1mod alter_table;
16mod create_table;
17
18use std::sync::Arc;
19use std::time::Duration;
20
21use async_trait::async_trait;
22use clap::Parser;
23use client::api::v1::CreateTableExpr;
24use client::client_manager::NodeClients;
25use client::{DEFAULT_CATALOG_NAME, DEFAULT_SCHEMA_NAME};
26use common_error::ext::{BoxedError, ErrorExt};
27use common_error::status_code::StatusCode;
28use common_grpc::channel_manager::ChannelConfig;
29use common_meta::error::Error as CommonMetaError;
30use common_meta::key::TableMetadataManager;
31use common_meta::kv_backend::KvBackendRef;
32use common_meta::node_manager::NodeManagerRef;
33use common_meta::peer::Peer;
34use common_meta::rpc::router::{RegionRoute, find_leaders};
35use common_telemetry::{error, info, warn};
36use futures::TryStreamExt;
37use snafu::{ResultExt, ensure};
38use store_api::storage::TableId;
39
40use crate::Tool;
41use crate::common::StoreConfig;
42use crate::error::{
43 InvalidArgumentsSnafu, Result, SendRequestToDatanodeSnafu, TableMetadataSnafu, UnexpectedSnafu,
44};
45use crate::metadata::utils::{FullTableMetadata, IteratorInput, TableMetadataIterator};
46
47#[derive(Debug, Default, Parser)]
49pub struct RepairLogicalTablesCommand {
50 #[clap(long, value_delimiter = ',', alias = "table-name")]
52 table_names: Vec<String>,
53
54 #[clap(long, value_delimiter = ',', alias = "table-id")]
56 table_ids: Vec<TableId>,
57
58 #[clap(long, default_value = DEFAULT_SCHEMA_NAME)]
60 schema_name: String,
61
62 #[clap(long, default_value = DEFAULT_CATALOG_NAME)]
64 catalog_name: String,
65
66 #[clap(long)]
68 fail_fast: bool,
69
70 #[clap(flatten)]
71 store: StoreConfig,
72
73 #[clap(long, default_value_t = 30)]
75 client_timeout_secs: u64,
76
77 #[clap(long, default_value_t = 3)]
79 client_connect_timeout_secs: u64,
80}
81
82impl RepairLogicalTablesCommand {
83 fn validate(&self) -> Result<()> {
84 ensure!(
85 !self.table_names.is_empty() || !self.table_ids.is_empty(),
86 InvalidArgumentsSnafu {
87 msg: "You must specify --table-names or --table-ids.",
88 }
89 );
90 Ok(())
91 }
92}
93
94impl RepairLogicalTablesCommand {
95 pub async fn build(&self) -> std::result::Result<Box<dyn Tool>, BoxedError> {
96 self.validate().map_err(BoxedError::new)?;
97 let kv_backend = self.store.build().await?;
98 let node_client_channel_config = ChannelConfig::new()
99 .timeout(Duration::from_secs(self.client_timeout_secs))
100 .connect_timeout(Duration::from_secs(self.client_connect_timeout_secs));
101 let node_manager = Arc::new(NodeClients::new(node_client_channel_config));
102
103 Ok(Box::new(RepairTool {
104 table_names: self.table_names.clone(),
105 table_ids: self.table_ids.clone(),
106 schema_name: self.schema_name.clone(),
107 catalog_name: self.catalog_name.clone(),
108 fail_fast: self.fail_fast,
109 kv_backend,
110 node_manager,
111 }))
112 }
113}
114
115struct RepairTool {
116 table_names: Vec<String>,
117 table_ids: Vec<TableId>,
118 schema_name: String,
119 catalog_name: String,
120 fail_fast: bool,
121 kv_backend: KvBackendRef,
122 node_manager: NodeManagerRef,
123}
124
125#[async_trait]
126impl Tool for RepairTool {
127 async fn do_work(&self) -> std::result::Result<(), BoxedError> {
128 self.repair_tables().await.map_err(BoxedError::new)
129 }
130}
131
132impl RepairTool {
133 fn generate_iterator_input(&self) -> Result<IteratorInput> {
134 if !self.table_names.is_empty() {
135 let table_names = &self.table_names;
136 let catalog = &self.catalog_name;
137 let schema_name = &self.schema_name;
138
139 let table_names = table_names
140 .iter()
141 .map(|table_name| (catalog.clone(), schema_name.clone(), table_name.clone()))
142 .collect::<Vec<_>>();
143 return Ok(IteratorInput::new_table_names(table_names));
144 } else if !self.table_ids.is_empty() {
145 return Ok(IteratorInput::new_table_ids(self.table_ids.clone()));
146 };
147
148 InvalidArgumentsSnafu {
149 msg: "You must specify --table-names or --table-id.",
150 }
151 .fail()
152 }
153
154 async fn repair_tables(&self) -> Result<()> {
155 let input = self.generate_iterator_input()?;
156 let mut table_metadata_iterator =
157 Box::pin(TableMetadataIterator::new(self.kv_backend.clone(), input).into_stream());
158 let table_metadata_manager = TableMetadataManager::new(self.kv_backend.clone());
159
160 let mut skipped_table = 0;
161 let mut success_table = 0;
162 while let Some(full_table_metadata) = table_metadata_iterator.try_next().await? {
163 let full_table_name = full_table_metadata.full_table_name();
164 if !full_table_metadata.is_metric_engine() {
165 warn!(
166 "Skipping repair for non-metric engine table: {}",
167 full_table_name
168 );
169 skipped_table += 1;
170 continue;
171 }
172
173 if full_table_metadata.is_physical_table() {
