fix(ecstore): repair lifecycle transition and restore flows (#2240)

Co-authored-by: heihutu <heihutu@gmail.com>
Co-authored-by: 安正超 <anzhengchao@gmail.com>
Co-authored-by: houseme <housemecn@gmail.com>
Co-authored-by: loverustfs <hello@rustfs.com>
Co-authored-by: weisd <im@weisd.in>
This commit is contained in:
cxymds
2026-03-23 12:29:13 +08:00
committed by GitHub
parent 2e7abfbd63
commit 236142a682
16 changed files with 1209 additions and 94 deletions
@@ -33,7 +33,9 @@ use crate::global::GLOBAL_LocalNodeName;
use crate::global::{GLOBAL_LifecycleSys, GLOBAL_TierConfigMgr, get_global_deployment_id};
use crate::store::ECStore;
use crate::store_api::StorageAPI;
use crate::store_api::{GetObjectReader, HTTPRangeSpec, ObjectInfo, ObjectOperations, ObjectOptions, ObjectToDelete};
use crate::store_api::{
GetObjectReader, HTTPRangeSpec, ListOperations, ObjectInfo, ObjectOperations, ObjectOptions, ObjectToDelete,
};
use crate::tier::warm_backend::WarmBackendGetOpts;
use async_channel::{Receiver as A_Receiver, Sender as A_Sender, bounded};
use bytes::BytesMut;
@@ -646,18 +648,48 @@ pub async fn validate_transition_tier(lc: &BucketLifecycleConfiguration) -> Resu
}
pub async fn enqueue_transition_immediate(oi: &ObjectInfo, src: LcEventSrc) {
let lc = GLOBAL_LifecycleSys.get(&oi.bucket).await;
if !lc.is_none() {
let event = lc.expect("err").eval(&oi.to_lifecycle_opts()).await;
match event.action {
IlmAction::TransitionAction | IlmAction::TransitionVersionAction => {
if oi.delete_marker || oi.is_dir {
return;
}
GLOBAL_TransitionState.queue_transition_task(oi, &event, &src).await;
}
_ => (),
if let Some(lc) = GLOBAL_LifecycleSys.get(&oi.bucket).await {
enqueue_transition_with_lifecycle(oi, &lc, &src).await;
}
}
pub async fn enqueue_transition_for_existing_objects(api: Arc<ECStore>, bucket: &str) -> Result<(), Error> {
let Some(lc) = GLOBAL_LifecycleSys.get(bucket).await else {
return Ok(());
};
let mut marker = None;
let mut version_marker = None;
let src = LcEventSrc::Scanner;
loop {
let page = api
.clone()
.list_object_versions(bucket, "", marker.clone(), version_marker.clone(), None, 1000)
.await?;
for object in &page.objects {
enqueue_transition_with_lifecycle(object, &lc, &src).await;
}
if !page.is_truncated {
return Ok(());
}
marker = page.next_marker;
version_marker = page.next_version_idmarker;
}
}
async fn enqueue_transition_with_lifecycle(oi: &ObjectInfo, lc: &BucketLifecycleConfiguration, src: &LcEventSrc) {
let event = lc.eval(&oi.to_lifecycle_opts()).await;
match event.action {
IlmAction::TransitionAction | IlmAction::TransitionVersionAction => {
if oi.delete_marker || oi.is_dir {
return;
}
GLOBAL_TransitionState.queue_transition_task(oi, &event, src).await;
}
_ => (),
}
}
@@ -768,7 +768,7 @@ impl LifecycleCalculate for NoncurrentVersionTransition {
#[async_trait::async_trait]
impl LifecycleCalculate for Transition {
fn next_due(&self, obj: &ObjectOpts) -> Option<OffsetDateTime> {
if !obj.is_latest || self.days.is_none() {
if !obj.is_latest {
return None;
}
@@ -1296,6 +1296,45 @@ mod tests {
assert_eq!(event.storage_class, "COLDTIER44");
}
#[tokio::test]
