mirror of
https://github.com/GreptimeTeam/greptimedb.git
synced 2026-05-14 12:00:40 +00:00
fix(mito2): introduce PartitionExprChange in staging flow and keep memtables on metadata-only updates (#7695)
* feat(mito2): add RegionMetaAction::PartitionExprChange Signed-off-by: WenyXu <wenymedia@gmail.com> * refactor(mito2): apply partition-expr action in manifest builder and manager Signed-off-by: WenyXu <wenymedia@gmail.com> * refactor(mito2): add partition-expr action merge rules and conflict guard Signed-off-by: WenyXu <wenymedia@gmail.com> * refactor(mito2): use partition-rule action in enter staging Signed-off-by: WenyXu <wenymedia@gmail.com> * fix(mito2): validate Change and route to metadata-only update on staging exit Signed-off-by: WenyXu <wenymedia@gmail.com> * test(mito2): cover partition-expr action staging flow and conflict cases Signed-off-by: WenyXu <wenymedia@gmail.com> * test(mito2): add apply-staging coverage for Change metadata validation Signed-off-by: WenyXu <wenymedia@gmail.com> * test(mito2): add apply-staging coverage for Change metadata validation Signed-off-by: WenyXu <wenymedia@gmail.com> * chore: fmt Signed-off-by: WenyXu <wenymedia@gmail.com> * chore: remove unused error Signed-off-by: WenyXu <wenymedia@gmail.com> * chore: apply suggestions Signed-off-by: WenyXu <wenymedia@gmail.com> * chore: add warn Signed-off-by: WenyXu <wenymedia@gmail.com> * chore: add comments Signed-off-by: WenyXu <wenymedia@gmail.com> * test: preserves unflushed memtable Signed-off-by: WenyXu <wenymedia@gmail.com> * chore: fmt Signed-off-by: WenyXu <wenymedia@gmail.com> --------- Signed-off-by: WenyXu <wenymedia@gmail.com>
This commit is contained in:
@@ -14,8 +14,10 @@
|
||||
|
||||
use std::assert_matches::assert_matches;
|
||||
use std::fs;
|
||||
use std::sync::Arc;
|
||||
|
||||
use api::v1::Rows;
|
||||
use common_function::utils::partition_rule_version;
|
||||
use common_recordbatch::RecordBatches;
|
||||
use datatypes::value::Value;
|
||||
use partition::expr::{PartitionExpr, col};
|
||||
@@ -23,14 +25,18 @@ use store_api::region_engine::{
|
||||
RegionEngine, RegionRole, RemapManifestsRequest, SettableRegionRoleState,
|
||||
};
|
||||
use store_api::region_request::{
|
||||
ApplyStagingManifestRequest, EnterStagingRequest, RegionFlushRequest, RegionRequest,
|
||||
ApplyStagingManifestRequest, EnterStagingRequest, RegionFlushRequest, RegionPutRequest,
|
||||
RegionRequest,
|
||||
};
|
||||
use store_api::storage::{FileId, RegionId};
|
||||
|
||||
use super::ScanRequest;
|
||||
use crate::config::MitoConfig;
|
||||
use crate::error::Error;
|
||||
use crate::manifest::action::RegionManifest;
|
||||
use crate::manifest::action::{
|
||||
RegionChange, RegionEdit, RegionManifest, RegionMetaAction, RegionMetaActionList,
|
||||
};
|
||||
use crate::sst::FormatType;
|
||||
use crate::sst::file::FileMeta;
|
||||
use crate::test_util::{CreateRequestBuilder, TestEnv, build_rows, put_rows, rows_schema};
|
||||
|
||||
@@ -40,6 +46,12 @@ fn range_expr(col_name: &str, start: i64, end: i64) -> PartitionExpr {
|
||||
.and(col(col_name).lt(Value::Int64(end)))
|
||||
}
|
||||
|
||||
fn float_range_expr(col_name: &str, start: f64, end: f64) -> PartitionExpr {
|
||||
col(col_name)
|
||||
.gt_eq(Value::Float64(start.into()))
|
||||
.and(col(col_name).lt(Value::Float64(end.into())))
|
||||
}
|
||||
|
||||
#[tokio::test]
|
||||
async fn test_apply_staging_manifest_invalid_region_state() {
|
||||
common_telemetry::init_default_ut_logging();
|
||||
@@ -454,6 +466,228 @@ async fn test_apply_staging_manifest_invalid_files_to_add_with_format(flat_forma
|
||||
);
|
||||
}
|
||||
|
||||
#[tokio::test]
|
||||
async fn test_apply_staging_manifest_change_edit_different_columns_fails() {
|
||||
test_apply_staging_manifest_change_edit_different_columns_fails_with_format(false).await;
|
||||
test_apply_staging_manifest_change_edit_different_columns_fails_with_format(true).await;
|
||||
}
|
||||
|
||||
async fn test_apply_staging_manifest_change_edit_different_columns_fails_with_format(
|
||||
flat_format: bool,
|
||||
) {
|
||||
let mut env = TestEnv::with_prefix("apply-change-edit-different-columns").await;
|
||||
let engine = env
|
||||
.create_engine(MitoConfig {
|
||||
default_experimental_flat_format: flat_format,
|
||||
..Default::default()
|
||||
})
|
||||
.await;
|
||||
let region_id = RegionId::new(2, 2);
|
||||
let request = CreateRequestBuilder::new().build();
|
||||
let column_schemas = rows_schema(&request);
