flow/
server.rs

1// Copyright 2023 Greptime Team
2//
3// Licensed under the Apache License, Version 2.0 (the "License");
4// you may not use this file except in compliance with the License.
5// You may obtain a copy of the License at
6//
7//     http://www.apache.org/licenses/LICENSE-2.0
8//
9// Unless required by applicable law or agreed to in writing, software
10// distributed under the License is distributed on an "AS IS" BASIS,
11// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
12// See the License for the specific language governing permissions and
13// limitations under the License.
14
15//! Implementation of grpc service for flow node
16
17use std::net::SocketAddr;
18use std::sync::Arc;
19
20use api::v1::flow::DirtyWindowRequests;
21use api::v1::{RowDeleteRequests, RowInsertRequests};
22use cache::{TABLE_FLOWNODE_SET_CACHE_NAME, TABLE_ROUTE_CACHE_NAME};
23use catalog::CatalogManagerRef;
24use common_base::Plugins;
25use common_error::ext::BoxedError;
26use common_meta::cache::{LayeredCacheRegistryRef, TableFlownodeSetCacheRef, TableRouteCacheRef};
27use common_meta::ddl::ProcedureExecutorRef;
28use common_meta::key::flow::FlowMetadataManagerRef;
29use common_meta::key::TableMetadataManagerRef;
30use common_meta::kv_backend::KvBackendRef;
31use common_meta::node_manager::{Flownode, NodeManagerRef};
32use common_query::Output;
33use common_runtime::JoinHandle;
34use common_telemetry::tracing::info;
35use futures::TryStreamExt;
36use greptime_proto::v1::flow::{flow_server, FlowRequest, FlowResponse, InsertRequests};
37use itertools::Itertools;
38use operator::delete::Deleter;
39use operator::insert::Inserter;
40use operator::statement::StatementExecutor;
41use partition::manager::PartitionRuleManager;
42use query::{QueryEngine, QueryEngineFactory};
43use servers::add_service;
44use servers::grpc::builder::GrpcServerBuilder;
45use servers::grpc::{GrpcServer, GrpcServerConfig};
46use servers::http::HttpServerBuilder;
47use servers::metrics_handler::MetricsHandler;
48use servers::server::{ServerHandler, ServerHandlers};
49use session::context::QueryContextRef;
50use snafu::{OptionExt, ResultExt};
51use tokio::sync::{broadcast, oneshot, Mutex};
52use tonic::codec::CompressionEncoding;
53use tonic::{Request, Response, Status};
54
55use crate::adapter::flownode_impl::{FlowDualEngine, FlowDualEngineRef};
56use crate::adapter::{create_worker, FlowStreamingEngineRef};
57use crate::batching_mode::engine::BatchingEngine;
58use crate::error::{
59    to_status_with_last_err, CacheRequiredSnafu, ExternalSnafu, IllegalAuthConfigSnafu,
60    ListFlowsSnafu, ParseAddrSnafu, ShutdownServerSnafu, StartServerSnafu, UnexpectedSnafu,
61};
62use crate::heartbeat::HeartbeatTask;
63use crate::metrics::{METRIC_FLOW_PROCESSING_TIME, METRIC_FLOW_ROWS};
64use crate::transform::register_function_to_query_engine;
65use crate::utils::{SizeReportSender, StateReportHandler};
66use crate::{Error, FlowAuthHeader, FlownodeOptions, FrontendClient, StreamingEngine};
67
68pub const FLOW_NODE_SERVER_NAME: &str = "FLOW_NODE_SERVER";
69/// wrapping flow node manager to avoid orphan rule with Arc<...>
70#[derive(Clone)]
71pub struct FlowService {
72    pub dual_engine: FlowDualEngineRef,
73}
74
75impl FlowService {
76    pub fn new(manager: FlowDualEngineRef) -> Self {
77        Self {
78            dual_engine: manager,
79        }
80    }
81}
82
83#[async_trait::async_trait]
84impl flow_server::Flow for FlowService {
85    async fn handle_create_remove(
86        &self,
87        request: Request<FlowRequest>,
88    ) -> Result<Response<FlowResponse>, Status> {
89        let _timer = METRIC_FLOW_PROCESSING_TIME
90            .with_label_values(&["ddl"])
91            .start_timer();
92
93        let request = request.into_inner();
94        self.dual_engine
95            .handle(request)
96            .await
97            .map_err(|err| {
98                common_telemetry::error!(err; "Failed to handle flow request");
