_mut());
if stats_ptr.is_null() {
return None;
}
let observation = CStr::from_ptr(stats_ptr).to_str().ok().and_then(parse_mimalloc_stats_json);
libmimalloc_sys::mi_free(stats_ptr.cast());
observation?
};
Some(AllocatorMemorySnapshot {
backend: crate::allocator_reclaim::allocator_backend(),
observation,
})
}
#[cfg(target_os = "windows")]
fn read_allocator_memory_snapshot() -> Option<AllocatorMemorySnapshot> {
None
}
fn configured_memory_observability_interval_secs() -> u64 {
rustfs_utils::get_env_u64(ENV_MEMORY_OBSERVABILITY_INTERVAL_SECS, DEFAULT_MEMORY_OBSERVABILITY_INTERVAL_SECS).max(1)
}
@@ -542,13 +566,6 @@ mod tests {
assert_eq!(parse_mimalloc_stats_json(r#"{ "allocator": "unknown" }"#), None);
}
#[test]
fn read_allocator_memory_snapshot_uses_mimalloc_stats_json() {
let snapshot = super::read_allocator_memory_snapshot();
#[cfg(not(target_os = "windows"))]
assert!(snapshot.is_some(), "allocator snapshot should be available on non-Windows");
}
#[test]
fn memory_observability_snapshot_reports_disabled_when_metrics_are_disabled() {
let snapshot = build_memory_observability_status_snapshot(false, 15, false);