|
||||
engine
|
||||
.handle_request(region_id, RegionRequest::Create(request))
|
||||
.await
|
||||
.unwrap();
|
||||
|
||||
let rows_data = Rows {
|
||||
schema: column_schemas,
|
||||
rows: build_rows(0, 3),
|
||||
};
|
||||
put_rows(&engine, region_id, rows_data).await;
|
||||
engine
|
||||
.handle_request(
|
||||
region_id,
|
||||
RegionRequest::Flush(RegionFlushRequest {
|
||||
row_group_size: None,
|
||||
}),
|
||||
)
|
||||
.await
|
||||
.unwrap();
|
||||
|
||||
let partition_expr = range_expr("tag_0", 0, 50).as_json_str().unwrap();
|
||||
engine
|
||||
.handle_request(
|
||||
region_id,
|
||||
RegionRequest::EnterStaging(EnterStagingRequest {
|
||||
partition_expr: partition_expr.clone(),
|
||||
}),
|
||||
)
|
||||
.await
|
||||
.unwrap();
|
||||
|
||||
let remap_result = engine
|
||||
.remap_manifests(RemapManifestsRequest {
|
||||
region_id,
|
||||
input_regions: vec![region_id],
|
||||
region_mapping: [(region_id, vec![region_id])].into_iter().collect(),
|
||||
new_partition_exprs: [(region_id, partition_expr.clone())].into_iter().collect(),
|
||||
})
|
||||
.await
|
||||
.unwrap();
|
||||
|
||||
let region = engine.get_region(region_id).unwrap();
|
||||
let mut manager = region.manifest_ctx.manifest_manager.write().await;
|
||||
let manifest_storage = manager.store();
|
||||
let blob_store = manifest_storage.staging_storage().blob_storage();
|
||||
let remap_manifest_path = remap_result.manifest_paths[®ion_id].clone();
|
||||
let remap_manifest_bytes = blob_store.get(&remap_manifest_path).await.unwrap();
|
||||
manager.clear_staging_manifest_and_dir().await.unwrap();
|
||||
|
||||
let mut changed_metadata = region.version().metadata.as_ref().clone();
|
||||
changed_metadata.column_metadatas.rotate_left(1);
|
||||
|
||||
manager
|
||||
.update(
|
||||
RegionMetaActionList::new(vec![
|
||||
RegionMetaAction::Change(RegionChange {
|
||||
metadata: Arc::new(changed_metadata),
|
||||
sst_format: FormatType::PrimaryKey,
|
||||
}),
|
||||
RegionMetaAction::Edit(RegionEdit {
|
||||
files_to_add: Vec::new(),
|
||||
files_to_remove: Vec::new(),
|
||||
timestamp_ms: None,
|
||||
compaction_time_window: None,
|
||||
flushed_entry_id: None,
|
||||
flushed_sequence: None,
|
||||
committed_sequence: None,
|
||||
}),
|
||||
]),
|
||||
true,
|
||||
)
|
||||
.await
|
||||
.unwrap();
|
||||
blob_store
|
||||
.put(&remap_manifest_path, remap_manifest_bytes)
|
||||
.await
|
||||
.unwrap();
|
||||
drop(manager);
|
||||
|
||||
let err = engine
|
||||
.handle_request(
|
||||
region_id,
|
||||
RegionRequest::ApplyStagingManifest(ApplyStagingManifestRequest {
|
||||
partition_expr,
|
||||
central_region_id: region_id,
|
||||
manifest_path: remap_manifest_path,
|
||||
}),
|
||||
)
|
||||
.await
|
||||
.unwrap_err();
|
||||
assert_matches!(
|
||||
err.into_inner().as_any().downcast_ref::<Error>().unwrap(),
|
||||
Error::Unexpected { .. }
|
||||
);
|
||||
}
|
||||
|
||||
#[tokio::test]
|
||||
async fn test_apply_staging_manifest_preserves_unflushed_memtable() {
|
||||
test_apply_staging_manifest_preserves_unflushed_memtable_with_format(false).await;
|
||||
test_apply_staging_manifest_preserves_unflushed_memtable_with_format(true).await;
|
||||
}
|
||||
|
||||
async fn test_apply_staging_manifest_preserves_unflushed_memtable_with_format(flat_format: bool) {
|
||||
let mut env = TestEnv::with_prefix("apply-preserve-memtable").await;
|
||||
let engine = env
|
||||
.create_engine(MitoConfig {
|
||||
default_experimental_flat_format: flat_format,
|
||||
..Default::default()
|
||||
})
|
||||
.await;
|
||||
|
||||
let region_id = RegionId::new(3, 1);
|
||||
let origin_partition_expr = float_range_expr("field_0", 0., 50.).as_json_str().unwrap();
|
||||
let request = CreateRequestBuilder::new()
|
||||
.partition_expr_json(Some(origin_partition_expr))
|
||||
.build();
|
||||
let column_schemas = rows_schema(&request);
|
||||
|
||||
engine
|
||||
.handle_request(region_id, RegionRequest::Create(request))
|
||||
.await
|
||||
.unwrap();
|
||||
|
||||
let base_rows = Rows {
|
||||
schema: column_schemas.clone(),
|
||||
rows: build_rows(0, 3),
|
||||
};
|
||||
put_rows(&engine, region_id, base_rows).await;
|
||||
engine
|
||||
.handle_request(
|
||||
region_id,
|
||||
RegionRequest::Flush(RegionFlushRequest {
|
||||
row_group_size: None,
|
||||
}),
|
||||
)
|
||||
.await
|
||||
.unwrap();
|
||||
|
||||
let partition_expr = float_range_expr("field_0", 0., 100.).as_json_str().unwrap();
|