99                err
100            })
101            .map(Response::new)
102            .map_err(to_status_with_last_err)
103    }
104
105    async fn handle_mirror_request(
106        &self,
107        request: Request<InsertRequests>,
108    ) -> Result<Response<FlowResponse>, Status> {
109        let _timer = METRIC_FLOW_PROCESSING_TIME
110            .with_label_values(&["insert"])
111            .start_timer();
112
113        let request = request.into_inner();
114        // TODO(discord9): fix protobuf import order shenanigans to remove this duplicated define
115        let mut row_count = 0;
116        let request = api::v1::region::InsertRequests {
117            requests: request
118                .requests
119                .into_iter()
120                .map(|insert| {
121                    insert.rows.as_ref().inspect(|x| row_count += x.rows.len());
122                    api::v1::region::InsertRequest {
123                        region_id: insert.region_id,
124                        rows: insert.rows,
125                    }
126                })
127                .collect_vec(),
128        };
129
130        METRIC_FLOW_ROWS
131            .with_label_values(&["in"])
132            .inc_by(row_count as u64);
133
134        self.dual_engine
135            .handle_inserts(request)
136            .await
137            .map(Response::new)
138            .map_err(to_status_with_last_err)
139    }
140
141    async fn handle_mark_dirty_time_window(
142        &self,
143        reqs: Request<DirtyWindowRequests>,
144    ) -> Result<Response<FlowResponse>, Status> {
145        self.dual_engine
146            .batching_engine()
147            .handle_mark_dirty_time_window(reqs.into_inner())
148            .await
149            .map(Response::new)
150            .map_err(to_status_with_last_err)
151    }
152}
153
154#[derive(Clone)]
155pub struct FlownodeServer {
156    inner: Arc<FlownodeServerInner>,
157}
158
159/// FlownodeServerInner is the inner state of FlownodeServer,
160/// this struct mostly useful for construct/start and stop the
161/// flow node server
162struct FlownodeServerInner {
163    /// worker shutdown signal, not to be confused with server_shutdown_tx
164    worker_shutdown_tx: Mutex<broadcast::Sender<()>>,
165    /// server shutdown signal for shutdown grpc server
166    server_shutdown_tx: Mutex<broadcast::Sender<()>>,
167    /// streaming task handler
168    streaming_task_handler: Mutex<Option<JoinHandle<()>>>,
169    flow_service: FlowService,
170}
171
172impl FlownodeServer {
173    pub fn new(flow_service: FlowService) -> Self {
174        let (tx, _rx) = broadcast::channel::<()>(1);
175        let (server_tx, _server_rx) = broadcast::channel::<()>(1);
176        Self {
177            inner: Arc::new(FlownodeServerInner {
178                flow_service,
179                worker_shutdown_tx: Mutex::new(tx),
180                server_shutdown_tx: Mutex::new(server_tx),
181                streaming_task_handler: Mutex::new(None),
182            }),
183        }
184    }
185
186    /// Start the background task for streaming computation.
187    ///
188    /// Should be called only after heartbeat is establish, hence can get cluster info
189    async fn start_workers(&self) -> Result<(), Error> {
190        let manager_ref = self.inner.flow_service.dual_engine.clone();
191        let handle = manager_ref
192            .streaming_engine()
193            .run_background(Some(self.inner.worker_shutdown_tx.lock().await.subscribe()));
194        self.inner
195            .streaming_task_handler
196            .lock()
197            .await
198            .replace(handle);
199
200        self.inner
201            .flow_service
202            .dual_engine
203            .start_flow_consistent_check_task()
204            .await?;
205
206        Ok(())
207    }
208
209    /// Stop the background task for streaming computation.
