flow/
adapter.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//! for getting data from source and sending results to sink
16//! and communicating with other parts of the database
17#![warn(unused_imports)]
18
19use std::collections::BTreeMap;
20use std::sync::Arc;
21use std::time::{Duration, Instant, SystemTime};
22
23use api::v1::{RowDeleteRequest, RowDeleteRequests, RowInsertRequest, RowInsertRequests};
24use common_config::Configurable;
25use common_error::ext::BoxedError;
26use common_meta::key::TableMetadataManagerRef;
27use common_runtime::JoinHandle;
28use common_telemetry::logging::{LoggingOptions, TracingOptions};
29use common_telemetry::{debug, info, trace};
30use datatypes::schema::ColumnSchema;
31use datatypes::value::Value;
32use greptime_proto::v1;
33use itertools::{EitherOrBoth, Itertools};
34use meta_client::MetaClientOptions;
35use query::options::QueryOptions;
36use query::QueryEngine;
37use serde::{Deserialize, Serialize};
38use servers::grpc::GrpcOptions;
39use servers::heartbeat_options::HeartbeatOptions;
40use servers::http::HttpOptions;
41use session::context::QueryContext;
42use snafu::{ensure, OptionExt, ResultExt};
43use store_api::storage::{ConcreteDataType, RegionId};
44use table::metadata::TableId;
45use tokio::sync::broadcast::error::TryRecvError;
46use tokio::sync::{broadcast, watch, Mutex, RwLock};
47
48pub(crate) use crate::adapter::node_context::FlownodeContext;
49use crate::adapter::refill::RefillTask;
50use crate::adapter::table_source::ManagedTableSource;
51use crate::adapter::util::relation_desc_to_column_schemas_with_fallback;
52pub(crate) use crate::adapter::worker::{create_worker, Worker, WorkerHandle};
53use crate::compute::ErrCollector;
54use crate::df_optimizer::sql_to_flow_plan;
55use crate::error::{EvalSnafu, ExternalSnafu, InternalSnafu, InvalidQuerySnafu, UnexpectedSnafu};
56use crate::expr::Batch;
57use crate::metrics::{METRIC_FLOW_INSERT_ELAPSED, METRIC_FLOW_ROWS, METRIC_FLOW_RUN_INTERVAL_MS};
58use crate::repr::{self, DiffRow, RelationDesc, Row, BATCH_SIZE};
59use crate::{CreateFlowArgs, FlowId, TableName};
60
61pub(crate) mod flownode_impl;
62mod parse_expr;
63pub(crate) mod refill;
64mod stat;
65#[cfg(test)]
66mod tests;
67pub(crate) mod util;
68mod worker;
69
70pub(crate) mod node_context;
71pub(crate) mod table_source;
72
73use crate::error::Error;
74use crate::utils::StateReportHandler;
75use crate::FrontendInvoker;
76
77// `GREPTIME_TIMESTAMP` is not used to distinguish when table is created automatically by flow
78pub const AUTO_CREATED_PLACEHOLDER_TS_COL: &str = "__ts_placeholder";
79
80pub const AUTO_CREATED_UPDATE_AT_TS_COL: &str = "update_at";
81
82/// Flow config that exists both in standalone&distributed mode
83#[derive(Clone, Debug, Serialize, Deserialize, PartialEq)]
84#[serde(default)]
85pub struct FlowConfig {
86    pub num_workers: usize,
87}
88
89impl Default for FlowConfig {
90    fn default() -> Self {
91        Self {
92            num_workers: (common_config::utils::get_cpus() / 2).max(1),
93        }
94    }
95}
96
97/// Options for flow node
98#[derive(Clone, Debug, PartialEq, Serialize, Deserialize)]
99#[serde(default)]
100pub struct FlownodeOptions {
101    pub node_id: Option<u64>,
102    pub flow: FlowConfig,
103    pub grpc: GrpcOptions,
104    pub http: HttpOptions,
105    pub meta_client: Option<MetaClientOptions>,
106    pub logging: LoggingOptions,
107    pub tracing: TracingOptions,
108    pub heartbeat: HeartbeatOptions,
109    pub query: QueryOptions,
110    pub user_provider: Option<String>,
111}
112
113impl Default for FlownodeOptions {
114    fn default() -> Self {
115        Self {
116            node_id: None,
117            flow: FlowConfig::default(),
118            grpc: GrpcOptions::default().with_bind_addr("127.0.0.1:3004"),
119            http: HttpOptions::default(),
120            meta_client: None,
121            logging: LoggingOptions::default(),
122            tracing: TracingOptions::default(),
123            heartbeat: HeartbeatOptions::default(),
124            // flownode's query option is set to 1 to throttle flow's query so
125            // that it won't use too much cpu or memory
126            query: QueryOptions { parallelism: 1 },
127            user_provider: None,
128        }
129    }
130}
131
132impl Configurable for FlownodeOptions {
133    fn validate_sanitize(&mut self) -> common_config::error::Result<()> {
134        if self.flow.num_workers == 0 {
135            self.flow.num_workers = (common_config::utils::get_cpus() / 2).max(1);
136        }
137        Ok(())
138    }
139}
140
141/// Arc-ed FlowNodeManager, cheaper to clone
142pub type FlowStreamingEngineRef = Arc<StreamingEngine>;
143
144/// FlowNodeManager manages the state of all tasks in the flow node, which should be run on the same thread
145///
146/// The choice of timestamp is just using current system timestamp for now
147///