||||
engine
|
||||
.handle_request(
|
||||
region_id,
|
||||
RegionRequest::EnterStaging(EnterStagingRequest {
|
||||
partition_expr: partition_expr.clone(),
|
||||
}),
|
||||
)
|
||||
.await
|
||||
.unwrap();
|
||||
|
||||
let expected_version = partition_rule_version(Some(&partition_expr));
|
||||
let unflushed_rows = Rows {
|
||||
schema: column_schemas,
|
||||
rows: build_rows(3, 6),
|
||||
};
|
||||
engine
|
||||
.handle_request(
|
||||
region_id,
|
||||
RegionRequest::Put(RegionPutRequest {
|
||||
rows: unflushed_rows,
|
||||
hint: None,
|
||||
partition_rule_version: Some(expected_version),
|
||||
}),
|
||||
)
|
||||
.await
|
||||
.unwrap();
|
||||
|
||||
let remap_result = engine
|
||||
.remap_manifests(RemapManifestsRequest {
|
||||
region_id,
|
||||
input_regions: vec![region_id],
|
||||
region_mapping: [(region_id, vec![region_id])].into_iter().collect(),
|
||||
new_partition_exprs: [(region_id, partition_expr.clone())].into_iter().collect(),
|
||||
})
|
||||
.await
|
||||
.unwrap();
|
||||
|
||||
engine
|
||||
.handle_request(
|
||||
region_id,
|
||||
RegionRequest::ApplyStagingManifest(ApplyStagingManifestRequest {
|
||||
partition_expr,
|
||||
central_region_id: region_id,
|
||||
manifest_path: remap_result.manifest_paths[®ion_id].clone(),
|
||||
}),
|
||||
)
|
||||
.await
|
||||
.unwrap();
|
||||
|
||||
let scanner = engine
|
||||
.scanner(region_id, ScanRequest::default())
|
||||
.await
|
||||
.unwrap();
|
||||
assert!(
|
||||
scanner.num_memtables() > 0,
|
||||
"unflushed memtable should be preserved after apply staging manifest"
|
||||
);
|
||||
let batches = RecordBatches::try_collect(scanner.scan().await.unwrap())
|
||||
.await
|
||||
.unwrap();
|
||||
let total_rows: usize = batches.iter().map(|rb| rb.num_rows()).sum();
|
||||
assert_eq!(total_rows, 6, "rows from memtable should remain readable");
|
||||
}
|
||||
|
||||
#[tokio::test]
|
||||
async fn test_split_repartition_causes_duplicate_data() {
|
||||
common_telemetry::init_default_ut_logging();
|
||||
|
||||
@@ -31,7 +31,9 @@ use object_store::layers::mock::{
|
||||
Result as MockResult, Write, Writer,
|
||||
};
|
||||
use partition::expr::{PartitionExpr, col};
|
||||
use store_api::region_engine::{RegionEngine, RemapManifestsRequest, SettableRegionRoleState};
|
||||
use store_api::region_engine::{
|
||||
RegionEngine, RemapManifestsRequest, SetRegionRoleStateResponse, SettableRegionRoleState,
|
||||
};
|
||||
use store_api::region_request::{
|
||||
ApplyStagingManifestRequest, EnterStagingRequest, RegionAlterRequest, RegionFlushRequest,
|
||||
RegionPutRequest, RegionRequest, RegionTruncateRequest,
|
||||
@@ -41,8 +43,12 @@ use store_api::storage::{RegionId, ScanRequest};
|
||||
use crate::config::MitoConfig;
|
||||
use crate::engine::listener::NotifyEnterStagingResultListener;
|
||||
use crate::error::Error;
|
||||
use crate::manifest::action::{
|
||||
RegionChange, RegionEdit, RegionMetaAction, RegionMetaActionList, RegionPartitionExprChange,
|
||||
};
|
||||
use crate::region::{RegionLeaderState, RegionRoleState, parse_partition_expr};
|
||||
use crate::request::WorkerRequest;
|
||||
use crate::sst::FormatType;
|
||||
use crate::test_util::{CreateRequestBuilder, TestEnv, build_rows, put_rows, rows_schema};
|
||||
|
||||
fn range_expr(col_name: &str, start: i64, end: i64) -> PartitionExpr {
|
||||
@@ -810,6 +816,147 @@ async fn test_staging_exit_success_with_manifests_with_format(flat_format: bool)
|
||||
assert!(sst_entries.iter().all(|e| e.visible));
|
||||
}
|
||||
|
||||
#[tokio::test]
|
||||
async fn test_enter_staging_writes_partition_expr_change_action() {
|
||||
test_enter_staging_writes_partition_expr_change_action_with_format(false).await;
|
||||
test_enter_staging_writes_partition_expr_change_action_with_format(true).await;
|
||||
}
|
||||
|
||||
async fn test_enter_staging_writes_partition_expr_change_action_with_format(flat_format: bool) {
|
||||
let mut env = TestEnv::new().await;
|
||||
let engine = env
|
||||
.create_engine(MitoConfig {
|
||||
default_experimental_flat_format: flat_format,
|
||||
..Default::default()
|
||||
})
|
||||
.await;
|
||||
|
||||
let region_id = RegionId::new(2000, 1);
|
||||
let request = CreateRequestBuilder::new().build();
|
||||
engine
|
||||
.handle_request(region_id, RegionRequest::Create(request))
|
||||
.await
|
||||
.unwrap();
|
||||
|
||||
let partition_expr = default_partition_expr();
|
||||
engine
|
||||
.handle_request(
|
||||
region_id,
|
||||
