risingwave_meta/barrier/
complete_task.rs

1// Copyright 2025 RisingWave Labs
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
15use std::collections::HashMap;
16use std::future::{Future, pending};
17use std::mem::replace;
18use std::sync::Arc;
19
20use anyhow::Context;
21use futures::future::try_join_all;
22use prometheus::HistogramTimer;
23use risingwave_common::catalog::DatabaseId;
24use risingwave_common::id::JobId;
25use risingwave_common::must_match;
26use risingwave_common::util::deployment::Deployment;
27use risingwave_pb::hummock::HummockVersionStats;
28use tokio::task::JoinHandle;
29
30use crate::barrier::checkpoint::CheckpointControl;
31use crate::barrier::command::CommandContext;
32use crate::barrier::context::GlobalBarrierWorkerContext;
33use crate::barrier::notifier::Notifier;
34use crate::barrier::progress::TrackingJob;
35use crate::barrier::rpc::ControlStreamManager;
36use crate::barrier::schedule::PeriodicBarriers;
37use crate::hummock::CommitEpochInfo;
38use crate::manager::MetaSrvEnv;
39use crate::rpc::metrics::GLOBAL_META_METRICS;
40use crate::{MetaError, MetaResult};
41
42pub(super) enum CompletingTask {
43    None,
44    Completing {
45        #[expect(clippy::type_complexity)]
46        /// `database_id` -> (`Some(database_graph_committed_epoch)`, [(`creating_job_id`, `creating_job_committed_epoch`)])
47        epochs_to_ack: HashMap<DatabaseId, (Option<u64>, Vec<(JobId, u64)>)>,
48
49        // The join handle of a spawned task that completes the barrier.
50        // The return value indicate whether there is some create streaming job command
51        // that has finished but not checkpointed. If there is any, we will force checkpoint on the next barrier
52        join_handle: JoinHandle<MetaResult<HummockVersionStats>>,
53    },
54    #[expect(dead_code)]
55    Err(MetaError),
56}
57
58/// Only for checkpoint barrier. For normal barrier, there won't be a task.
59#[derive(Default)]
60pub(super) struct CompleteBarrierTask {
61    pub(super) commit_info: CommitEpochInfo,
62    pub(super) finished_jobs: Vec<TrackingJob>,
63    pub(super) finished_cdc_table_backfill: Vec<JobId>,
64    pub(super) notifiers: Vec<Notifier>,
65    /// `database_id` -> (Some((`command_ctx`, `enqueue_time`)), vec!((`creating_job_id`, `epoch`)))
66    #[expect(clippy::type_complexity)]
67    pub(super) epoch_infos:
68        HashMap<DatabaseId, (Option<(CommandContext, HistogramTimer)>, Vec<(JobId, u64)>)>,
69    /// Source IDs that have finished listing data and need `ListFinish` commands
70    pub(super) list_finished_source_ids: Vec<u32>,
71    /// Source IDs that have finished loading data and need `LoadFinish` commands
72    pub(super) load_finished_source_ids: Vec<u32>,
73    /// Table IDs that have finished materialize refresh and need completion signaling
74    pub(super) refresh_finished_table_job_ids: Vec<JobId>,
75}
76
77impl CompleteBarrierTask {
78    #[expect(clippy::type_complexity)]
79    pub(super) fn epochs_to_ack(&self) -> HashMap<DatabaseId, (Option<u64>, Vec<(JobId, u64)>)> {
80        self.epoch_infos
81            .iter()
82            .map(|(database_id, (command_context, creating_job_epochs))| {
83                (
84                    *database_id,
85                    (
86                        command_context
87                            .as_ref()
88                            .map(|(command, _)| command.barrier_info.prev_epoch.value().0),
89                        creating_job_epochs.clone(),
90                    ),
91                )
92            })
93            .collect()
94    }
95}
96
97impl CompleteBarrierTask {
98    pub(super) async fn complete_barrier(
99        self,
100        context: &impl GlobalBarrierWorkerContext,
101        env: MetaSrvEnv,
102    ) -> MetaResult<HummockVersionStats> {
103        let result: MetaResult<HummockVersionStats> = try {
104            let wait_commit_timer = GLOBAL_META_METRICS
105                .barrier_wait_commit_latency
106                .start_timer();
107            let version_stats = context.commit_epoch(self.commit_info).await?;
108
109            // Handle list finished source IDs for refreshable batch sources
110            // Spawn this asynchronously to avoid deadlock during barrier collection
111            //
112            // This step is for fs-like refreshable-batch sources, which need to list the data first finishing loading. It guarantees finishing listing before loading.