148pub struct StreamingEngine {
149    /// The handler to the worker that will run the dataflow
150    /// which is `!Send` so a handle is used
151    pub worker_handles: Vec<WorkerHandle>,
152    /// The selector to select a worker to run the dataflow
153    worker_selector: Mutex<usize>,
154    /// The query engine that will be used to parse the query and convert it to a dataflow plan
155    pub query_engine: Arc<dyn QueryEngine>,
156    /// Getting table name and table schema from table info manager
157    table_info_source: ManagedTableSource,
158    frontend_invoker: RwLock<Option<FrontendInvoker>>,
159    /// contains mapping from table name to global id, and table schema
160    node_context: RwLock<FlownodeContext>,
161    /// Contains all refill tasks
162    refill_tasks: RwLock<BTreeMap<FlowId, RefillTask>>,
163    flow_err_collectors: RwLock<BTreeMap<FlowId, ErrCollector>>,
164    src_send_buf_lens: RwLock<BTreeMap<TableId, watch::Receiver<usize>>>,
165    tick_manager: FlowTickManager,
166    /// This node id is only available in distributed mode, on standalone mode this is guaranteed to be `None`
167    pub node_id: Option<u32>,
168    /// Lock for flushing, will be `read` by `handle_inserts` and `write` by `flush_flow`
169    ///
170    /// So that a series of event like `inserts -> flush` can be handled correctly
171    flush_lock: RwLock<()>,
172    /// receive a oneshot sender to send state size report
173    state_report_handler: RwLock<Option<StateReportHandler>>,
174}
175
176/// Building FlownodeManager
177impl StreamingEngine {
178    /// set frontend invoker
179    pub async fn set_frontend_invoker(&self, frontend: FrontendInvoker) {
180        *self.frontend_invoker.write().await = Some(frontend);
181    }
182
183    /// Create **without** setting `frontend_invoker`
184    pub fn new(
185        node_id: Option<u32>,
186        query_engine: Arc<dyn QueryEngine>,
187        table_meta: TableMetadataManagerRef,
188    ) -> Self {
189        let srv_map = ManagedTableSource::new(
190            table_meta.table_info_manager().clone(),
191            table_meta.table_name_manager().clone(),
192        );
193        let node_context = FlownodeContext::new(Box::new(srv_map.clone()) as _);
194        let tick_manager = FlowTickManager::new();
195        let worker_handles = Vec::new();
196        StreamingEngine {
197            worker_handles,
198            worker_selector: Mutex::new(0),
199            query_engine,
200            table_info_source: srv_map,
201            frontend_invoker: RwLock::new(None),
202            node_context: RwLock::new(node_context),
203            refill_tasks: Default::default(),
204            flow_err_collectors: Default::default(),
205            src_send_buf_lens: Default::default(),
206            tick_manager,
207            node_id,
208            flush_lock: RwLock::new(()),
209            state_report_handler: RwLock::new(None),
210        }
211    }
212
213    pub async fn with_state_report_handler(self, handler: StateReportHandler) -> Self {
214        *self.state_report_handler.write().await = Some(handler);
215        self
216    }
217
218    /// Create a flownode manager with one worker
219    pub fn new_with_workers<'s>(
220        node_id: Option<u32>,
221        query_engine: Arc<dyn QueryEngine>,
222        table_meta: TableMetadataManagerRef,
223        num_workers: usize,
224    ) -> (Self, Vec<Worker<'s>>) {
225        let mut zelf = Self::new(node_id, query_engine, table_meta);
226
227        let workers: Vec<_> = (0..num_workers)
228            .map(|_| {
229                let (handle, worker) = create_worker();
230                zelf.add_worker_handle(handle);
231                worker
232            })
233            .collect();
234        (zelf, workers)
235    }
236
237    /// add a worker handler to manager, meaning this corresponding worker is under it's manage
238    pub fn add_worker_handle(&mut self, handle: WorkerHandle) {
239        self.worker_handles.push(handle);
240    }
241}
242
243#[derive(Debug)]
244pub enum DiffRequest {
245    Insert(Vec<(Row, repr::Timestamp)>),
246    Delete(Vec<(Row, repr::Timestamp)>),
247}
248
249impl DiffRequest {
250    pub fn len(&self) -> usize {
251        match self {
252            Self::Insert(v) => v.len(),
253            Self::Delete(v) => v.len(),
254        }
255    }
256
257    pub fn is_empty(&self) -> bool {
258        self.len() == 0
259    }
260}
261
262pub fn batches_to_rows_req(batches: Vec<Batch>) -> Result<Vec<DiffRequest>, Error> {
263    let mut reqs = Vec::new();
264    for batch in batches {
265        let mut rows = Vec::with_capacity(batch.row_count());
266        for i in 0..batch.row_count() {
267            let row = batch.get_row(i).context(EvalSnafu)?;
268            rows.push((Row::new(row), 0));
269        }
270        reqs.push(DiffRequest::Insert(rows));
271    }
272    Ok(reqs)