#[serial]
async fn eval_inner_transitions_latest_object_after_date_due() {
let base_time = OffsetDateTime::from_unix_timestamp(1_000_000).unwrap();
let transition_date = base_time - Duration::days(1);
let lc = BucketLifecycleConfiguration {
expiry_updated_at: None,
rules: vec![LifecycleRule {
status: ExpirationStatus::from_static(ExpirationStatus::ENABLED),
expiration: None,
abort_incomplete_multipart_upload: None,
del_marker_expiration: None,
filter: None,
id: Some("transition-date".to_string()),
noncurrent_version_expiration: None,
noncurrent_version_transitions: None,
prefix: None,
transitions: Some(vec![Transition {
days: None,
date: Some(transition_date.into()),
storage_class: Some(TransitionStorageClass::from_static("WARM")),
}]),
}],
};
let opts = ObjectOpts {
name: "obj".to_string(),
mod_time: Some(base_time),
is_latest: true,
transition_status: "".to_string(),
..Default::default()
};
let event = lc.eval_inner(&opts, base_time + Duration::days(1), 0).await;
assert_eq!(event.action, IlmAction::TransitionAction);
assert_eq!(event.rule_id, "transition-date");
assert_eq!(event.storage_class, "WARM");
}
#[tokio::test]
#[serial]
async fn eval_inner_expires_noncurrent_version_after_due() {
@@ -2453,7 +2453,7 @@ impl ReplicateObjectInfoExt for ReplicateObjectInfo {
return rinfo;
}
let sopts = StatObjectOptions {
let mut sopts = StatObjectOptions {
version_id: object_info.version_id.map(|v| v.to_string()).unwrap_or_default(),
internal: AdvancedGetOptions {
replication_proxy_request: "false".to_string(),
@@ -2462,7 +2462,9 @@ impl ReplicateObjectInfoExt for ReplicateObjectInfo {
..Default::default()
};
sopts.set(AMZ_TAGGING_DIRECTIVE, "ACCESS");
if let Err(err) = sopts.set(AMZ_TAGGING_DIRECTIVE, "ACCESS") {
warn!("failed to set replication tagging directive header: {err}");
}
match tgt_client
.head_object(&tgt_client.bucket, &object, self.version_id.map(|v| v.to_string()))
+63 -14
View File
@@ -64,20 +64,32 @@ impl GetObjectOptions {
pub fn header(&self) -> HeaderMap {
let mut headers: HeaderMap = HeaderMap::with_capacity(self.headers.len());
for (k, v) in &self.headers {
if let Ok(header_name) = HeaderName::from_bytes(k.as_bytes()) {
headers.insert(header_name, v.parse().expect("err"));
} else {
warn!("Invalid header name: {}", k);
match (HeaderName::from_bytes(k.as_bytes()), HeaderValue::from_str(v)) {
(Ok(header_name), Ok(header_value)) => {
headers.insert(header_name, header_value);
}
(Err(_), _) => {
warn!("Invalid header name: {}", k);
}
(_, Err(_)) => {
warn!("Invalid header value for {}: {:?}", k, v);
}
}
}
if self.checksum {
headers.insert("x-amz-checksum-mode", "ENABLED".parse().expect("err"));
headers.insert(HeaderName::from_static("x-amz-checksum-mode"), HeaderValue::from_static("ENABLED"));
}
headers
}
pub fn set(&self, key: &str, value: &str) {
//self.headers[http.CanonicalHeaderKey(key)] = value;
pub fn set(&mut self, key: &str, value: &str) -> Result<(), std::io::Error> {
let header_name = HeaderName::from_bytes(key.as_bytes())
.map_err(|err| std::io::Error::other(err_invalid_argument(&format!("Invalid header name {key}: {err}"))))?;
