mirror of
https://github.com/rustfs/rustfs.git
synced 2026-08-05 12:57:42 +00:00
fix(cache): key the object body cache on data_dir for write-uniqueness (#4703)
* refactor(object-data-cache): derive Default for ObjectDataCacheGetRequest The GET request literal is hand-listed field-by-field across ~13 test sites in two crates. Adding `mod_time_unix_nanos` in backlog#1111 had to touch every one and still missed a literal, producing a compile error caught only in a later CI lane. Derive `Default` and spread the engine-crate literals so the next field addition is absorbed rather than fanned out. Co-Authored-By: heihutu <heihutu@gmail.com> * fix(cache): key the object body cache on data_dir for write-uniqueness The cache key's correctness rested on being write-unique, but its only write-scoped component was `mod_time` — a wall-clock timestamp that is not monotonic and can be absent. Two writes that collide on MD5 (same etag+size) and land on an equal mod_time (clock skew, same-tick, or absent) derived the same key, so a node that never saw the overwrite could serve the previous body for up to the TTL, and the same collision turned the fill-after-invalidation race into a serving bug. This is the store's strong-read-after-write guarantee leaning on a probabilistic argument. Add `data_dir` — the xl.meta directory UUID ecstore regenerates on every body write — as the primary write-unique anchor: - surface `data_dir: Option<Uuid>` on ObjectInfo, copied from FileInfo in the single GET-path constructor `from_file_info`; - carry it into ObjectDataCacheKey as `data_dir_u128` (held as u128 to keep the engine crate free of a uuid dependency), derived in the one planner site both the ecstore hook and the usecase layer share, so both produce an identical key by construction; - keep `mod_time` as a second anchor and `etag+size` as belt-and-braces; an absent data_dir falls back to the prior behavior — strict improvement, no regression. Two writes distinct only by data_dir now derive different keys even under an MD5 collision with identical mod_time — the case mod_time alone cannot cover. Blast radius is compiler-guarded: ObjectInfo derives Default and every real construction site uses `..Default::default()`, so only from_file_info and one full-literal test needed the field. The three P0 body_cache_hook_e2e regressions, engine (80), and app (36) suites pass unchanged; a mutation that severs the planner wiring fails planner_key_changes_with_data_dir. Refs: backlog#1111, backlog#1118 Co-Authored-By: heihutu <heihutu@gmail.com> * fix(cache): thread data_dir field through the merged mutation-hook test Merging main (which landed the object-mutation-hook work, backlog#1131) brought in a GetRequest test literal that predates the data_dir field. Spread it via `..Default::default()` — the derive(Default) added here means this is the last such hand-listed literal to need touching. Co-Authored-By: heihutu <heihutu@gmail.com> --------- Co-authored-by: heihutu <heihutu@gmail.com>
This commit is contained in:
@@ -144,12 +144,13 @@ impl ObjectDataCache {
|
||||
record_plan_decision(self.backend.as_metric_label(), self.config.mode, "cacheable", "eligible", request.size);
|
||||
|
||||
ObjectDataCacheGetPlan::Cacheable {
|
||||
key: ObjectDataCacheKey::with_mod_time(
|
||||
key: ObjectDataCacheKey::with_write_anchors(
|
||||
request.bucket,
|
||||
request.object,
|
||||
request.version_id.as_deref(),
|
||||
request.etag,
|
||||
request.size,
|
||||
request.data_dir_u128,
|
||||
request.mod_time_unix_nanos,
|
||||
request.body_variant,
|
||||
),
|
||||
@@ -393,7 +394,11 @@ const fn invalidation_outcome(result: &ObjectDataCacheInvalidationResult) -> &'s
|
||||
}
|
||||
|
||||
/// Protocol-neutral GET request metadata for cache planning.
|
||||
#[derive(Debug, Clone)]
|
||||
///
|
||||
/// Derives `Default` so tests construct it with `..Default::default()`; a new
|
||||
/// field then does not have to be added to every literal (which is how the
|
||||
/// `mod_time` seam bug arose during backlog#1111).
|
||||
#[derive(Debug, Clone, Default)]
|
||||
pub struct ObjectDataCacheGetRequest<'a> {
|
||||
/// Bucket name.
