Skip to main content

datanode/
heartbeat.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
15use std::collections::HashMap;
16use std::sync::Arc;
17use std::sync::atomic::{AtomicBool, Ordering};
18use std::time::Duration;
19
20use api::v1::meta::heartbeat_request::NodeWorkloads;
21use api::v1::meta::{DatanodeWorkloads, HeartbeatRequest, NodeInfo, Peer, RegionRole, RegionStat};
22use common_base::Plugins;
23use common_meta::cache_invalidator::CacheInvalidatorRef;
24use common_meta::datanode::{EnvVars, REGION_STATISTIC_KEY};
25use common_meta::distributed_time_constants::BASE_HEARTBEAT_INTERVAL;
26use common_meta::heartbeat::handler::invalidate_table_cache::InvalidateCacheHandler;
27use common_meta::heartbeat::handler::parse_mailbox_message::ParseMailboxMessageHandler;
28use common_meta::heartbeat::handler::suspend::SuspendHandler;
29use common_meta::heartbeat::handler::{
30    HandlerGroupExecutor, HeartbeatResponseHandlerContext, HeartbeatResponseHandlerExecutorRef,
31};
32use common_meta::heartbeat::mailbox::{HeartbeatMailbox, MailboxRef};
33use common_meta::heartbeat::utils::outgoing_message_to_mailbox_message;
34use common_meta::kv_backend::KvBackendRef;
35use common_stat::ResourceStatRef;
36use common_telemetry::{debug, error, info, trace, warn};
37use common_workload::DatanodeWorkloadType;
38use meta_client::MetaClientRef;
39use meta_client::client::heartbeat::HeartbeatConfig;
40use meta_client::client::{HeartbeatSender, MetaClient};
41use servers::addrs;
42use snafu::{OptionExt as _, ResultExt};
43use tokio::sync::{Notify, mpsc};
44use tokio::time::Instant;
45
46use self::handler::RegionHeartbeatResponseHandler;
47use crate::alive_keeper::{CountdownTaskHandlerExtRef, RegionAliveKeeper};
48use crate::config::DatanodeOptions;
49use crate::error::{self, MetaClientInitSnafu, RegionEngineNotFoundSnafu, Result};
50use crate::event_listener::RegionServerEventReceiver;
51use crate::metrics::{self, HEARTBEAT_RECV_COUNT, HEARTBEAT_SENT_COUNT};
52use crate::region_server::RegionServer;
53
54pub(crate) mod handler;
55pub(crate) mod task_tracker;
56
57/// The datanode heartbeat task which sending `[HeartbeatRequest]` to Metasrv periodically in background.
58pub struct HeartbeatTask {
59    node_id: u64,
60    workload_types: Vec<DatanodeWorkloadType>,
61    node_epoch: u64,
62    peer_addr: String,
63    running: Arc<AtomicBool>,
64    meta_client: MetaClientRef,
65    region_server: RegionServer,
66    resp_handler_executor: HeartbeatResponseHandlerExecutorRef,
67    region_alive_keeper: Arc<RegionAliveKeeper>,
68    resource_stat: ResourceStatRef,
69    env_vars: EnvVars,
70}
71
72impl Drop for HeartbeatTask {
73    fn drop(&mut self) {
74        self.running.store(false, Ordering::Release);
75    }
76}
77
78impl HeartbeatTask {
79    /// Create a new heartbeat task instance.
80    pub async fn try_new(
81        opts: &DatanodeOptions,
82        region_server: RegionServer,
83        meta_client: MetaClientRef,
84        kv_backend: KvBackendRef,
85        cache_invalidator: CacheInvalidatorRef,
86        plugins: Plugins,
87        resource_stat: ResourceStatRef,
88    ) -> Result<Self> {
89        let countdown_task_handler_ext = plugins.get::<CountdownTaskHandlerExtRef>();
90        let region_alive_keeper = Arc::new(RegionAliveKeeper::new(
91            region_server.clone(),
92            countdown_task_handler_ext,
93            BASE_HEARTBEAT_INTERVAL,
94        ));
95        let resp_handler_executor = Arc::new(HandlerGroupExecutor::new(vec![
96            region_alive_keeper.clone(),
97            Arc::new(ParseMailboxMessageHandler),
98            Arc::new(SuspendHandler::new(region_server.suspend_state())),
99            Arc::new(
100                RegionHeartbeatResponseHandler::new(region_server.clone(), kv_backend)
101                    .with_open_region_parallelism(opts.init_regions_parallelism),
102            ),
103            Arc::new(InvalidateCacheHandler::new(cache_invalidator)),
104        ]));
105
106        Ok(Self {
107            node_id: opts.node_id.unwrap_or(0),
108            workload_types: opts.workload_types.clone(),
109            // We use datanode's start time millis as the node's epoch.