RegionRequest::EnterStaging(EnterStagingRequest {
|
||||
partition_expr: partition_expr.clone(),
|
||||
}),
|
||||
)
|
||||
.await
|
||||
.unwrap();
|
||||
|
||||
let region = engine.get_region(region_id).unwrap();
|
||||
let manager = region.manifest_ctx.manifest_manager.read().await;
|
||||
let staging_manifests = manager.store().fetch_staging_manifests().await.unwrap();
|
||||
assert!(!staging_manifests.is_empty());
|
||||
|
||||
let mut found_partition_expr_change = false;
|
||||
let mut found_change = false;
|
||||
for (_, raw_action_list) in staging_manifests {
|
||||
let action_list = RegionMetaActionList::decode(&raw_action_list).unwrap();
|
||||
for action in action_list.actions {
|
||||
match action {
|
||||
RegionMetaAction::PartitionExprChange(change) => {
|
||||
found_partition_expr_change = true;
|
||||
assert_eq!(change.partition_expr, Some(partition_expr.clone()));
|
||||
}
|
||||
RegionMetaAction::Change(_) => {
|
||||
found_change = true;
|
||||
}
|
||||
_ => {}
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
assert!(found_partition_expr_change);
|
||||
assert!(!found_change);
|
||||
}
|
||||
|
||||
#[tokio::test]
|
||||
async fn test_staging_exit_conflict_partition_expr_change_and_change() {
|
||||
test_staging_exit_conflict_partition_expr_change_and_change_with_format(false).await;
|
||||
test_staging_exit_conflict_partition_expr_change_and_change_with_format(true).await;
|
||||
}
|
||||
|
||||
async fn test_staging_exit_conflict_partition_expr_change_and_change_with_format(
|
||||
flat_format: bool,
|
||||
) {
|
||||
let mut env = TestEnv::new().await;
|
||||
let engine = env
|
||||
.create_engine(MitoConfig {
|
||||
default_experimental_flat_format: flat_format,
|
||||
..Default::default()
|
||||
})
|
||||
.await;
|
||||
|
||||
let region_id = RegionId::new(2000, 2);
|
||||
let request = CreateRequestBuilder::new().build();
|
||||
engine
|
||||
.handle_request(region_id, RegionRequest::Create(request))
|
||||
.await
|
||||
.unwrap();
|
||||
|
||||
let partition_expr = default_partition_expr();
|
||||
engine
|
||||
.handle_request(
|
||||
region_id,
|
||||
RegionRequest::EnterStaging(EnterStagingRequest {
|
||||
partition_expr: partition_expr.clone(),
|
||||
}),
|
||||
)
|
||||
.await
|
||||
.unwrap();
|
||||
|
||||
let region = engine.get_region(region_id).unwrap();
|
||||
let mut changed_metadata = region.version().metadata.as_ref().clone();
|
||||
changed_metadata.set_partition_expr(Some(partition_expr.clone()));
|
||||
|
||||
let mut manager = region.manifest_ctx.manifest_manager.write().await;
|
||||
manager
|
||||
.update(
|
||||
RegionMetaActionList::new(vec![
|
||||
RegionMetaAction::PartitionExprChange(RegionPartitionExprChange {
|
||||
partition_expr: Some(partition_expr),
|
||||
}),
|
||||
RegionMetaAction::Change(RegionChange {
|
||||
metadata: Arc::new(changed_metadata),
|
||||
sst_format: FormatType::PrimaryKey,
|
||||
}),
|
||||
RegionMetaAction::Edit(RegionEdit {
|
||||
files_to_add: Vec::new(),
|
||||
files_to_remove: Vec::new(),
|
||||
timestamp_ms: None,
|
||||
compaction_time_window: None,
|
||||
flushed_entry_id: None,
|
||||
flushed_sequence: None,
|
||||
committed_sequence: None,
|
||||
}),
|
||||
]),
|
||||
true,
|
||||
)
|
||||
.await
|
||||
.unwrap();
|
||||
drop(manager);
|
||||
|
||||
let response = engine
|
||||
.set_region_role_state_gracefully(region_id, SettableRegionRoleState::Leader)
|
||||
.await
|
||||
.unwrap();
|
||||
match response {
|
||||
SetRegionRoleStateResponse::InvalidTransition(err) => {
|
||||
assert_matches!(
|
||||
err.as_any().downcast_ref::<Error>().unwrap(),
|
||||
Error::Unexpected { .. }
|
||||
);
|
||||
}
|
||||
_ => panic!("Expected InvalidTransition response, got: {response:?}"),
|
||||
}
|
||||
}
|
||||
|
||||
#[tokio::test(flavor = "multi_thread")]
|
||||
async fn test_write_stall_on_enter_staging() {
|
||||
test_write_stall_on_enter_staging_with_format(false).await;
|
||||
|
||||
@@ -18,6 +18,7 @@ use std::collections::{HashMap, HashSet};
|
||||
use std::time::Duration;
|
||||
|
||||
use chrono::Utc;
|
||||
use common_telemetry::warn;
|
||||
use serde::{Deserialize, Serialize};
|
||||
use snafu::{OptionExt, ResultExt};
|
||||
use store_api::ManifestVersion;
|
||||
@@ -37,6 +38,8 @@ use crate::wal::EntryId;
|
||||
pub enum RegionMetaAction {
|
||||
/// Change region's metadata for request like ALTER
|
||||
Change(RegionChange),
|
||||
/// Change only region partition expression metadata.
