fix: complete pending close after close-time flush

Signed-off-by: jeremyhi <fengjiachun@gmail.com>
This commit is contained in:
jeremyhi
2026-07-08 12:40:26 +08:00
parent e9efd591a3
commit 673dc39444
2 changed files with 94 additions and 2 deletions
+60
View File
@@ -287,6 +287,66 @@ async fn test_close_region_skip_wal_while_flush_in_flight_closes_region() {
assert!(!engine.is_region_exists(region_id));
}
#[tokio::test]
async fn test_concurrent_close_region_skip_wal_while_flush_in_flight_succeeds() {
common_telemetry::init_default_ut_logging();
let mut env = TestEnv::with_prefix("concurrent-close-skip-wal-while-flush-in-flight").await;
let listener = Arc::new(AlterFlushListener::default());
let engine = env
.create_engine_with(MitoConfig::default(), None, Some(listener.clone()), None)
.await;
let region_id = RegionId::new(1, 1);
let mut request = CreateRequestBuilder::new().build();
request.options.insert(
WAL_OPTIONS_KEY.to_string(),
serde_json::to_string(&WalOptions::Noop).unwrap(),
);
engine
.handle_request(region_id, RegionRequest::Create(request.clone()))
.await
.unwrap();
let rows = Rows {
schema: rows_schema(&request),
rows: build_rows(0, 3),
};
put_rows(&engine, region_id, rows).await;
let request_count = listener.request_count();
let engine_cloned = engine.clone();
let first_close = tokio::spawn(async move {
engine_cloned
.handle_request(region_id, RegionRequest::Close(RegionCloseRequest {}))
.await
.unwrap();
});
listener.wait_flush_begin().await;
listener.wait_request_count(request_count + 1).await;
let engine_cloned = engine.clone();
let second_close = tokio::spawn(async move {
engine_cloned
.handle_request(region_id, RegionRequest::Close(RegionCloseRequest {}))
.await
.unwrap();
});
listener.wait_request_count(request_count + 2).await;
listener.wake_flush();
tokio::time::timeout(Duration::from_secs(5), first_close)
.await
.unwrap()
.unwrap();
tokio::time::timeout(Duration::from_secs(5), second_close)
.await
.unwrap()
.unwrap();
assert!(!engine.is_region_exists(region_id));
}
#[tokio::test]
async fn test_close_region_after_truncate_skip_wal() {
common_telemetry::init_default_ut_logging();
+34 -2
View File
@@ -57,7 +57,7 @@ use crate::region::options::{IndexOptions, MergeMode, RegionOptions};
use crate::region::version::{VersionControlData, VersionControlRef, VersionRef};
use crate::region::{ManifestContextRef, RegionLeaderState, RegionRoleState, parse_partition_expr};
use crate::request::{
BackgroundNotify, FlushFailed, FlushFinished, OnFailure, OptionOutputTx, OutputTx,
BackgroundNotify, DdlRequest, FlushFailed, FlushFinished, OnFailure, OptionOutputTx, OutputTx,
SenderBulkRequest, SenderDdlRequest, SenderWriteRequest, WorkerRequest, WorkerRequestWithTime,
};
use crate::schedule::scheduler::{Job, SchedulerRef};
@@ -1283,7 +1283,11 @@ impl FlushScheduler {
/// Notifies the scheduler that the region is closed.
pub(crate) fn on_region_closed(&mut self, region_id: RegionId) {
self.remove_region_on_failure(region_id, Arc::new(RegionClosedSnafu { region_id }.build()));
let Some(flush_status) = self.region_status.remove(&region_id) else {
return;
};
flush_status.on_region_closed(Arc::new(RegionClosedSnafu { region_id }.build()));
}
/// Notifies the scheduler that the region is truncated.
@@ -1408,6 +1412,34 @@ impl FlushStatus {
}));
}
}
fn on_region_closed(self, err: Arc<Error>) {
if let Some(mut task) = self.pending_task {
if task.reason == FlushReason::Closing {
task.on_success();
} else {
task.on_failure(err.clone());
}
}
for ddl in self.pending_ddls {
if matches!(ddl.request, DdlRequest::Close(_)) {
ddl.sender.send(Ok(0));
} else {
ddl.sender.send(Err(err.clone()).context(FlushRegionSnafu {
region_id: self.region_id,
}));
}
}
for write_req in self.pending_writes {
write_req
.sender
.send(Err(err.clone()).context(FlushRegionSnafu {
region_id: self.region_id,
}));
}
}
}
#[cfg(test)]