273}
274
275/// This impl block contains methods to send writeback requests to frontend
276impl StreamingEngine {
277    /// Return the number of requests it made
278    pub async fn send_writeback_requests(&self) -> Result<usize, Error> {
279        let all_reqs = self.generate_writeback_request().await?;
280        if all_reqs.is_empty() || all_reqs.iter().all(|v| v.1.is_empty()) {
281            return Ok(0);
282        }
283        let mut req_cnt = 0;
284        for (table_name, reqs) in all_reqs {
285            if reqs.is_empty() {
286                continue;
287            }
288            let (catalog, schema) = (table_name[0].clone(), table_name[1].clone());
289            let ctx = Arc::new(QueryContext::with(&catalog, &schema));
290
291            let (is_ts_placeholder, proto_schema) = match self
292                .try_fetch_existing_table(&table_name)
293                .await?
294                .context(UnexpectedSnafu {
295                    reason: format!("Table not found: {}", table_name.join(".")),
296                }) {
297                Ok(r) => r,
298                Err(e) => {
299                    if self
300                        .table_info_source
301                        .get_opt_table_id_from_name(&table_name)
302                        .await?
303                        .is_none()
304                    {
305                        // deal with both flow&sink table no longer exists
306                        // but some output is still in output buf
307                        common_telemetry::warn!(e; "Table `{}` no longer exists, skip writeback", table_name.join("."));
308                        continue;
309                    } else {
310                        return Err(e);
311                    }
312                }
313            };
314            let schema_len = proto_schema.len();
315
316            let total_rows = reqs.iter().map(|r| r.len()).sum::<usize>();
317            trace!(
318                "Sending {} writeback requests to table {}, reqs total rows={}",
319                reqs.len(),
320                table_name.join("."),
321                reqs.iter().map(|r| r.len()).sum::<usize>()
322            );
323
324            METRIC_FLOW_ROWS
325                .with_label_values(&["out-streaming"])
326                .inc_by(total_rows as u64);
327
328            let now = self.tick_manager.tick();
329            for req in reqs {
330                match req {
331                    DiffRequest::Insert(insert) => {
332                        let rows_proto: Vec<v1::Row> = insert
333                            .into_iter()
334                            .map(|(mut row, _ts)| {
335                                // extend `update_at` col if needed
336                                // if schema include a millisecond timestamp here, and result row doesn't have it, add it
337                                if row.len() < proto_schema.len()
338                                    && proto_schema[row.len()].datatype
339                                        == greptime_proto::v1::ColumnDataType::TimestampMillisecond
340                                            as i32
341                                {
342                                    row.extend([Value::from(
343                                        common_time::Timestamp::new_millisecond(now),
344                                    )]);
345                                }
346                                // ts col, if auto create
347                                if is_ts_placeholder {
348                                    ensure!(
349                                        row.len() == schema_len - 1,
350                                        InternalSnafu {
351                                            reason: format!(
352                                                "Row len mismatch, expect {} got {}",
353                                                schema_len - 1,
354                                                row.len()
355                                            )
356                                        }
357                                    );
358                                    row.extend([Value::from(
359                                        common_time::Timestamp::new_millisecond(0),
360                                    )]);
361                                }
362                                if row.len() != proto_schema.len() {
363                                    UnexpectedSnafu {
364                                        reason: format!(
365                                            "Flow output row length mismatch, expect {} got {}, the columns in schema are: {:?}",
366                                            proto_schema.len(),
367                                            row.len(),
368                                            proto_schema.iter().map(|c|&c.column_name).collect_vec()