|
||||
PartitionExprChange(RegionPartitionExprChange),
|
||||
/// Edit region's state for changing options or file list.
|
||||
Edit(RegionEdit),
|
||||
/// Remove the region.
|
||||
@@ -45,6 +48,12 @@ pub enum RegionMetaAction {
|
||||
Truncate(RegionTruncate),
|
||||
}
|
||||
|
||||
#[derive(Serialize, Deserialize, Clone, Debug, PartialEq, Eq)]
|
||||
pub struct RegionPartitionExprChange {
|
||||
/// Partition expression serialized as JSON.
|
||||
pub partition_expr: Option<String>,
|
||||
}
|
||||
|
||||
impl RegionMetaAction {
|
||||
/// Returns true if the action is a change action.
|
||||
pub fn is_change(&self) -> bool {
|
||||
@@ -55,6 +64,11 @@ impl RegionMetaAction {
|
||||
pub fn is_edit(&self) -> bool {
|
||||
matches!(self, RegionMetaAction::Edit(_))
|
||||
}
|
||||
|
||||
/// Returns true if the action is a partition expr change action.
|
||||
pub fn is_partition_expr_change(&self) -> bool {
|
||||
matches!(self, RegionMetaAction::PartitionExprChange(_))
|
||||
}
|
||||
}
|
||||
|
||||
#[derive(Serialize, Deserialize, Clone, Debug, PartialEq, Eq)]
|
||||
@@ -203,6 +217,29 @@ impl RegionManifestBuilder {
|
||||
self.sst_format = change.sst_format;
|
||||
}
|
||||
|
||||
/// Applies a partition-expression-only metadata change.
|
||||
///
|
||||
/// This path updates only `partition_expr` (and its derived
|
||||
/// `partition_rule_version`) on current metadata and does not touch
|
||||
/// `sst_format`.
|
||||
pub fn apply_partition_expr_change(
|
||||
&mut self,
|
||||
manifest_version: ManifestVersion,
|
||||
change: RegionPartitionExprChange,
|
||||
) {
|
||||
if let Some(metadata) = &self.metadata {
|
||||
let mut metadata = metadata.as_ref().clone();
|
||||
metadata.set_partition_expr(change.partition_expr);
|
||||
self.metadata = Some(metadata.into());
|
||||
self.manifest_version = manifest_version;
|
||||
} else {
|
||||
warn!(
|
||||
"metadata is not set in region manifest builder, ignore partition expr change: {:?}",
|
||||
change
|
||||
);
|
||||
}
|
||||
}
|
||||
|
||||
pub fn apply_edit(&mut self, manifest_version: ManifestVersion, edit: RegionEdit) {
|
||||
self.manifest_version = manifest_version;
|
||||
|
||||
@@ -485,8 +522,14 @@ impl RegionMetaActionList {
|
||||
Self { actions }
|
||||
}
|
||||
|
||||
/// Split the actions into a region change and an edit.
|
||||
pub fn split_region_change_and_edit(self) -> (Option<RegionChange>, RegionEdit) {
|
||||
/// Split the actions into a partition expr change, a region change and an edit.
|
||||
pub fn split_region_change_and_edit(
|
||||
self,
|
||||
) -> (
|
||||
Option<RegionPartitionExprChange>,
|
||||
Option<RegionChange>,
|
||||
RegionEdit,
|
||||
) {
|
||||
let mut edit = RegionEdit {
|
||||
files_to_add: Vec::new(),
|
||||
files_to_remove: Vec::new(),
|
||||
@@ -496,9 +539,13 @@ impl RegionMetaActionList {
|
||||
flushed_sequence: None,
|
||||
committed_sequence: None,
|
||||
};
|
||||
let mut partition_expr_change = None;
|
||||
let mut region_change = None;
|
||||
for action in self.actions {
|
||||
match action {
|
||||
RegionMetaAction::PartitionExprChange(change) => {
|
||||
partition_expr_change = Some(change);
|
||||
}
|
||||
RegionMetaAction::Change(change) => {
|
||||
region_change = Some(change);
|
||||
}
|
||||
@@ -528,7 +575,7 @@ impl RegionMetaActionList {
|
||||
}
|
||||
}
|
||||
|
||||
(region_change, edit)
|
||||
(partition_expr_change, region_change, edit)
|
||||
}
|
||||
}
|
||||
|
||||
@@ -599,6 +646,12 @@ mod tests {
|
||||
|
||||
let region_remove = r#"{"region_id":42}"#;
|
||||
let _ = serde_json::from_str::<RegionRemove>(region_remove).unwrap();
|
||||
|
||||
let region_partition_expr_change = r#"{
|
||||
"partition_expr": "{\"expr\":\"x < 100\"}"
|
||||
}"#;
|
||||