210    async fn stop_workers(&self) -> Result<(), Error> {
211        let tx = self.inner.worker_shutdown_tx.lock().await;
212        if tx.send(()).is_err() {
213            info!("Receiver dropped, the flow node server has already shutdown");
214        }
215        self.inner
216            .flow_service
217            .dual_engine
218            .stop_flow_consistent_check_task()
219            .await?;
220        Ok(())
221    }
222}
223
224impl FlownodeServer {
225    pub fn create_flow_service(&self) -> flow_server::FlowServer<impl flow_server::Flow> {
226        flow_server::FlowServer::new(self.inner.flow_service.clone())
227            .accept_compressed(CompressionEncoding::Gzip)
228            .send_compressed(CompressionEncoding::Gzip)
229            .accept_compressed(CompressionEncoding::Zstd)
230            .send_compressed(CompressionEncoding::Zstd)
231    }
232}
233
234/// The flownode server instance.
235pub struct FlownodeInstance {
236    flownode_server: FlownodeServer,
237    services: ServerHandlers,
238    heartbeat_task: Option<HeartbeatTask>,
239}
240
241impl FlownodeInstance {
242    pub async fn start(&mut self) -> Result<(), crate::Error> {
243        if let Some(task) = &self.heartbeat_task {
244            task.start().await?;
245        }
246
247        self.flownode_server.start_workers().await?;
248
249        self.services.start_all().await.context(StartServerSnafu)?;
250
251        Ok(())
252    }
253    pub async fn shutdown(&mut self) -> Result<(), Error> {
254        self.services
255            .shutdown_all()
256            .await
257            .context(ShutdownServerSnafu)?;
258
259        self.flownode_server.stop_workers().await?;
260
261        if let Some(task) = &self.heartbeat_task {
262            task.shutdown();
263        }
264
265        Ok(())
266    }
267
268    pub fn flownode_server(&self) -> &FlownodeServer {
269        &self.flownode_server
270    }
271
272    pub fn flow_engine(&self) -> FlowDualEngineRef {
273        self.flownode_server.inner.flow_service.dual_engine.clone()
274    }
275
276    pub fn setup_services(&mut self, services: ServerHandlers) {
277        self.services = services;
278    }
279}
280
281pub fn get_flow_auth_options(fn_opts: &FlownodeOptions) -> Result<Option<FlowAuthHeader>, Error> {
282    if let Some(user_provider) = fn_opts.user_provider.as_ref() {
283        let static_provider = auth::static_user_provider_from_option(user_provider)
284            .context(IllegalAuthConfigSnafu)?;
285
286        let (usr, pwd) = static_provider
287            .get_one_user_pwd()
288            .context(IllegalAuthConfigSnafu)?;
289        let auth_header = FlowAuthHeader::from_user_pwd(&usr, &pwd);
290        return Ok(Some(auth_header));
291    }
292
293    Ok(None)
294}
295
296/// [`FlownodeInstance`] Builder
297pub struct FlownodeBuilder {
298    opts: FlownodeOptions,
299    plugins: Plugins,
300    table_meta: TableMetadataManagerRef,
301    catalog_manager: CatalogManagerRef,
302    flow_metadata_manager: FlowMetadataManagerRef,
303    heartbeat_task: Option<HeartbeatTask>,
304    /// receive a oneshot sender to send state size report
305    state_report_handler: Option<StateReportHandler>,
306    frontend_client: Arc<FrontendClient>,
307}
308
309impl FlownodeBuilder {
310    /// init flownode builder
311    pub fn new(
312        opts: FlownodeOptions,
313        plugins: Plugins,
314        table_meta: TableMetadataManagerRef,
315        catalog_manager: CatalogManagerRef,
316        flow_metadata_manager: FlowMetadataManagerRef,
317        frontend_client: Arc<FrontendClient>,
318    ) -> Self {
319        Self {
320            opts,
321            plugins,
322            table_meta,
323            catalog_manager,
324            flow_metadata_manager,
325            heartbeat_task: None,
326            state_report_handler: None,
327            frontend_client,
328        }
329    }
330
331    pub fn with_heartbeat_task(self, heartbeat_task: HeartbeatTask) -> Self {
332        let (sender, receiver) = SizeReportSender::new();
333        Self {
334            heartbeat_task: Some(heartbeat_task.with_query_stat_size(sender)),
335            state_report_handler: Some(receiver),
336            ..self
337        }
338    }
339
340    pub async fn build(mut self) -> Result<FlownodeInstance, Error> {
341        // TODO(discord9): does this query engine need those?