369                                        ),
370                                    }
371                                    .fail()?;
372                                }
373                                Ok(row.into())
374                            })
375                            .collect::<Result<Vec<_>, Error>>()?;
376                        let table_name = table_name.last().unwrap().clone();
377                        let req = RowInsertRequest {
378                            table_name,
379                            rows: Some(v1::Rows {
380                                schema: proto_schema.clone(),
381                                rows: rows_proto,
382                            }),
383                        };
384                        req_cnt += 1;
385                        self.frontend_invoker
386                            .read()
387                            .await
388                            .as_ref()
389                            .with_context(|| UnexpectedSnafu {
390                                reason: "Expect a frontend invoker for flownode to write back",
391                            })?
392                            .row_inserts(RowInsertRequests { inserts: vec![req] }, ctx.clone())
393                            .await
394                            .map_err(BoxedError::new)
395                            .with_context(|_| ExternalSnafu {})?;
396                    }
397                    DiffRequest::Delete(remove) => {
398                        info!("original remove rows={:?}", remove);
399                        let rows_proto: Vec<v1::Row> = remove
400                            .into_iter()
401                            .map(|(mut row, _ts)| {
402                                row.extend(Some(Value::from(
403                                    common_time::Timestamp::new_millisecond(0),
404                                )));
405                                row.into()
406                            })
407                            .collect::<Vec<_>>();
408                        let table_name = table_name.last().unwrap().clone();
409                        let req = RowDeleteRequest {
410                            table_name,
411                            rows: Some(v1::Rows {
412                                schema: proto_schema.clone(),
413                                rows: rows_proto,
414                            }),
415                        };
416
417                        req_cnt += 1;
418                        self.frontend_invoker
419                            .read()
420                            .await
421                            .as_ref()
422                            .with_context(|| UnexpectedSnafu {
423                                reason: "Expect a frontend invoker for flownode to write back",
424                            })?
425                            .row_deletes(RowDeleteRequests { deletes: vec![req] }, ctx.clone())
426                            .await
427                            .map_err(BoxedError::new)
428                            .with_context(|_| ExternalSnafu {})?;
429                    }
430                }
431            }
432        }
433        Ok(req_cnt)
434    }
435
436    /// Generate writeback request for all sink table
437    pub async fn generate_writeback_request(
438        &self,
439    ) -> Result<BTreeMap<TableName, Vec<DiffRequest>>, Error> {
440        trace!("Start to generate writeback request");
441        let mut output = BTreeMap::new();
442        let mut total_row_count = 0;
443        for (name, sink_recv) in self
444            .node_context
445            .write()
446            .await
447            .sink_receiver
448            .iter_mut()
449            .map(|(n, (_s, r))| (n, r))
450        {
451            let mut batches = Vec::new();
452            while let Ok(batch) = sink_recv.try_recv() {
453                total_row_count += batch.row_count();
454                batches.push(batch);
455            }
456            let reqs = batches_to_rows_req(batches)?;
457            output.insert(name.clone(), reqs);
458        }
459        trace!("Prepare writeback req: total row count={}", total_row_count);
460        Ok(output)
461    }
462
463    /// Fetch table schema and primary key from table info source, if table not exist return None
464    async fn fetch_table_pk_schema(
465        &self,
466        table_name: &TableName,
467    ) -> Result<Option<(Vec<String>, Option<usize>, Vec<ColumnSchema>)>, Error> {
468        if let Some(table_id) = self
469            .table_info_source
470            .get_opt_table_id_from_name(table_name)
471            .await?
472        {
473            let table_info = self
474                .table_info_source
475                .get_table_info_value(&table_id)
476                .await?