let _ = serde_json::from_str::<RegionPartitionExprChange>(region_partition_expr_change)
|
||||
.unwrap();
|
||||
}
|
||||
|
||||
#[test]
|
||||
|
||||
@@ -300,6 +300,9 @@ impl RegionManifestManager {
|
||||
RegionMetaAction::Change(action) => {
|
||||
manifest_builder.apply_change(manifest_version, action);
|
||||
}
|
||||
RegionMetaAction::PartitionExprChange(action) => {
|
||||
manifest_builder.apply_partition_expr_change(manifest_version, action);
|
||||
}
|
||||
RegionMetaAction::Edit(action) => {
|
||||
manifest_builder.apply_edit(manifest_version, action);
|
||||
}
|
||||
@@ -442,6 +445,9 @@ impl RegionManifestManager {
|
||||
RegionMetaAction::Change(action) => {
|
||||
manifest_builder.apply_change(manifest_version, action);
|
||||
}
|
||||
RegionMetaAction::PartitionExprChange(action) => {
|
||||
manifest_builder.apply_partition_expr_change(manifest_version, action);
|
||||
}
|
||||
RegionMetaAction::Edit(action) => {
|
||||
manifest_builder.apply_edit(manifest_version, action);
|
||||
}
|
||||
@@ -548,6 +554,9 @@ impl RegionManifestManager {
|
||||
RegionMetaAction::Change(action) => {
|
||||
manifest_builder.apply_change(version, action);
|
||||
}
|
||||
RegionMetaAction::PartitionExprChange(action) => {
|
||||
manifest_builder.apply_partition_expr_change(version, action);
|
||||
}
|
||||
RegionMetaAction::Edit(action) => {
|
||||
manifest_builder.apply_edit(version, action);
|
||||
}
|
||||
|
||||
@@ -715,11 +715,21 @@ impl MitoRegion {
|
||||
}
|
||||
};
|
||||
let expect_change = merged_actions.actions.iter().any(|a| a.is_change());
|
||||
let expect_partition_expr_change = merged_actions
|
||||
.actions
|
||||
.iter()
|
||||
.any(|a| a.is_partition_expr_change());
|
||||
let expect_edit = merged_actions.actions.iter().any(|a| a.is_edit());
|
||||
ensure!(
|
||||
expect_change,
|
||||
!(expect_change && expect_partition_expr_change),
|
||||
UnexpectedSnafu {
|
||||
reason: "expect a change action in merged actions"
|
||||
reason: "unexpected both change and partition expr change actions in merged actions"
|
||||
}
|
||||
);
|
||||
ensure!(
|
||||
expect_change || expect_partition_expr_change,
|
||||
UnexpectedSnafu {
|
||||
reason: "expect a change or partition expr change action in merged actions"
|
||||
}
|
||||
);
|
||||
ensure!(
|
||||
@@ -729,20 +739,38 @@ impl MitoRegion {
|
||||
}
|
||||
);
|
||||
|
||||
let (merged_partition_expr_change, merged_change, merged_edit) =
|
||||
merged_actions.clone().split_region_change_and_edit();
|
||||
if let Some(change) = &merged_change {
|
||||
// In staging exit we only allow metadata-only updates. A `Change`
|
||||
// action is accepted only when column definitions are unchanged;
|
||||
// otherwise it is treated as a schema change and rejected.
|
||||
let current_column_metadatas = &self.version().metadata.column_metadatas;
|
||||
ensure!(
|
||||
change.metadata.column_metadatas == *current_column_metadatas,
|
||||
UnexpectedSnafu {
|
||||
reason: "change action alters column metadata in staging exit"
|
||||
}
|
||||
);
|
||||
}
|
||||
|
||||
// Submit merged actions using the manifest manager's update method
|
||||
// Pass the `false` so it saves to normal directory, not staging
|
||||
let new_version = manager.update(merged_actions.clone(), false).await?;
|
||||
|
||||
let new_version = manager.update(merged_actions, false).await?;
|
||||
info!(
|
||||
"Successfully submitted merged staged manifests for region {}, new version: {}",
|
||||
self.region_id, new_version
|
||||
);
|
||||
|
||||
// Apply the merged changes to in-memory version control
|
||||
let (merged_change, merged_edit) = merged_actions.split_region_change_and_edit();
|
||||
// Safety: we have already ensured that there is a change action in the merged actions.