HeaderValue::from_str(value)
.map_err(|err| std::io::Error::other(err_invalid_argument(&format!("Invalid header value for {key}: {err}"))))?;
self.headers.insert(header_name.as_str().to_string(), value.to_string());
Ok(())
}
pub fn set_req_param(&mut self, key: &str, value: &str) {
@@ -89,12 +101,12 @@ impl GetObjectOptions {
}
pub fn set_match_etag(&mut self, etag: &str) -> Result<(), std::io::Error> {
self.set("If-Match", &format!("\"{etag}\""));
self.set("If-Match", &format!("\"{etag}\""))?;
Ok(())
}
pub fn set_match_etag_except(&mut self, etag: &str) -> Result<(), std::io::Error> {
self.set("If-None-Match", &format!("\"{etag}\""));
self.set("If-None-Match", &format!("\"{etag}\""))?;
Ok(())
}
@@ -102,7 +114,7 @@ impl GetObjectOptions {
if mod_time.unix_timestamp() == 0 {
return Err(std::io::Error::other(err_invalid_argument("Modified since cannot be empty.")));
}
self.set("If-Unmodified-Since", &mod_time.to_string());
self.set("If-Unmodified-Since", &mod_time.to_string())?;
Ok(())
}
@@ -110,17 +122,17 @@ impl GetObjectOptions {
if mod_time.unix_timestamp() == 0 {
return Err(std::io::Error::other(err_invalid_argument("Modified since cannot be empty.")));
}
self.set("If-Modified-Since", &mod_time.to_string());
self.set("If-Modified-Since", &mod_time.to_string())?;
Ok(())
}
pub fn set_range(&mut self, start: i64, end: i64) -> Result<(), std::io::Error> {
if start == 0 && end < 0 {
self.set("Range", &format!("bytes={}", end));
self.set("Range", &format!("bytes={}", end))?;
} else if 0 < start && end == 0 {
self.set("Range", &format!("bytes={}-", start));
self.set("Range", &format!("bytes={}-", start))?;
} else if 0 <= start && start <= end {
self.set("Range", &format!("bytes={}-{}", start, end));
self.set("Range", &format!("bytes={}-{}", start, end))?;
} else {
return Err(std::io::Error::other(err_invalid_argument(&format!(
"Invalid range specified: start={} end={}",
@@ -146,3 +158,40 @@ impl GetObjectOptions {
url_values
}
}
#[cfg(test)]
mod tests {
use super::GetObjectOptions;
#[test]
fn set_range_populates_range_header() {
let mut opts = GetObjectOptions::default();
opts.set_range(5, 9).expect("valid range should succeed");
let headers = opts.header();
let range = headers.get("range").expect("range header should be present");
assert_eq!(range.to_str().expect("range header must be valid ascii"), "bytes=5-9");
}
#[test]
fn set_rejects_invalid_header_value() {
let mut opts = GetObjectOptions::default();
let err = opts
.set("Range", "bytes=5-\n9")
.expect_err("invalid header value should fail");
assert!(err.to_string().contains("Invalid header value"));
assert!(opts.headers.is_empty(), "invalid headers must not be stored");
}
#[test]
fn header_skips_invalid_prepopulated_header_value() {
let mut opts = GetObjectOptions::default();
opts.headers.insert("Range".to_string(), "bytes=5-\n9".to_string());
let headers = opts.header();
assert!(headers.get("range").is_none(), "invalid stored header values should be ignored");
}
}
+1 -1
View File
@@ -375,7 +375,7 @@ impl TransitionClient {
//debug!("http_resp_body: {}", String::from_utf8(b).unwrap());