477                .unwrap();
478            let meta = table_info.table_info.meta;
479            let primary_keys = meta
480                .primary_key_indices
481                .into_iter()
482                .map(|i| meta.schema.column_schemas[i].name.clone())
483                .collect_vec();
484            let schema = meta.schema.column_schemas;
485            let time_index = meta.schema.timestamp_index;
486            Ok(Some((primary_keys, time_index, schema)))
487        } else {
488            Ok(None)
489        }
490    }
491
492    /// return (primary keys, schema and if the table have a placeholder timestamp column)
493    /// schema of the table comes from flow's output plan
494    ///
495    /// adjust to add `update_at` column and ts placeholder if needed
496    async fn adjust_auto_created_table_schema(
497        &self,
498        schema: &RelationDesc,
499    ) -> Result<(Vec<String>, Vec<ColumnSchema>, bool), Error> {
500        // TODO(discord9): condiser remove buggy auto create by schema
501
502        // TODO(discord9): use default key from schema
503        let primary_keys = schema
504            .typ()
505            .keys
506            .first()
507            .map(|v| {
508                v.column_indices
509                    .iter()
510                    .map(|i| {
511                        schema
512                            .get_name(*i)
513                            .clone()
514                            .unwrap_or_else(|| format!("col_{i}"))
515                    })
516                    .collect_vec()
517            })
518            .unwrap_or_default();
519        let update_at = ColumnSchema::new(
520            AUTO_CREATED_UPDATE_AT_TS_COL,
521            ConcreteDataType::timestamp_millisecond_datatype(),
522            true,
523        );
524
525        let original_schema = relation_desc_to_column_schemas_with_fallback(schema);
526
527        let mut with_auto_added_col = original_schema.clone();
528        with_auto_added_col.push(update_at);
529
530        // if no time index, add one as placeholder
531        let no_time_index = schema.typ().time_index.is_none();
532        if no_time_index {
533            let ts_col = ColumnSchema::new(
534                AUTO_CREATED_PLACEHOLDER_TS_COL,
535                ConcreteDataType::timestamp_millisecond_datatype(),
536                true,
537            )
538            .with_time_index(true);
539            with_auto_added_col.push(ts_col);
540        }
541
542        Ok((primary_keys, with_auto_added_col, no_time_index))
543    }
544}
545
546/// Flow Runtime related methods
547impl StreamingEngine {
548    /// Start state report handler, which will receive a sender from HeartbeatTask to send state size report back
549    ///
550    /// if heartbeat task is shutdown, this future will exit too
551    async fn start_state_report_handler(self: Arc<Self>) -> Option<JoinHandle<()>> {
552        let state_report_handler = self.state_report_handler.write().await.take();
553        if let Some(mut handler) = state_report_handler {
554            let zelf = self.clone();
555            let handler = common_runtime::spawn_global(async move {
556                while let Some(ret_handler) = handler.recv().await {
557                    let state_report = zelf.gen_state_report().await;
558                    ret_handler.send(state_report).unwrap_or_else(|err| {
559                        common_telemetry::error!(err; "Send state size report error");
560                    });
561                }
562            });
563            Some(handler)
564        } else {
565            None
566        }
567    }
568
569    /// run in common_runtime background runtime
570    pub fn run_background(
571        self: Arc<Self>,
572        shutdown: Option<broadcast::Receiver<()>>,
573    ) -> JoinHandle<()> {
574        info!("Starting flownode manager's background task");
575        common_runtime::spawn_global(async move {
576            let _state_report_handler = self.clone().start_state_report_handler().await;
577            self.run(shutdown).await;
578        })
579    }
580
581    /// log all flow errors
582    pub async fn log_all_errors(&self) {
583        for (f_id, f_err) in self.flow_err_collectors.read().await.iter() {
584            let all_errors = f_err.get_all().await;
585            if !all_errors.is_empty() {
586                let all_errors = all_errors
587                    .into_iter()
588                    .map(|i| format!("{:?}", i))
589                    .join("\n");
590                common_telemetry::error!("Flow {} has following errors: {}", f_id, all_errors);
591            }
592        }
593    }
594
595    /// Trigger dataflow running, and then send writeback request to the source sender
596    ///
597    /// note that this method didn't handle input mirror request, as this should be handled by grpc server
598    pub async fn run(&self, mut shutdown: Option<broadcast::Receiver<()>>) {
599        debug!("Starting to run");
600        let default_interval = Duration::from_secs(1);