174 warn!("Skipping repair for physical table: {}", full_table_name);
175 skipped_table += 1;
176 continue;
177 }
178
179 let (physical_table_id, physical_table_route) = table_metadata_manager
180 .table_route_manager()
181 .get_physical_table_route(full_table_metadata.table_id)
182 .await
183 .context(TableMetadataSnafu)?;
184
185 if let Err(err) = self
186 .repair_table(
187 &full_table_metadata,
188 physical_table_id,
189 &physical_table_route.region_routes,
190 )
191 .await
192 {
193 error!(
194 err;
195 "Failed to repair table: {}, skipped table: {}",
196 full_table_name,
197 skipped_table,
198 );
199
200 if self.fail_fast {
201 return Err(err);
202 }
203 } else {
204 success_table += 1;
205 }
206 }
207
208 info!(
209 "Repair logical tables result: {} tables repaired, {} tables skipped",
210 success_table, skipped_table
211 );
212
213 Ok(())
214 }
215
216 async fn alter_table_on_datanodes(
217 &self,
218 full_table_metadata: &FullTableMetadata,
219 physical_region_routes: &[RegionRoute],
220 ) -> Result<Vec<(Peer, CommonMetaError)>> {
221 let logical_table_id = full_table_metadata.table_id;
222 let alter_table_expr = alter_table::generate_alter_table_expr_for_all_columns(
223 &full_table_metadata.table_info,
224 )?;
225 let node_manager = self.node_manager.clone();
226
227 let mut failed_peers = Vec::new();
228 info!(
229 "Sending alter table requests to all datanodes for table: {}, number of regions:{}.",
230 full_table_metadata.full_table_name(),
231 physical_region_routes.len()
232 );
233 let leaders = find_leaders(physical_region_routes);
234 for peer in &leaders {
235 let alter_table_request = alter_table::make_alter_region_request_for_peer(
236 logical_table_id,
237 &alter_table_expr,
238 peer,
239 physical_region_routes,
240 )?;
241 let datanode = node_manager.datanode(peer).await;
242 if let Err(err) = datanode.handle(alter_table_request).await {
243 failed_peers.push((peer.clone(), err));
244 }
245 }
246
247 Ok(failed_peers)
248 }
249
250 async fn create_table_on_datanode(
251 &self,
252 create_table_expr: &CreateTableExpr,
253 logical_table_id: TableId,
254 physical_table_id: TableId,
255 peer: &Peer,
256 physical_region_routes: &[RegionRoute],
257 ) -> Result<()> {
258 let node_manager = self.node_manager.clone();
259 let datanode = node_manager.datanode(peer).await;
260 let create_table_request = create_table::make_create_region_request_for_peer(
261 logical_table_id,
262 physical_table_id,
263 create_table_expr,
264 peer,
265 physical_region_routes,
266 )?;
267
268 datanode
269 .handle(create_table_request)
270 .await
271 .with_context(|_| SendRequestToDatanodeSnafu { peer: peer.clone() })?;
272
273 Ok(())
274 }
275
276 async fn repair_table(
277 &self,
278 full_table_metadata: &FullTableMetadata,
279 physical_table_id: TableId,
280 physical_region_routes: &[RegionRoute],
281 ) -> Result<()> {
282 let full_table_name = full_table_metadata.full_table_name();
283 let failed_peers = self
285 .alter_table_on_datanodes(full_table_metadata, physical_region_routes)
286 .await?;
287
288 if failed_peers.is_empty() {
289 info!(
290 "All alter table requests sent successfully for table: {}",
291 full_table_name
292 );
293 return Ok(());
294 }
295 warn!(
296 "Sending alter table requests to datanodes for table: {} failed for the datanodes: {:?}",
297 full_table_name,
298 failed_peers
299 .iter()
300 .map(|(peer, _)| peer.id)
301 .collect::<Vec<_>>()
302 );
303
304 let create_table_expr =
305 create_table::generate_create_table_expr(&full_table_metadata.table_info)?;
306
307 let mut errors = Vec::new();
308 for (peer, err) in failed_peers {
309 if err.status_code() != StatusCode::RegionNotFound {
310 error!(
311 err;
312 "Sending alter table requests to datanode: {} for table: {} failed",
313 peer.id,
314 full_table_name,
315 );
316 continue;
317 }
318 info!(
319 "Region not found for table: {}, datanode: {}, trying to create the logical table on that datanode",
320 full_table_name, peer.id
321 );
322
323 if let Err(err) = self
326 .create_table_on_datanode(
327 &create_table_expr,
328 full_table_metadata.table_id,
329 physical_table_id,
330 &peer,
331 physical_region_routes,
332 )
333 .await
334 {
335 error!(
336 err;
337 "Failed to create table on datanode: {} for table: {}",
338 peer.id, full_table_name
339 );
340 errors.push(err);
341 if self.fail_fast {
342 break;
343 }
344 } else {
345 info!(
346 "Created table on datanode: {} for table: {}",
347 peer.id, full_table_name
348 );
349 }
350 }
351
352 if !errors.is_empty() {
353 return UnexpectedSnafu {
354 msg: format!(
355 "Failed to create table on datanodes for table: {}",
356 full_table_name,
357 ),
358 }
359 .fail();
360 }
361
362 Ok(())
363 }
364}