342        let query_engine_factory = QueryEngineFactory::new_with_plugins(
343            // query engine in flownode is only used for translate plan with resolved table source.
344            self.catalog_manager.clone(),
345            None,
346            None,
347            None,
348            None,
349            false,
350            Default::default(),
351            self.opts.query.clone(),
352        );
353        let manager = Arc::new(
354            self.build_manager(query_engine_factory.query_engine())
355                .await?,
356        );
357        let batching = Arc::new(BatchingEngine::new(
358            self.frontend_client.clone(),
359            query_engine_factory.query_engine(),
360            self.flow_metadata_manager.clone(),
361            self.table_meta.clone(),
362            self.catalog_manager.clone(),
363        ));
364        let dual = FlowDualEngine::new(
365            manager.clone(),
366            batching,
367            self.flow_metadata_manager.clone(),
368            self.catalog_manager.clone(),
369            self.plugins.clone(),
370        );
371
372        let server = FlownodeServer::new(FlowService::new(Arc::new(dual)));
373
374        let heartbeat_task = self.heartbeat_task;
375
376        let instance = FlownodeInstance {
377            flownode_server: server,
378            services: ServerHandlers::default(),
379            heartbeat_task,
380        };
381        Ok(instance)
382    }
383
384    /// build [`FlowWorkerManager`], note this doesn't take ownership of `self`,
385    /// nor does it actually start running the worker.
386    async fn build_manager(
387        &mut self,
388        query_engine: Arc<dyn QueryEngine>,
389    ) -> Result<StreamingEngine, Error> {
390        let table_meta = self.table_meta.clone();
391
392        register_function_to_query_engine(&query_engine);
393
394        let num_workers = self.opts.flow.num_workers;
395
396        let node_id = self.opts.node_id.map(|id| id as u32);
397
398        let mut man = StreamingEngine::new(node_id, query_engine, table_meta);
399        for worker_id in 0..num_workers {
400            let (tx, rx) = oneshot::channel();
401
402            let _handle = std::thread::Builder::new()
403                .name(format!("flow-worker-{}", worker_id))
404                .spawn(move || {
405                    let (handle, mut worker) = create_worker();
406                    let _ = tx.send(handle);
407                    info!("Flow Worker started in new thread");
408                    worker.run();
409                });
410            let worker_handle = rx.await.map_err(|e| {
411                UnexpectedSnafu {
412                    reason: format!("Failed to receive worker handle: {}", e),
413                }
414                .build()
415            })?;
416            man.add_worker_handle(worker_handle);
417        }
418        if let Some(handler) = self.state_report_handler.take() {
419            man = man.with_state_report_handler(handler).await;
420        }
421        info!("Flow Node Manager started");
422        Ok(man)
423    }
424}
425
426/// Useful in distributed mode
427pub struct FlownodeServiceBuilder<'a> {
428    opts: &'a FlownodeOptions,
429    grpc_server: Option<GrpcServer>,
430    enable_http_service: bool,
431}
432
433impl<'a> FlownodeServiceBuilder<'a> {
434    pub fn new(opts: &'a FlownodeOptions) -> Self {
435        Self {
436            opts,
437            grpc_server: None,
438            enable_http_service: false,
439        }
440    }
441
442    pub fn enable_http_service(self) -> Self {
443        Self {
444            enable_http_service: true,
445            ..self
446        }
447    }
448
449    pub fn with_grpc_server(self, grpc_server: GrpcServer) -> Self {