|
||||
pub bucket: &'a str,
|
||||
@@ -405,9 +410,11 @@ pub struct ObjectDataCacheGetRequest<'a> {
|
||||
pub etag: &'a str,
|
||||
/// Object size in bytes.
|
||||
pub size: u64,
|
||||
/// Resolved version's `data_dir` UUID as a `u128`, or `None`. Primary
|
||||
/// write-unique key component (backlog#1111 / ODC-06).
|
||||
pub data_dir_u128: Option<u128>,
|
||||
/// Resolved version's modification time as Unix nanoseconds, or `0` when
|
||||
/// absent. Carried into the key so it is write-unique (backlog#1111 /
|
||||
/// ODC-06).
|
||||
/// absent. Second write-unique anchor.
|
||||
pub mod_time_unix_nanos: i128,
|
||||
/// Supported response body variant.
|
||||
pub body_variant: ObjectDataCacheBodyVariant,
|
||||
@@ -546,7 +553,7 @@ mod tests {
|
||||
ObjectDataCacheInvalidationReason, ObjectDataCacheInvalidationResult, ObjectDataCacheLookup,
|
||||
};
|
||||
use crate::config::{ObjectDataCacheConfig, ObjectDataCacheMode};
|
||||
use crate::key::{ObjectDataCacheBodyVariant, ObjectDataCacheIdentity};
|
||||
use crate::key::ObjectDataCacheIdentity;
|
||||
use bytes::Bytes;
|
||||
use metrics_util::MetricKind;
|
||||
use metrics_util::debugging::{DebugValue, DebuggingRecorder};
|
||||
@@ -576,11 +583,9 @@ mod tests {
|
||||
ObjectDataCacheGetRequest {
|
||||
bucket,
|
||||
object,
|
||||
version_id: None,
|
||||
etag,
|
||||
size,
|
||||
mod_time_unix_nanos: 0,
|
||||
body_variant: ObjectDataCacheBodyVariant::FullObjectPlainV1,
|
||||
..Default::default()
|
||||
}
|
||||
}
|
||||
|
||||
@@ -846,11 +851,9 @@ mod tests {
|
||||
let plan = cache.plan_get(ObjectDataCacheGetRequest {
|
||||
bucket: "bucket",
|
||||
object: "object",
|
||||
version_id: None,
|
||||
etag: "etag",
|
||||
size: 5,
|
||||
mod_time_unix_nanos: 0,
|
||||
body_variant: ObjectDataCacheBodyVariant::FullObjectPlainV1,
|
||||
..Default::default()
|
||||
});
|
||||
|
||||
let fill = cache.fill_body(&plan, Bytes::from_static(b"oops")).await;
|
||||
@@ -900,11 +903,10 @@ mod tests {
|
||||
let request = ObjectDataCacheGetRequest {
|
||||
bucket: "bucket",
|
||||
object: "object",
|
||||
version_id: None,
|
||||
etag: "etag",
|
||||
size: 5,
|
||||
mod_time_unix_nanos: 42,
|
||||
body_variant: ObjectDataCacheBodyVariant::FullObjectPlainV1,
|
||||
..Default::default()
|
||||
};
|
||||
let ObjectDataCacheGetPlan::Cacheable { key } = cache.plan_get(request) else {
|
||||
panic!("plan should be cacheable");
|
||||
|
||||
@@ -32,10 +32,15 @@ pub enum ObjectDataCacheBodyVariant {
|
||||
/// two same-length payloads that collide on MD5 would derive the identical key,
|
||||
/// so a GET on a node that never observed the overwrite could serve the old
|
||||
/// bytes for up to the TTL, and the same collision turns the
|
||||
/// fill-after-invalidation race (backlog#1118) into a serving bug. Including
|
||||
/// the resolved version's modification time distinguishes two writes even under
|
||||
/// an MD5 collision, because an overwrite advances `mod_time`. `etag + size`
|
||||
/// stay in the key as belt-and-braces.
|
||||
/// fill-after-invalidation race (backlog#1118) into a serving bug.
|
||||
///
|
||||
/// The write-unique anchor is `data_dir` — the xl.meta directory UUID that
|
||||
/// ecstore regenerates on *every* body write. Two distinct writes of the same
|
||||
/// object therefore never share a `data_dir`, so the key changes on overwrite
|
||||
/// even when `etag + size` collide (an MD5 collision) *and* `mod_time` is equal
|
||||
/// (clock skew or same-tick writes). `mod_time_unix_nanos` remains as a second
|
||||
/// anchor for reads where `data_dir` is unavailable, and `etag + size` stay as
|
||||
/// belt-and-braces.