//if self.is_trace_enabled && !(self.trace_errors_only && resp.status() == StatusCode::OK) {
if resp.status() != StatusCode::OK {
if !resp.status().is_success() {
//self.dump_http(&cloned_req, &resp)?;
let mut body_vec = Vec::new();
let mut body = resp.into_body();
+44 -41
View File
@@ -791,14 +791,17 @@ mod test {
panic!("No non-loop back IP address found for this host");
}
let non_loop_back_ip = non_loop_back_i_ps[0];
let remote_ip1 = "192.0.2.10";
let remote_ip2 = "192.0.2.11";
let remote_ip3 = "192.0.2.12";
let case1_endpoint1 = format!("http://{non_loop_back_ip}/d1");
let case1_endpoint2 = format!("http://{non_loop_back_ip}/d2");
let args = vec![
format!("http://{}:10000/d1", non_loop_back_ip),
format!("http://{}:10000/d2", non_loop_back_ip),
"http://example.org:10000/d3".to_string(),
"http://example.com:10000/d4".to_string(),
format!("http://{remote_ip1}:10000/d3"),
format!("http://{remote_ip2}:10000/d4"),
];
let (case1_ur_ls, case1_local_flags) = get_expected_endpoints(args, format!("http://{non_loop_back_ip}:10000/"));
@@ -807,26 +810,26 @@ mod test {
let args = vec![
format!("http://{}:10000/d1", non_loop_back_ip),
format!("http://{}:9000/d2", non_loop_back_ip),
"http://example.org:10000/d3".to_string(),
"http://example.com:10000/d4".to_string(),
format!("http://{remote_ip1}:10000/d3"),
format!("http://{remote_ip2}:10000/d4"),
];
let (case2_ur_ls, case2_local_flags) = get_expected_endpoints(args, format!("http://{non_loop_back_ip}:10000/"));
let case3_endpoint1 = format!("http://{non_loop_back_ip}/d1");
let args = vec![
format!("http://{}:80/d1", non_loop_back_ip),
"http://example.org:9000/d2".to_string(),
"http://example.com:80/d3".to_string(),
"http://example.net:80/d4".to_string(),
format!("http://{remote_ip1}:9000/d2"),
format!("http://{remote_ip2}:80/d3"),
format!("http://{remote_ip3}:80/d4"),
];
let (case3_ur_ls, case3_local_flags) = get_expected_endpoints(args, format!("http://{non_loop_back_ip}:80/"));
let case4_endpoint1 = format!("http://{non_loop_back_ip}/d1");
let args = vec![
format!("http://{}:9000/d1", non_loop_back_ip),
"http://example.org:9000/d2".to_string(),
"http://example.com:9000/d3".to_string(),
"http://example.net:9000/d4".to_string(),
format!("http://{remote_ip1}:9000/d2"),
format!("http://{remote_ip2}:9000/d3"),
format!("http://{remote_ip3}:9000/d4"),
];
let (case4_ur_ls, case4_local_flags) = get_expected_endpoints(args, format!("http://{non_loop_back_ip}:9000/"));
@@ -844,8 +847,8 @@ mod test {
let case6_endpoint1 = format!("http://{non_loop_back_ip}:9003/d4");
let args = vec![
"http://localhost:9000/d1".to_string(),
"http://localhost:9001/d2".to_string(),
"http://127.0.0.1:9000/d1".to_string(),
"http://127.0.0.1:9001/d2".to_string(),
"http://127.0.0.1:9002/d3".to_string(),
case6_endpoint1.clone(),
];
@@ -864,8 +867,8 @@ mod test {
// Erasure Single Drive
TestCase {
num: 2,
server_addr: "localhost:9000",
args: vec!["http://localhost/d1"],
server_addr: "127.0.0.1:9000",
args: vec!["http://127.0.0.1/d1"],
expected_err: Some(Error::other("use path style endpoint for single node setup")),
..Default::default()
},
@@ -885,7 +888,7 @@ mod test {
},