113            // The other sources can skip this step.
114
115            if !self.list_finished_source_ids.is_empty() {
116                context
117                    .handle_list_finished_source_ids(self.list_finished_source_ids.clone())
118                    .await?;
119            }
120
121            // Handle load finished source IDs for refreshable batch sources
122            // Spawn this asynchronously to avoid deadlock during barrier collection
123            if !self.load_finished_source_ids.is_empty() {
124                context
125                    .handle_load_finished_source_ids(self.load_finished_source_ids.clone())
126                    .await?;
127            }
128
129            // Handle refresh finished table IDs for materialized view refresh completion
130            if !self.refresh_finished_table_job_ids.is_empty() {
131                context
132                    .handle_refresh_finished_table_ids(self.refresh_finished_table_job_ids.clone())
133                    .await?;
134            }
135
136            for command_ctx in self
137                .epoch_infos
138                .values()
139                .flat_map(|(command, _)| command.as_ref().map(|(command, _)| command))
140            {
141                context.post_collect_command(command_ctx).await?;
142            }
143
144            wait_commit_timer.observe_duration();
145            version_stats
146        };
147
148        let version_stats = {
149            let version_stats = match result {
150                Ok(version_stats) => version_stats,
151                Err(e) => {
152                    for notifier in self.notifiers {
153                        notifier.notify_collection_failed(e.clone());
154                    }
155                    return Err(e);
156                }
157            };
158            self.notifiers.into_iter().for_each(|notifier| {
159                notifier.notify_collected();
160            });
161            try_join_all(
162                self.finished_jobs
163                    .into_iter()
164                    .map(|finished_job| context.finish_creating_job(finished_job)),
165            )
166            .await?;
167            try_join_all(
168                self.finished_cdc_table_backfill
169                    .into_iter()
170                    .map(|job_id| context.finish_cdc_table_backfill(job_id)),
171            )
172            .await?;
173            for (database_id, (command, _)) in self.epoch_infos {
174                if let Some((command_ctx, enqueue_time)) = command {
175                    let duration_sec = enqueue_time.stop_and_record();
176                    Self::report_complete_event(&env, duration_sec, &command_ctx);
177                    GLOBAL_META_METRICS
178                        .last_committed_barrier_time
179                        .with_label_values(&[database_id.to_string().as_str()])
180                        .set(command_ctx.barrier_info.curr_epoch.value().as_unix_secs() as i64);
181                }
182            }
183            version_stats
184        };
185
186        Ok(version_stats)
187    }
188}
189
190impl CompleteBarrierTask {
191    fn report_complete_event(env: &MetaSrvEnv, duration_sec: f64, command_ctx: &CommandContext) {
192        // Record barrier latency in event log.
193        use risingwave_pb::meta::event_log;
194        let event = event_log::EventBarrierComplete {
195            prev_epoch: command_ctx.barrier_info.prev_epoch(),
196            cur_epoch: command_ctx.barrier_info.curr_epoch.value().0,
197            duration_sec,
198            command: command_ctx
199                .command
200                .as_ref()
201                .map(|command| command.to_string())
202                .unwrap_or_else(|| "barrier".to_owned()),
203            barrier_kind: command_ctx.barrier_info.kind.as_str_name().to_owned(),
204        };
205        if cfg!(debug_assertions) || Deployment::current().is_ci() {
206            // Add a warning log so that debug mode / CI can observe it
207            if duration_sec > 5.0 {
208                tracing::warn!(event = ?event,"high barrier latency observed!")