|
||||
#[derive(Debug, Clone, PartialEq, Eq, Hash)]
|
||||
pub struct ObjectDataCacheKey {
|
||||
/// Bucket name.
|
||||
@@ -48,10 +53,15 @@ pub struct ObjectDataCacheKey {
|
||||
pub etag: Arc<str>,
|
||||
/// Object size in bytes.
|
||||
pub size: u64,
|
||||
/// Resolved version's `data_dir` UUID as a `u128` (`Uuid::as_u128`), or
|
||||
/// `None` when the read did not resolve one. This is the primary
|
||||
/// write-unique component: ecstore regenerates `data_dir` on every body
|
||||
/// write, so it is distinct across overwrites regardless of content or
|
||||
/// timestamp. Held as `u128` to keep this crate free of a `uuid` dependency.
|
||||
pub data_dir_u128: Option<u128>,
|
||||
/// Resolved version's modification time as Unix nanoseconds
|
||||
/// (`OffsetDateTime::unix_timestamp_nanos`), or `0` when absent. This is the
|
||||
/// write-unique component: an overwrite advances `mod_time`, so the key
|
||||
/// changes even when `etag + size` are unchanged (an MD5 collision).
|
||||
/// (`OffsetDateTime::unix_timestamp_nanos`), or `0` when absent. Second
|
||||
/// write-unique anchor, used when `data_dir` is unavailable.
|
||||
pub mod_time_unix_nanos: i128,
|
||||
/// Cached body semantics.
|
||||
pub body_variant: ObjectDataCacheBodyVariant,
|
||||
@@ -67,16 +77,17 @@ pub struct ObjectDataCacheIdentity {
|
||||
}
|
||||
|
||||
impl ObjectDataCacheKey {
|
||||
/// Creates a stable object data cache key with an explicit modification
|
||||
/// time. This is the full constructor; the production GET planner uses it so
|
||||
/// the key is write-unique (backlog#1111 / ODC-06).
|
||||
/// Creates a stable object data cache key with explicit write anchors
|
||||
/// (`data_dir` and `mod_time`). This is the full constructor; the production
|
||||
/// GET planner uses it so the key is write-unique (backlog#1111 / ODC-06).
|
||||
#[allow(clippy::too_many_arguments)]
|
||||
pub fn with_mod_time(
|
||||
pub fn with_write_anchors(
|
||||
bucket: impl Into<Arc<str>>,
|
||||
object: impl Into<Arc<str>>,
|
||||
version_id: Option<&str>,
|
||||
etag: impl Into<Arc<str>>,
|
||||
size: u64,
|
||||
data_dir_u128: Option<u128>,
|
||||
mod_time_unix_nanos: i128,
|
||||
body_variant: ObjectDataCacheBodyVariant,
|
||||
) -> Self {
|
||||
@@ -86,14 +97,15 @@ impl ObjectDataCacheKey {
|
||||
version_id: version_id.map_or_else(|| Arc::<str>::from(NULL_VERSION_ID), Arc::<str>::from),
|
||||
etag: etag.into(),
|
||||
size,
|
||||
data_dir_u128,
|
||||
mod_time_unix_nanos,
|
||||
body_variant,
|
||||
}
|
||||
}
|
||||
|
||||
/// Creates a stable object data cache key with no modification time
|
||||
/// (`mod_time_unix_nanos == 0`). Retained for callers that key purely by
|
||||
/// content identity (index/backend tests).
|
||||
/// Creates a stable object data cache key with no write anchors (`data_dir`
|
||||
/// absent, `mod_time_unix_nanos == 0`). Retained for callers that key purely
|
||||
/// by content identity (index/backend tests).
|
||||
pub fn new(
|
||||
bucket: impl Into<Arc<str>>,
|
||||
object: impl Into<Arc<str>>,
|
||||
@@ -102,7 +114,7 @@ impl ObjectDataCacheKey {
|
||||
size: u64,
|
||||
body_variant: ObjectDataCacheBodyVariant,
|
||||
) -> Self {
|
||||
Self::with_mod_time(bucket, object, version_id, etag, size, 0, body_variant)
|
||||
Self::with_write_anchors(bucket, object, version_id, etag, size, None, 0, body_variant)
|
||||
}
|
||||
|
||||
/// Returns true when this key targets the canonical unversioned body variant.