110            node_epoch: common_time::util::current_time_millis() as u64,
111            peer_addr: addrs::resolve_addr(&opts.grpc.bind_addr, Some(&opts.grpc.server_addr)),
112            running: Arc::new(AtomicBool::new(false)),
113            meta_client,
114            region_server,
115            resp_handler_executor,
116            region_alive_keeper,
117            resource_stat,
118            env_vars: EnvVars::from_config(&opts.heartbeat_env_vars),
119        })
120    }
121
122    pub async fn create_streams(
123        meta_client: &MetaClient,
124        running: Arc<AtomicBool>,
125        handler_executor: HeartbeatResponseHandlerExecutorRef,
126        mailbox: MailboxRef,
127        mut notify: Option<Arc<Notify>>,
128        quit_signal: Arc<Notify>,
129    ) -> Result<(HeartbeatSender, HeartbeatConfig)> {
130        let client_id = meta_client.id();
131        let (tx, mut rx, config) = meta_client.heartbeat().await.context(MetaClientInitSnafu)?;
132
133        let mut last_received_lease = Instant::now();
134
135        let _handle = common_runtime::spawn_hb(async move {
136            while let Some(res) = rx.message().await.unwrap_or_else(|e| {
137                error!(e; "Error while reading heartbeat response");
138                None
139            }) {
140                if let Some(msg) = res.mailbox_message.as_ref() {
141                    info!("Received mailbox message: {msg:?}, meta_client id: {client_id:?}");
142                }
143                if let Some(lease) = res.region_lease.as_ref() {
144                    metrics::LAST_RECEIVED_HEARTBEAT_ELAPSED
145                        .set(last_received_lease.elapsed().as_millis() as i64);
146                    // Resets the timer.
147                    last_received_lease = Instant::now();
148
149                    let mut leader_region_lease_count = 0;
150                    let mut follower_region_lease_count = 0;
151                    for lease in &lease.regions {
152                        match lease.role() {
153                            RegionRole::Leader
154                            | RegionRole::StagingLeader
155                            | RegionRole::DowngradingLeader => leader_region_lease_count += 1,
156                            RegionRole::Follower => follower_region_lease_count += 1,
157                        }
158                    }
159
160                    metrics::HEARTBEAT_REGION_LEASES
161                        .with_label_values(&["leader"])
162                        .set(leader_region_lease_count);
163                    metrics::HEARTBEAT_REGION_LEASES
164                        .with_label_values(&["follower"])
165                        .set(follower_region_lease_count);
166                }
167                let ctx = HeartbeatResponseHandlerContext::new(mailbox.clone(), res);
168                if let Err(e) = Self::handle_response(ctx, handler_executor.clone()).await {
169                    error!(e; "Error while handling heartbeat response");
170                }
171                if let Some(notify) = notify.take() {
172                    notify.notify_one();
173                }
174                if !running.load(Ordering::Acquire) {
175                    info!("Heartbeat task shutdown");
176                }
177            }
178            quit_signal.notify_one();
179            info!("Heartbeat handling loop exit.");
180        });
181        Ok((tx, config))
182    }
183
184    async fn handle_response(
185        ctx: HeartbeatResponseHandlerContext,
186        handler_executor: HeartbeatResponseHandlerExecutorRef,
187    ) -> Result<()> {
188        trace!("Heartbeat response: {:?}", ctx.response);
189        handler_executor
190            .handle(ctx)
191            .await
192            .context(error::HandleHeartbeatResponseSnafu)
193    }
194
195    #[allow(deprecated)]
196    /// Start heartbeat task, spawn background task.