601        let mut tick_interval = tokio::time::interval(default_interval);
602        // burst mode, so that if we miss a tick, we will run immediately to fully utilize the cpu
603        tick_interval.set_missed_tick_behavior(tokio::time::MissedTickBehavior::Burst);
604        let mut avg_spd = 0; // rows/sec
605        let mut since_last_run = tokio::time::Instant::now();
606        let run_per_trace = 10;
607        let mut run_cnt = 0;
608        loop {
609            // TODO(discord9): only run when new inputs arrive or scheduled to
610            let row_cnt = self.run_available(false).await.unwrap_or_else(|err| {
611                common_telemetry::error!(err;"Run available errors");
612                0
613            });
614
615            if let Err(err) = self.send_writeback_requests().await {
616                common_telemetry::error!(err;"Send writeback request errors");
617            };
618            self.log_all_errors().await;
619
620            // determine if need to shutdown
621            match &shutdown.as_mut().map(|s| s.try_recv()) {
622                Some(Ok(())) => {
623                    info!("Shutdown flow's main loop");
624                    break;
625                }
626                Some(Err(TryRecvError::Empty)) => (),
627                Some(Err(TryRecvError::Closed)) => {
628                    common_telemetry::error!("Shutdown channel is closed");
629                    break;
630                }
631                Some(Err(TryRecvError::Lagged(num))) => {
632                    common_telemetry::error!("Shutdown channel is lagged by {}, meaning multiple shutdown cmd have been issued", num);
633                    break;
634                }
635                None => (),
636            }
637
638            // for now we want to batch rows until there is around `BATCH_SIZE` rows in send buf
639            // before trigger a run of flow's worker
640            let wait_for = since_last_run.elapsed();
641
642            // last runs insert speed
643            let cur_spd = row_cnt * 1000 / wait_for.as_millis().max(1) as usize;
644            // rapid increase, slow decay
645            avg_spd = if cur_spd > avg_spd {
646                cur_spd
647            } else {
648                (9 * avg_spd + cur_spd) / 10
649            };
650            let new_wait = BATCH_SIZE * 1000 / avg_spd.max(1); //in ms
651            let new_wait = Duration::from_millis(new_wait as u64).min(default_interval);
652
653            // print trace every `run_per_trace` times so that we can see if there is something wrong
654            // but also not get flooded with trace
655            if run_cnt >= run_per_trace {
656                trace!("avg_spd={} r/s, cur_spd={} r/s", avg_spd, cur_spd);
657                trace!("Wait for {} ms, row_cnt={}", new_wait.as_millis(), row_cnt);
658                run_cnt = 0;
659            } else {
660                run_cnt += 1;
661            }
662
663            METRIC_FLOW_RUN_INTERVAL_MS.set(new_wait.as_millis() as i64);
664            since_last_run = tokio::time::Instant::now();
665            tokio::select! {
666                _ = tick_interval.tick() => (),
667                _ = tokio::time::sleep(new_wait) => ()
668            }
669        }
670        // flow is now shutdown, drop frontend_invoker early so a ref cycle(in standalone mode) can be prevent:
671        // FlowWorkerManager.frontend_invoker -> FrontendInvoker.inserter
672        // -> Inserter.node_manager -> NodeManager.flownode -> Flownode.flow_streaming_engine.frontend_invoker
673        self.frontend_invoker.write().await.take();
674    }
675
676    /// Run all available subgraph in the flow node
677    /// This will try to run all dataflow in this node
678    ///
679    /// set `blocking` to true to wait until worker finish running
680    /// false to just trigger run and return immediately
681    /// return numbers of rows send to worker(Inaccuary)
682    /// TODO(discord9): add flag for subgraph that have input since last run
683    pub async fn run_available(&self, blocking: bool) -> Result<usize, Error> {
684        let mut row_cnt = 0;
685
686        let now = self.tick_manager.tick();
687        for worker in self.worker_handles.iter() {
688            // TODO(discord9): consider how to handle error in individual worker
689            worker.run_available(now, blocking).await?;
690        }
691        // check row send and rows remain in send buf
692        let flush_res = if blocking {
693            let ctx = self.node_context.read().await;
694            ctx.flush_all_sender().await
695        } else {
696            match self.node_context.try_read() {
697                Ok(ctx) => ctx.flush_all_sender().await,
698                Err(_) => return Ok(row_cnt),
699            }
700        };
701        match flush_res {
702            Ok(r) => {
703                common_telemetry::trace!("Total flushed {} rows", r);
704                row_cnt += r;
705            }
706            Err(err) => {