TestCase {
num: 4,
server_addr: "localhost:10000",
server_addr: "127.0.0.1:10000",
args: vec!["/d1"],
expected_endpoints: Some(Endpoints(vec![Endpoint {
url: must_file_path("/d1"),
@@ -899,12 +902,12 @@ mod test {
},
TestCase {
num: 5,
server_addr: "localhost:9000",
server_addr: "127.0.0.1:9000",
args: vec![
"https://127.0.0.1:9000/d1",
"https://localhost:9001/d1",
"https://example.com/d1",
"https://example.com/d2",
"https://127.0.0.1:9001/d1",
"https://192.0.2.1/d1",
"https://192.0.2.1/d2",
],
expected_err: Some(Error::other("same path '/d1' can not be served by different port on same address")),
..Default::default()
@@ -952,35 +955,35 @@ mod test {
num: 7,
server_addr: "0.0.0.0:9000",
args: vec![
"http://localhost/d1",
"http://localhost/d2",
"http://localhost/d3",
"http://localhost/d4",
"http://127.0.0.1/d1",
"http://127.0.0.1/d2",
"http://127.0.0.1/d3",
"http://127.0.0.1/d4",
],
expected_endpoints: Some(Endpoints(vec![
Endpoint {
url: must_url("http://localhost:9000/d1"),
url: must_url("http://127.0.0.1:9000/d1"),
is_local: true,
pool_idx: 0,
set_idx: 0,
disk_idx: 0,
},
Endpoint {
url: must_url("http://localhost:9000/d2"),
url: must_url("http://127.0.0.1:9000/d2"),
is_local: true,
pool_idx: 0,
set_idx: 0,
disk_idx: 0,
},
Endpoint {
url: must_url("http://localhost:9000/d3"),
url: must_url("http://127.0.0.1:9000/d3"),
is_local: true,
pool_idx: 0,
set_idx: 0,
disk_idx: 0,
},
Endpoint {
url: must_url("http://localhost:9000/d4"),
url: must_url("http://127.0.0.1:9000/d4"),
is_local: true,
pool_idx: 0,
set_idx: 0,
@@ -995,8 +998,8 @@ mod test {
num: 8,
server_addr: "127.0.0.1:10000",
args: vec![
"http://localhost/d1",
"http://localhost/d2",
"http://[::1]/d1",
"http://[::1]/d2",
"http://127.0.0.1/d3",
"http://127.0.0.1/d4",
],
@@ -1034,8 +1037,8 @@ mod test {
args: vec![
case1_endpoint1.as_str(),
case1_endpoint2.as_str(),
"http://example.org/d3",
"http://example.com/d4",
"http://192.0.2.10/d3",
"http://192.0.2.11/d4",
],
expected_endpoints: Some(Endpoints(vec![
Endpoint {
@@ -1076,8 +1079,8 @@ mod test {
args: vec![
case2_endpoint1.as_str(),
case2_endpoint2.as_str(),
"http://example.org/d3",
"http://example.com/d4",
"http://192.0.2.10/d3",
"http://192.0.2.11/d4",
],
expected_endpoints: Some(Endpoints(vec![
Endpoint {
@@ -1117,9 +1120,9 @@ mod test {
server_addr: "0.0.0.0:80",
args: vec![
case3_endpoint1.as_str(),
"http://example.org:9000/d2",
"http://example.com/d3",
"http://example.net/d4",
"http://192.0.2.10:9000/d2",
"http://192.0.2.11/d3",
"http://192.0.2.12/d4",
],
expected_endpoints: Some(Endpoints(vec![
Endpoint {
@@ -1159,9 +1162,9 @@ mod test {
server_addr: "0.0.0.0:9000",
args: vec![
case4_endpoint1.as_str(),
"http://example.org/d2",
"http://example.com/d3",
"http://example.net/d4",
"http://192.0.2.10/d2",
"http://192.0.2.11/d3",
"http://192.0.2.12/d4",
],
expected_endpoints: Some(Endpoints(vec![
Endpoint {
@@ -1242,8 +1245,8 @@ mod test {
num: 16,
server_addr: "0.0.0.0:9003",
args: vec![
"http://localhost:9000/d1",
"http://localhost:9001/d2",
"http://127.0.0.1:9000/d1",
"http://127.0.0.1:9001/d2",
"http://127.0.0.1:9002/d3",
case6_endpoint1.as_str(),
],