450        Self {
451            grpc_server: Some(grpc_server),
452            ..self
453        }
454    }
455
456    pub fn with_default_grpc_server(mut self, flownode_server: &FlownodeServer) -> Self {
457        let grpc_server = Self::grpc_server_builder(self.opts, flownode_server).build();
458        self.grpc_server = Some(grpc_server);
459        self
460    }
461
462    pub fn build(mut self) -> Result<ServerHandlers, Error> {
463        let handlers = ServerHandlers::default();
464        if let Some(grpc_server) = self.grpc_server.take() {
465            let addr: SocketAddr = self.opts.grpc.bind_addr.parse().context(ParseAddrSnafu {
466                addr: &self.opts.grpc.bind_addr,
467            })?;
468            let handler: ServerHandler = (Box::new(grpc_server), addr);
469            handlers.insert(handler);
470        }
471
472        if self.enable_http_service {
473            let http_server = HttpServerBuilder::new(self.opts.http.clone())
474                .with_metrics_handler(MetricsHandler)
475                .build();
476            let addr: SocketAddr = self.opts.http.addr.parse().context(ParseAddrSnafu {
477                addr: &self.opts.http.addr,
478            })?;
479            let handler: ServerHandler = (Box::new(http_server), addr);
480            handlers.insert(handler);
481        }
482        Ok(handlers)
483    }
484
485    pub fn grpc_server_builder(
486        opts: &FlownodeOptions,
487        flownode_server: &FlownodeServer,
488    ) -> GrpcServerBuilder {
489        let config = GrpcServerConfig {
490            max_recv_message_size: opts.grpc.max_recv_message_size.as_bytes() as usize,
491            max_send_message_size: opts.grpc.max_send_message_size.as_bytes() as usize,
492            tls: opts.grpc.tls.clone(),
493        };
494        let service = flownode_server.create_flow_service();
495        let runtime = common_runtime::global_runtime();
496        let mut builder = GrpcServerBuilder::new(config, runtime);
497        add_service!(builder, service);
498        builder
499    }
500}
501
502/// Basically a tiny frontend that communicates with datanode, different from [`FrontendClient`] which
503/// connect to a real frontend instead, this is used for flow's streaming engine. And is for simple query.
504///
505/// For heavy query use [`FrontendClient`] which offload computation to frontend, lifting the load from flownode
506#[derive(Clone)]
507pub struct FrontendInvoker {
508    inserter: Arc<Inserter>,
509    deleter: Arc<Deleter>,
510    statement_executor: Arc<StatementExecutor>,
511}
512
513impl FrontendInvoker {
514    pub fn new(
515        inserter: Arc<Inserter>,
516        deleter: Arc<Deleter>,
517        statement_executor: Arc<StatementExecutor>,
518    ) -> Self {
519        Self {
520            inserter,
521            deleter,
522            statement_executor,
523        }
524    }
525
526    pub async fn build_from(
527        flow_streaming_engine: FlowStreamingEngineRef,
528        catalog_manager: CatalogManagerRef,
529        kv_backend: KvBackendRef,
530        layered_cache_registry: LayeredCacheRegistryRef,
531        procedure_executor: ProcedureExecutorRef,
532        node_manager: NodeManagerRef,
533    ) -> Result<FrontendInvoker, Error> {
534        let table_route_cache: TableRouteCacheRef =
535            layered_cache_registry.get().context(CacheRequiredSnafu {
536                name: TABLE_ROUTE_CACHE_NAME,
537            })?;
538
539        let partition_manager = Arc::new(PartitionRuleManager::new(
540            kv_backend.clone(),
541            table_route_cache.clone(),
542        ));
543
544        let table_flownode_cache: TableFlownodeSetCacheRef =