209            }
210        }
211        env.event_log_manager_ref()
212            .add_event_logs(vec![event_log::Event::BarrierComplete(event)]);
213    }
214}
215
216pub(super) struct BarrierCompleteOutput {
217    #[expect(clippy::type_complexity)]
218    /// `database_id` -> (`Some(database_graph_committed_epoch)`, [(`creating_job_id`, `creating_job_committed_epoch`)])
219    pub epochs_to_ack: HashMap<DatabaseId, (Option<u64>, Vec<(JobId, u64)>)>,
220    pub hummock_version_stats: HummockVersionStats,
221}
222
223impl CompletingTask {
224    pub(super) fn next_completed_barrier<'a>(
225        &'a mut self,
226        periodic_barriers: &mut PeriodicBarriers,
227        checkpoint_control: &mut CheckpointControl,
228        control_stream_manager: &mut ControlStreamManager,
229        context: &Arc<impl GlobalBarrierWorkerContext>,
230        env: &MetaSrvEnv,
231    ) -> impl Future<Output = MetaResult<BarrierCompleteOutput>> + 'a {
232        // If there is no completing barrier, try to start completing the earliest barrier if
233        // it has been collected.
234        if let CompletingTask::None = self
235            && let Some(task) = checkpoint_control
236                .next_complete_barrier_task(Some((periodic_barriers, control_stream_manager)))
237        {
238            {
239                let epochs_to_ack = task.epochs_to_ack();
240                let context = context.clone();
241                let await_tree_reg = env.await_tree_reg().clone();
242                let env = env.clone();
243
244                let fut = async move { task.complete_barrier(&*context, env).await };
245                let fut = await_tree_reg
246                    .register_derived_root("Barrier Completion Task")
247                    .instrument(fut);
248                let join_handle = tokio::spawn(fut);
249
250                *self = CompletingTask::Completing {
251                    epochs_to_ack,
252                    join_handle,
253                };
254            }
255        }
256
257        async move {
258            if !matches!(self, CompletingTask::Completing { .. }) {
259                return pending().await;
260            };
261            self.next_completed_barrier_inner().await
262        }
263    }
264
265    #[await_tree::instrument]
266    pub(super) async fn wait_completing_task(
267        &mut self,
268    ) -> MetaResult<Option<BarrierCompleteOutput>> {
269        match self {
270            CompletingTask::None => Ok(None),
271            CompletingTask::Completing { .. } => {
272                self.next_completed_barrier_inner().await.map(Some)
273            }
274            CompletingTask::Err(_) => {
275                unreachable!("should not be called on previous err")
276            }
277        }
278    }
279
280    async fn next_completed_barrier_inner(&mut self) -> MetaResult<BarrierCompleteOutput> {
281        let CompletingTask::Completing { join_handle, .. } = self else {
282            unreachable!()
283        };
284
285        {
286            {
287                let join_result: MetaResult<_> = try {
288                    join_handle
289                        .await
290                        .context("failed to join completing command")??
291                };
292                // It's important to reset the completing_command after await no matter the result is err
293                // or not, and otherwise the join handle will be polled again after ready.
294                let next_completing_command_status = if let Err(e) = &join_result {
295                    CompletingTask::Err(e.clone())
296                } else {
297                    CompletingTask::None
298                };
299                let completed_command = replace(self, next_completing_command_status);
300                let hummock_version_stats = join_result?;
301
302                must_match!(completed_command, CompletingTask::Completing {
303                    epochs_to_ack,
304                    ..
305                } => {
306                    Ok(BarrierCompleteOutput {
307                        epochs_to_ack,
308                        hummock_version_stats,
309                    })
310                })
311            }
312        }
313    }
314}
risingwave_meta/barrier/complete_task.rs

risingwave_meta/barrier/
complete_task.rs