707                common_telemetry::error!("Flush send buf errors: {:?}", err);
708            }
709        };
710
711        Ok(row_cnt)
712    }
713
714    /// send write request to related source sender
715    pub async fn handle_write_request(
716        &self,
717        region_id: RegionId,
718        rows: Vec<DiffRow>,
719        batch_datatypes: &[ConcreteDataType],
720    ) -> Result<(), Error> {
721        let rows_len = rows.len();
722        let table_id = region_id.table_id();
723        let _timer = METRIC_FLOW_INSERT_ELAPSED
724            .with_label_values(&[table_id.to_string().as_str()])
725            .start_timer();
726        self.node_context
727            .read()
728            .await
729            .send(table_id, rows, batch_datatypes)
730            .await?;
731        trace!(
732            "Handling write request for table_id={} with {} rows",
733            table_id,
734            rows_len
735        );
736        Ok(())
737    }
738}
739
740/// Create&Remove flow
741impl StreamingEngine {
742    /// remove a flow by it's id
743    pub async fn remove_flow_inner(&self, flow_id: FlowId) -> Result<(), Error> {
744        for handle in self.worker_handles.iter() {
745            if handle.contains_flow(flow_id).await? {
746                handle.remove_flow(flow_id).await?;
747                break;
748            }
749        }
750        self.node_context.write().await.remove_flow(flow_id);
751        Ok(())
752    }
753
754    /// Return task id if a new task is created, otherwise return None
755    ///
756    /// steps to create task:
757    /// 1. parse query into typed plan(and optional parse expire_after expr)
758    /// 2. render source/sink with output table id and used input table id
759    pub async fn create_flow_inner(&self, args: CreateFlowArgs) -> Result<Option<FlowId>, Error> {
760        let CreateFlowArgs {
761            flow_id,
762            sink_table_name,
763            source_table_ids,
764            create_if_not_exists,
765            or_replace,
766            expire_after,
767            comment,
768            sql,
769            flow_options,
770            query_ctx,
771        } = args;
772
773        let mut node_ctx = self.node_context.write().await;
774        // assign global id to source and sink table
775        for source in &source_table_ids {
776            node_ctx
777                .assign_global_id_to_table(&self.table_info_source, None, Some(*source))
778                .await?;
779        }
780        node_ctx
781            .assign_global_id_to_table(&self.table_info_source, Some(sink_table_name.clone()), None)
782            .await?;
783
784        node_ctx.register_task_src_sink(flow_id, &source_table_ids, sink_table_name.clone());
785
786        node_ctx.query_context = query_ctx.map(Arc::new);
787        // construct a active dataflow state with it
788        let flow_plan = sql_to_flow_plan(&mut node_ctx, &self.query_engine, &sql).await?;
789
790        debug!("Flow {:?}'s Plan is {:?}", flow_id, flow_plan);
791
792        // check schema against actual table schema if exists
793        // if not exist create sink table immediately
794        if let Some((_, _, real_schema)) = self.fetch_table_pk_schema(&sink_table_name).await? {
795            let auto_schema = relation_desc_to_column_schemas_with_fallback(&flow_plan.schema);
796
797            // for column schema, only `data_type` need to be check for equality
798            // since one can omit flow's column name when write flow query
799            // print a user friendly error message about mismatch and how to correct them
800            for (idx, zipped) in auto_schema
801                .iter()
802                .zip_longest(real_schema.iter())
803                .enumerate()
804            {
805                match zipped {
806                    EitherOrBoth::Both(auto, real) => {
807                        if auto.data_type != real.data_type {
808                            InvalidQuerySnafu {
809                                    reason: format!(
810                                        "Column {}(name is '{}', flow inferred name is '{}')'s data type mismatch, expect {:?} got {:?}",
811                                        idx,
812                                        real.name,
813                                        auto.name,
814                                        real.data_type,
815                                        auto.data_type
816                                    ),
817                                }
818                                .fail()?;
819                        }
820                    }
821                    EitherOrBoth::Right(real) if real.data_type.is_timestamp() => {
822                        // if table is auto created, the last one or two column should be timestamp(update at and ts placeholder)
823                        continue;
824                    }
825                    _ => InvalidQuerySnafu {
826                        reason: format!(
827                            "schema length mismatched, expected {} found {}",