545            layered_cache_registry.get().context(CacheRequiredSnafu {
546                name: TABLE_FLOWNODE_SET_CACHE_NAME,
547            })?;
548
549        let inserter = Arc::new(Inserter::new(
550            catalog_manager.clone(),
551            partition_manager.clone(),
552            node_manager.clone(),
553            table_flownode_cache,
554        ));
555
556        let deleter = Arc::new(Deleter::new(
557            catalog_manager.clone(),
558            partition_manager.clone(),
559            node_manager.clone(),
560        ));
561
562        let query_engine = flow_streaming_engine.query_engine.clone();
563
564        let statement_executor = Arc::new(StatementExecutor::new(
565            catalog_manager.clone(),
566            query_engine.clone(),
567            procedure_executor.clone(),
568            kv_backend.clone(),
569            layered_cache_registry.clone(),
570            inserter.clone(),
571            table_route_cache,
572            None,
573        ));
574
575        let invoker = FrontendInvoker::new(inserter, deleter, statement_executor);
576        Ok(invoker)
577    }
578}
579
580impl FrontendInvoker {
581    pub async fn row_inserts(
582        &self,
583        requests: RowInsertRequests,
584        ctx: QueryContextRef,
585    ) -> common_frontend::error::Result<Output> {
586        let _timer = METRIC_FLOW_PROCESSING_TIME
587            .with_label_values(&["output_insert"])
588            .start_timer();
589
590        self.inserter
591            .handle_row_inserts(requests, ctx, &self.statement_executor, false, false)
592            .await
593            .map_err(BoxedError::new)
594            .context(common_frontend::error::ExternalSnafu)
595    }
596
597    pub async fn row_deletes(
598        &self,
599        requests: RowDeleteRequests,
600        ctx: QueryContextRef,
601    ) -> common_frontend::error::Result<Output> {
602        let _timer = METRIC_FLOW_PROCESSING_TIME
603            .with_label_values(&["output_delete"])
604            .start_timer();
605
606        self.deleter
607            .handle_row_deletes(requests, ctx)
608            .await
609            .map_err(BoxedError::new)
610            .context(common_frontend::error::ExternalSnafu)
611    }
612
613    pub fn statement_executor(&self) -> Arc<StatementExecutor> {
614        self.statement_executor.clone()
615    }
616}
617
618/// get all flow ids in this flownode
619pub(crate) async fn get_all_flow_ids(
620    flow_metadata_manager: &FlowMetadataManagerRef,
621    catalog_manager: &CatalogManagerRef,
622    nodeid: Option<u64>,
623) -> Result<Vec<u32>, Error> {
624    let ret = if let Some(nodeid) = nodeid {
625        let flow_ids_one_node = flow_metadata_manager
626            .flownode_flow_manager()
627            .flows(nodeid)
628            .try_collect::<Vec<_>>()
629            .await
630            .context(ListFlowsSnafu { id: Some(nodeid) })?;
631        flow_ids_one_node.into_iter().map(|(id, _)| id).collect()
632    } else {
633        let all_catalogs = catalog_manager
634            .catalog_names()
635            .await
636            .map_err(BoxedError::new)
637            .context(ExternalSnafu)?;
638        let mut all_flow_ids = vec![];
639        for catalog in all_catalogs {
640            let flows = flow_metadata_manager
641                .flow_name_manager()
642                .flow_names(&catalog)
643                .await
644                .try_collect::<Vec<_>>()
645                .await
646                .map_err(BoxedError::new)
647                .context(ExternalSnafu)?;
648
649            all_flow_ids.extend(flows.into_iter().map(|(_, id)| id.flow_id()));
650        }
651        all_flow_ids
652    };
653
654    Ok(ret)
655}
flow/server.rs

flow/
server.rs