|
||||
@@ -151,33 +163,36 @@ mod tests {
|
||||
// ODC-06 (backlog#1111): two writes with identical etag + size (an MD5
|
||||
// collision on an unversioned overwrite) must derive different keys once
|
||||
// the modification time differs, so a stale node cannot serve old bytes.
|
||||
let old = ObjectDataCacheKey::with_mod_time(
|
||||
let old = ObjectDataCacheKey::with_write_anchors(
|
||||
"bucket",
|
||||
"object",
|
||||
None,
|
||||
"etag",
|
||||
42,
|
||||
None,
|
||||
1_000,
|
||||
ObjectDataCacheBodyVariant::FullObjectPlainV1,
|
||||
);
|
||||
let new = ObjectDataCacheKey::with_mod_time(
|
||||
let new = ObjectDataCacheKey::with_write_anchors(
|
||||
"bucket",
|
||||
"object",
|
||||
None,
|
||||
"etag",
|
||||
42,
|
||||
None,
|
||||
2_000,
|
||||
ObjectDataCacheBodyVariant::FullObjectPlainV1,
|
||||
);
|
||||
|
||||
assert_ne!(old, new, "keys differing only by mod_time must not collide");
|
||||
// Same mod_time still collapses to one key (a true re-read of one write).
|
||||
let new_again = ObjectDataCacheKey::with_mod_time(
|
||||
let new_again = ObjectDataCacheKey::with_write_anchors(
|
||||
"bucket",
|
||||
"object",
|
||||
None,
|
||||
"etag",
|
||||
42,
|
||||
None,
|
||||
2_000,
|
||||
ObjectDataCacheBodyVariant::FullObjectPlainV1,
|
||||
);
|
||||
@@ -185,10 +200,54 @@ mod tests {
|
||||
}
|
||||
|
||||
#[test]
|
||||
fn key_new_defaults_mod_time_to_zero() {
|
||||
fn key_distinguishes_writes_by_data_dir() {
|
||||
// ODC-06 (backlog#1111): `data_dir` is regenerated on every body write,
|
||||
// so two writes distinct only by `data_dir` — identical etag + size (MD5
|
||||
// collision) AND identical mod_time (clock skew or same-tick writes) —
|
||||
// must still derive different keys. This is the case mod_time alone
|
||||
// cannot cover.
|
||||
let first = ObjectDataCacheKey::with_write_anchors(
|
||||
"bucket",
|
||||
"object",
|
||||
None,
|
||||
"etag",
|
||||
42,
|
||||
Some(1),
|
||||
1_000,
|
||||
ObjectDataCacheBodyVariant::FullObjectPlainV1,
|
||||
);
|
||||
let second = ObjectDataCacheKey::with_write_anchors(
|
||||
"bucket",
|
||||
"object",
|
||||
None,
|
||||
"etag",
|
||||
42,
|
||||
Some(2),
|
||||
1_000,
|
||||
ObjectDataCacheBodyVariant::FullObjectPlainV1,
|
||||
);
|
||||
assert_ne!(first, second, "keys differing only by data_dir must not collide");
|
||||
|
||||
// Absent data_dir falls back to the remaining anchors (no regression).
|
||||
let no_data_dir = ObjectDataCacheKey::with_write_anchors(
|
||||
"bucket",
|
||||
"object",
|
||||
None,
|
||||
"etag",
|
||||
42,
|
||||
None,
|
||||
1_000,
|
||||
ObjectDataCacheBodyVariant::FullObjectPlainV1,
|
||||
);
|
||||
assert_ne!(first, no_data_dir);
|
||||
}
|
||||
|
||||
#[test]
|
||||
fn key_new_defaults_write_anchors() {
|
||||
let key = ObjectDataCacheKey::new("bucket", "object", None, "etag", 42, ObjectDataCacheBodyVariant::FullObjectPlainV1);
|
||||
|
||||
assert_eq!(key.mod_time_unix_nanos, 0);
|
||||
assert_eq!(key.data_dir_u128, None);
|
||||
}
|
||||
|
||||
#[test]
|
||||
|
||||
Reference in New Issue
Block a user