828                            real_schema.len(),
829                            auto_schema.len()
830                        ),
831                    }
832                    .fail()?,
833                }
834            }
835        } else {
836            // assign inferred schema to sink table
837            // create sink table
838            let did_create = self
839                .create_table_from_relation(
840                    &format!("flow-id={flow_id}"),
841                    &sink_table_name,
842                    &flow_plan.schema,
843                )
844                .await?;
845            if !did_create {
846                UnexpectedSnafu {
847                    reason: format!("Failed to create table {:?}", sink_table_name),
848                }
849                .fail()?;
850            }
851        }
852
853        node_ctx.add_flow_plan(flow_id, flow_plan.clone());
854
855        let _ = comment;
856        let _ = flow_options;
857
858        // TODO(discord9): add more than one handles
859        let sink_id = node_ctx.table_repr.get_by_name(&sink_table_name).unwrap().1;
860        let sink_sender = node_ctx.get_sink_by_global_id(&sink_id)?;
861
862        let source_ids = source_table_ids
863            .iter()
864            .map(|id| node_ctx.table_repr.get_by_table_id(id).unwrap().1)
865            .collect_vec();
866        let source_receivers = source_ids
867            .iter()
868            .map(|id| {
869                node_ctx
870                    .get_source_by_global_id(id)
871                    .map(|s| s.get_receiver())
872            })
873            .collect::<Result<Vec<_>, _>>()?;
874        let err_collector = ErrCollector::default();
875        self.flow_err_collectors
876            .write()
877            .await
878            .insert(flow_id, err_collector.clone());
879        // TODO(discord9): load balance?
880        let handle = self.get_worker_handle_for_create_flow().await;
881        let create_request = worker::Request::Create {
882            flow_id,
883            plan: flow_plan,
884            sink_id,
885            sink_sender,
886            source_ids,
887            src_recvs: source_receivers,
888            expire_after,
889            or_replace,
890            create_if_not_exists,
891            err_collector,
892        };
893
894        handle.create_flow(create_request).await?;
895        info!("Successfully create flow with id={}", flow_id);
896        Ok(Some(flow_id))
897    }
898
899    pub async fn flush_flow_inner(&self, flow_id: FlowId) -> Result<usize, Error> {
900        debug!("Starting to flush flow_id={:?}", flow_id);
901        // lock to make sure writes before flush are written to flow
902        // and immediately drop to prevent following writes to be blocked
903        drop(self.flush_lock.write().await);
904        let flushed_input_rows = self.node_context.read().await.flush_all_sender().await?;
905        let rows_send = self.run_available(true).await?;
906        let row = self.send_writeback_requests().await?;
907        debug!(
908            "Done to flush flow_id={:?} with {} input rows flushed, {} rows sent and {} output rows flushed",
909            flow_id, flushed_input_rows, rows_send, row
910        );
911        Ok(row)
912    }
913
914    pub async fn flow_exist_inner(&self, flow_id: FlowId) -> Result<bool, Error> {
915        let mut exist = false;
916        for handle in self.worker_handles.iter() {
917            if handle.contains_flow(flow_id).await? {
918                exist = true;
919                break;
920            }
921        }
922        Ok(exist)
923    }
924}
925
926/// FlowTickManager is a manager for flow tick, which trakc flow execution progress
927///
928/// TODO(discord9): better way to do it, and not expose flow tick even to other flow to avoid
929/// TSO coord mess
930#[derive(Clone, Debug)]
931pub struct FlowTickManager {
932    /// The starting instant of the flow, used with `start_timestamp` to calculate the current timestamp
933    start: Instant,
934    /// The timestamp when the flow started
935    start_timestamp: repr::Timestamp,
936}
937
938impl Default for FlowTickManager {
939    fn default() -> Self {
940        Self::new()
941    }
942}
943
944impl FlowTickManager {
945    pub fn new() -> Self {
946        FlowTickManager {
947            start: Instant::now(),
948            start_timestamp: SystemTime::now()
949                .duration_since(SystemTime::UNIX_EPOCH)
950                .unwrap()
951                .as_millis() as repr::Timestamp,
952        }
953    }
954
955    /// Return the current timestamp in milliseconds
956    ///
957    /// TODO(discord9): reconsider since `tick()` require a monotonic clock and also need to survive recover later
958    pub fn tick(&self) -> repr::Timestamp {
959        let current = Instant::now();
960        let since_the_epoch = current - self.start;
961        since_the_epoch.as_millis() as repr::Timestamp + self.start_timestamp
962    }
963}
flow/adapter.rs

flow/
adapter.rs