|
||||
let new_metadata = merged_change.as_ref().unwrap().metadata.clone();
|
||||
self.version_control.alter_schema(new_metadata);
|
||||
if let Some(change) = merged_partition_expr_change {
|
||||
let mut new_metadata = self.version().metadata.as_ref().clone();
|
||||
new_metadata.set_partition_expr(change.partition_expr);
|
||||
self.version_control.alter_metadata(new_metadata.into());
|
||||
}
|
||||
if let Some(change) = merged_change {
|
||||
self.version_control.alter_metadata(change.metadata);
|
||||
}
|
||||
self.version_control
|
||||
.apply_edit(Some(merged_edit), &[], self.file_purger.clone());
|
||||
|
||||
@@ -1364,7 +1392,10 @@ mod tests {
|
||||
use store_api::storage::RegionId;
|
||||
|
||||
use crate::access_layer::AccessLayer;
|
||||
use crate::manifest::action::RegionMetaActionList;
|
||||
use crate::error::Error;
|
||||
use crate::manifest::action::{
|
||||
RegionChange, RegionEdit, RegionMetaAction, RegionMetaActionList, RegionPartitionExprChange,
|
||||
};
|
||||
use crate::manifest::manager::{RegionManifestManager, RegionManifestOptions};
|
||||
use crate::region::{
|
||||
ManifestContext, ManifestStats, MitoRegion, RegionLeaderState, RegionRoleState,
|
||||
@@ -1381,6 +1412,194 @@ mod tests {
|
||||
assert!(AtomicCell::<RegionRoleState>::is_lock_free());
|
||||
}
|
||||
|
||||
async fn build_test_region(env: &SchedulerEnv) -> MitoRegion {
|
||||
let builder = VersionControlBuilder::new();
|
||||
let version_control = Arc::new(builder.build());
|
||||
let metadata = version_control.current().version.metadata.clone();
|
||||
|
||||
let manager = RegionManifestManager::new(
|
||||
metadata.clone(),
|
||||
0,
|
||||
RegionManifestOptions {
|
||||
manifest_dir: "".to_string(),
|
||||
object_store: env.access_layer.object_store().clone(),
|
||||
compress_type: CompressionType::Uncompressed,
|
||||
checkpoint_distance: 10,
|
||||
remove_file_options: Default::default(),
|
||||
manifest_cache: None,
|
||||
},
|
||||
FormatType::PrimaryKey,
|
||||
&Default::default(),
|
||||
)
|
||||
.await
|
||||
.unwrap();
|
||||
|
||||
let manifest_ctx = Arc::new(ManifestContext::new(
|
||||
manager,
|
||||
RegionRoleState::Leader(RegionLeaderState::Writable),
|
||||
));
|
||||
|
||||
MitoRegion {
|
||||
region_id: metadata.region_id,
|
||||
version_control,
|
||||
access_layer: env.access_layer.clone(),
|
||||
manifest_ctx,
|
||||
file_purger: crate::test_util::new_noop_file_purger(),
|
||||
provider: Provider::noop_provider(),
|
||||
last_flush_millis: Default::default(),
|
||||
last_compaction_millis: Default::default(),
|
||||
time_provider: Arc::new(StdTimeProvider),
|
||||
topic_latest_entry_id: Default::default(),
|
||||
written_bytes: Arc::new(AtomicU64::new(0)),
|
||||
stats: ManifestStats::default(),
|
||||
staging_partition_info: Mutex::new(None),
|
||||
}
|
||||
}
|
||||
|
||||
fn empty_edit() -> RegionEdit {
|
||||
RegionEdit {
|
||||
files_to_add: Vec::new(),
|
||||
files_to_remove: Vec::new(),
|
||||
timestamp_ms: None,
|
||||
compaction_time_window: None,
|
||||
flushed_entry_id: None,
|
||||
flushed_sequence: None,
|
||||
committed_sequence: None,
|
||||
}
|
||||
}
|
||||
|
||||
#[tokio::test]
|
||||
async fn test_exit_staging_partition_expr_change_and_edit_success() {
|
||||
let env = SchedulerEnv::new().await;
|
||||
let region = build_test_region(&env).await;
|
||||
|
||||
let mut manager = region.manifest_ctx.manifest_manager.write().await;
|
||||
region.set_staging(&mut manager).await.unwrap();
|
||||
manager
|
||||
.update(
|
||||
RegionMetaActionList::new(vec![
|
||||
RegionMetaAction::PartitionExprChange(RegionPartitionExprChange {
|
||||
partition_expr: Some("expr_a".to_string()),
|
||||
}),
|
||||
RegionMetaAction::Edit(empty_edit()),
|
||||
]),
|
||||
true,
|
||||
)
|
||||
.await
|
||||
.unwrap();
|
||||
|
||||
region.exit_staging_on_success(&mut manager).await.unwrap();
|
||||
drop(manager);
|
||||
|
||||
assert_eq!(
|
||||
region.version().metadata.partition_expr.as_deref(),
|
||||
Some("expr_a")
|
||||
);
|
||||
assert_eq!(
|
||||
region.state(),
|
||||
RegionRoleState::Leader(RegionLeaderState::Writable)
|
||||
);
|
||||
}
|
||||
|
||||
#[tokio::test]
|
||||
async fn test_exit_staging_change_with_same_columns_success() {
|
||||
let env = SchedulerEnv::new().await;
|
||||
let region = build_test_region(&env).await;
|
||||
|
||||
let mut manager = region.manifest_ctx.manifest_manager.write().await;
|
||||
region.set_staging(&mut manager).await.unwrap();
|
||||
|
||||
let mut changed_metadata = region.version().metadata.as_ref().clone();
|
||||
changed_metadata.set_partition_expr(Some("expr_b".to_string()));
|
||||
|
||||
manager
|
||||
.update(
|
||||
RegionMetaActionList::new(vec![
|
||||
RegionMetaAction::Change(RegionChange {
|
||||
metadata: Arc::new(changed_metadata),
|
||||
sst_format: FormatType::PrimaryKey,
|
||||
}),
|
||||
RegionMetaAction::Edit(empty_edit()),
|
||||
]),
|
||||
true,
|
||||
)
|
||||
.await
|
||||
.unwrap();
|
||||
|
||||