+47 -12
View File
@@ -1943,19 +1943,54 @@ impl ObjectOperations for SetDisks {
//}
let mut uploaded_parts: Vec<CompletePart> = vec![];
let rs: Option<HTTPRangeSpec> = None;
let gr = get_transitioned_object_reader(bucket, object, &rs, &HeaderMap::new(), &oi, opts).await;
if let Err(err) = gr {
return set_restore_header_fn(&mut oi, Some(StorageError::Io(err))).await;
}
let gr = gr.unwrap();
for part_info in &oi.parts {
let reader = BufReader::new(Cursor::new(vec![] /*gr.stream*/));
let parts = oi.parts.clone();
let mut part_offset: i64 = 0;
for part_info in &parts {
let mut part_opts = opts.clone();
part_opts.part_number = Some(part_info.number);
if part_info.actual_size <= 0 {
return set_restore_header_fn(
&mut oi,
Some(Error::other(format!("invalid multipart restore part size {}", part_info.actual_size))),
)
.await;
}
let part_end = match part_offset.checked_add(part_info.actual_size - 1) {
Some(end) => end,
None => {
return set_restore_header_fn(
&mut oi,
Some(Error::other("multipart restore part range overflow".to_string())),
)
.await;
}
};
let rs = Some(HTTPRangeSpec {
is_suffix_length: false,
start: part_offset,
end: part_end,
});
part_offset = match part_end.checked_add(1) {
Some(next) => next,
None => {
return set_restore_header_fn(
&mut oi,
Some(Error::other("multipart restore part offset overflow".to_string())),
)
.await;
}
};
let gr = match get_transitioned_object_reader(bucket, object, &rs, &HeaderMap::new(), &oi, &part_opts).await {
Ok(reader) => reader,
Err(err) => {
return set_restore_header_fn(&mut oi, Some(StorageError::Io(err))).await;
}
};
let reader = BufReader::new(gr.stream);
let hash_reader = HashReader::new(
Box::new(WarpReader::new(reader)),
part_info.size as i64,
part_info.size as i64,
part_info.actual_size,
part_info.actual_size,
None,
None,
false,
@@ -1968,7 +2003,7 @@ impl ObjectOperations for SetDisks {
//if let Err(err) = p_info {
// return set_restore_header_fn(&mut oi, err).await;
//}
if p_info.size != part_info.size {
if p_info.size as i64 != part_info.actual_size {
return set_restore_header_fn(
&mut oi,
Some(Error::other(ObjectApiError::InvalidObjectState(GenericError {
+1 -1
View File
@@ -27,7 +27,7 @@ use crate::bucket::utils::check_new_multipart_args;
use crate::bucket::utils::check_object_args;
use crate::bucket::utils::check_put_object_args;
use crate::bucket::utils::check_put_object_part_args;
use crate::bucket::utils::{check_valid_bucket_name, check_valid_bucket_name_strict, is_meta_bucketname};
use crate::bucket::utils::{check_valid_bucket_name, check_valid_bucket_name_strict};
use crate::config::GLOBAL_STORAGE_CLASS;
use crate::config::storageclass;
use crate::disk::endpoint::{Endpoint, EndpointType};
+1
View File
@@ -13,6 +13,7 @@
// limitations under the License.
use super::*;
use crate::bucket::utils::is_meta_bucketname;
fn should_override_created_from_metadata(created: OffsetDateTime) -> bool {
created != OffsetDateTime::UNIX_EPOCH
+1
View File
@@ -13,6 +13,7 @@
// limitations under the License.
use super::*;
use crate::bucket::utils::is_meta_bucketname;
impl ECStore {
#[instrument(level = "debug", skip(self))]