region.exit_staging_on_success(&mut manager).await.unwrap();
|
||||
drop(manager);
|
||||
|
||||
assert_eq!(
|
||||
region.version().metadata.partition_expr.as_deref(),
|
||||
Some("expr_b")
|
||||
);
|
||||
assert_eq!(
|
||||
region.state(),
|
||||
RegionRoleState::Leader(RegionLeaderState::Writable)
|
||||
);
|
||||
}
|
||||
|
||||
#[tokio::test]
|
||||
async fn test_exit_staging_change_with_different_columns_fails() {
|
||||
let env = SchedulerEnv::new().await;
|
||||
let region = build_test_region(&env).await;
|
||||
|
||||
let mut manager = region.manifest_ctx.manifest_manager.write().await;
|
||||
region.set_staging(&mut manager).await.unwrap();
|
||||
|
||||
let mut changed_metadata = region.version().metadata.as_ref().clone();
|
||||
changed_metadata.column_metadatas.rotate_left(1);
|
||||
|
||||
manager
|
||||
.update(
|
||||
RegionMetaActionList::new(vec![
|
||||
RegionMetaAction::Change(RegionChange {
|
||||
metadata: Arc::new(changed_metadata),
|
||||
sst_format: FormatType::PrimaryKey,
|
||||
}),
|
||||
RegionMetaAction::Edit(empty_edit()),
|
||||
]),
|
||||
true,
|
||||
)
|
||||
.await
|
||||
.unwrap();
|
||||
|
||||
let result = region.exit_staging_on_success(&mut manager).await;
|
||||
assert!(matches!(result, Err(Error::Unexpected { .. })));
|
||||
}
|
||||
|
||||
#[tokio::test]
|
||||
async fn test_exit_staging_partition_expr_change_and_change_conflict_fails() {
|
||||
let env = SchedulerEnv::new().await;
|
||||
let region = build_test_region(&env).await;
|
||||
|
||||
let mut manager = region.manifest_ctx.manifest_manager.write().await;
|
||||
region.set_staging(&mut manager).await.unwrap();
|
||||
|
||||
let mut changed_metadata = region.version().metadata.as_ref().clone();
|
||||
changed_metadata.set_partition_expr(Some("expr_c".to_string()));
|
||||
|
||||
manager
|
||||
.update(
|
||||
RegionMetaActionList::new(vec![
|
||||
RegionMetaAction::PartitionExprChange(RegionPartitionExprChange {
|
||||
partition_expr: Some("expr_c".to_string()),
|
||||
}),
|
||||
RegionMetaAction::Change(RegionChange {
|
||||
metadata: Arc::new(changed_metadata),
|
||||
sst_format: FormatType::PrimaryKey,
|
||||
}),
|
||||
RegionMetaAction::Edit(empty_edit()),
|
||||
]),
|
||||
true,
|
||||
)
|
||||
.await
|
||||
.unwrap();
|
||||
|
||||
let result = region.exit_staging_on_success(&mut manager).await;
|
||||
assert!(matches!(result, Err(Error::Unexpected { .. })));
|
||||
}
|
||||
|
||||
#[tokio::test]
|
||||
async fn test_set_region_state() {
|
||||
let env = SchedulerEnv::new().await;
|
||||
|
||||
@@ -198,6 +198,19 @@ impl VersionControl {
|
||||
version_data.version = new_version;
|
||||
}
|
||||
|
||||
/// Alter metadata of the region without rebuilding memtables.
|
||||
pub(crate) fn alter_metadata(&self, metadata: RegionMetadataRef) {
|
||||
let version = self.current().version;
|
||||
let new_version = Arc::new(
|
||||
VersionBuilder::from_version(version)
|
||||
.metadata(metadata)
|
||||
.build(),
|
||||
);
|
||||
|
||||
let mut version_data = self.data.write().unwrap();
|
||||
version_data.version = new_version;
|
||||
}
|
||||
|
||||
/// Alter schema and format of the region.
|
||||
///
|
||||
/// It replaces existing mutable memtable with a memtable that uses the
|
||||
|
||||
@@ -12,7 +12,6 @@
|
||||
// See the License for the specific language governing permissions and
|
||||
// limitations under the License.
|
||||
|
||||
use std::sync::Arc;
|
||||
use std::time::Instant;
|
||||
|
||||
use common_base::hash::partition_rule_version;
|
||||
@@ -23,7 +22,7 @@ use store_api::storage::RegionId;
|
||||
|
||||
use crate::error::{RegionNotFoundSnafu, Result, StagingPartitionExprMismatchSnafu};
|
||||
use crate::flush::FlushReason;
|
||||
use crate::manifest::action::{RegionChange, RegionMetaAction, RegionMetaActionList};
|
||||
use crate::manifest::action::{RegionMetaAction, RegionMetaActionList, RegionPartitionExprChange};
|
||||
use crate::region::{MitoRegionRef, RegionLeaderState, StagingPartitionInfo};
|
||||
use crate::request::{
|
||||
BackgroundNotify, DdlRequest, EnterStagingResult, OptionOutputTx, SenderDdlRequest,
|
||||
@@ -125,14 +124,11 @@ impl<S: LogStore> RegionWorkerLoop<S> {
|
||||
}
|
||||
|
||||
// Second step: write new staging manifest.
|
||||
let mut new_meta = (*region.metadata()).clone();
|
||||
new_meta.set_partition_expr(Some(partition_expr.clone()));
|
||||
let sst_format = region.version().options.sst_format.unwrap_or_default();
|
||||
let change = RegionChange {
|
||||
metadata: Arc::new(new_meta),
|
||||
sst_format,
|
||||
let change = RegionPartitionExprChange {
|
||||
partition_expr: Some(partition_expr.clone()),
|
||||
};
|
||||
let action_list = RegionMetaActionList::with_action(RegionMetaAction::Change(change));
|
||||
let action_list =
|
||||
RegionMetaActionList::with_action(RegionMetaAction::PartitionExprChange(change));
|
||||
region
|
||||
.manifest_ctx
|
||||
.update_manifest(RegionLeaderState::EnteringStaging, action_list, true)
|
||||
|
||||
Reference in New Issue
Block a user