197    pub async fn start(
198        &self,
199        event_receiver: RegionServerEventReceiver,
200        notify: Option<Arc<Notify>>,
201    ) -> Result<()> {
202        let running = self.running.clone();
203        if running
204            .compare_exchange(false, true, Ordering::AcqRel, Ordering::Acquire)
205            .is_err()
206        {
207            warn!("Heartbeat task started multiple times");
208            return Ok(());
209        }
210        let node_id = self.node_id;
211        let node_epoch = self.node_epoch;
212        let addr = &self.peer_addr;
213
214        let meta_client = self.meta_client.clone();
215        let region_server_clone = self.region_server.clone();
216
217        let handler_executor = self.resp_handler_executor.clone();
218
219        let (outgoing_tx, mut outgoing_rx) = mpsc::channel(16);
220        let mailbox = Arc::new(HeartbeatMailbox::new(outgoing_tx));
221
222        let quit_signal = Arc::new(Notify::new());
223
224        let (mut tx, config) = Self::create_streams(
225            &meta_client,
226            running.clone(),
227            handler_executor.clone(),
228            mailbox.clone(),
229            notify,
230            quit_signal.clone(),
231        )
232        .await?;
233
234        let interval = config.interval.as_millis() as u64;
235        let mut retry_interval = config.retry_interval;
236
237        // Update RegionAliveKeeper with the interval from Metasrv
238        self.region_alive_keeper.update_heartbeat_interval(interval);
239
240        info!(
241            "Starting heartbeat to Metasrv with config: {}. My node id is {}, address is {}.",
242            config, node_id, addr
243        );
244
245        let self_peer = Some(Peer {
246            id: node_id,
247            addr: addr.clone(),
248        });
249        let epoch = self.region_alive_keeper.epoch();
250        let workload_types = self.workload_types.clone();
251
252        self.region_alive_keeper.start(Some(event_receiver)).await?;
253        let mut last_sent = Instant::now();
254        let total_cpu_millicores = self.resource_stat.get_total_cpu_millicores();
255        let total_memory_bytes = self.resource_stat.get_total_memory_bytes();
256        let resource_stat = self.resource_stat.clone();
257        let region_alive_keeper = self.region_alive_keeper.clone();
258        let gc_limiter = self
259            .region_server
260            .mito_engine()
261            .context(RegionEngineNotFoundSnafu { name: "mito" })?
262            .gc_limiter();
263        let mut env_var_extensions = HashMap::new();
264        self.env_vars.into_extensions(&mut env_var_extensions);
265
266        common_runtime::spawn_hb(async move {
267            let sleep = tokio::time::sleep(Duration::from_millis(0));
268            tokio::pin!(sleep);
269
270            let build_info = common_version::build_info();
271
272            let heartbeat_request = HeartbeatRequest {
273                peer: self_peer,
274                node_epoch,
275                info: Some(NodeInfo {
276                    version: build_info.version.to_string(),
277                    git_commit: build_info.commit_short.to_string(),
278                    start_time_ms: node_epoch,
279                    total_cpu_millicores,
280                    total_memory_bytes,
281                    cpu_usage_millicores: 0,
282                    memory_usage_bytes: 0,
283                    // TODO(zyy17): Remove these deprecated fields when the deprecated fields are removed from the proto.
284                    cpus: total_cpu_millicores as u32,
285                    memory_bytes: total_memory_bytes as u64,
286                    hostname: hostname::get()
287                        .unwrap_or_default()
288                        .to_string_lossy()
289                        .to_string(),
290                }),
291                node_workloads: Some(NodeWorkloads::Datanode(DatanodeWorkloads {
292                    types: workload_types.iter().map(|w| w.to_i32()).collect(),
293                })),
294                ..Default::default()
295            };
296
297            loop {
298                if !running.load(Ordering::Relaxed) {
299                    info!("shutdown heartbeat task");
300                    break;
301                }
302                let req = tokio::select! {
303                    message = outgoing_rx.recv() => {
304                        if let Some(message) = message {
305                            match outgoing_message_to_mailbox_message(message) {
306                                Ok(message) => {
307                                    let mut extensions = env_var_extensions.clone();
308                                    let gc_stat = gc_limiter.gc_stat();
309                                    gc_stat.into_extensions(&mut extensions);
310
311                                    let req = HeartbeatRequest {
312                                        mailbox_message: Some(message),
313                                        extensions,
314                                        ..heartbeat_request.clone()
315                                    };
316                                    HEARTBEAT_RECV_COUNT.with_label_values(&["success"]).inc();
317                                    Some(req)
318                                }
319                                Err(e) => {
320                                    error!(e; "Failed to encode mailbox messages!");
321                                    HEARTBEAT_RECV_COUNT.with_label_values(&["error"]).inc();
322                                    None
323                                }
324                            }
325                        } else {
326                            None
327                        }
328                    }
329                    _ = &mut sleep => {
330                        let region_stats = Self::load_region_stats(&region_server_clone);
331                        let topic_stats = region_server_clone.topic_stats();
332                        let now = Instant::now();
333                        let duration_since_epoch = (now - epoch).as_millis() as u64;
334
335                        let mut extensions = env_var_extensions.clone();
336                        let gc_stat = gc_limiter.gc_stat();
337                        gc_stat.into_extensions(&mut extensions);
338
339                        let mut req = HeartbeatRequest {
340                            region_stats,
341                            topic_stats,
342                            duration_since_epoch,
343                            extensions,
344                            ..heartbeat_request.clone()
345                        };
346
347                        if let Some(info) = req.info.as_mut() {
348                            info.cpu_usage_millicores = resource_stat.get_cpu_usage_millicores();
349                            info.memory_usage_bytes = resource_stat.get_memory_usage_bytes();
350                        }
351
352                        sleep.as_mut().reset(now + Duration::from_millis(interval));
353                        Some(req)
354                    }
355                    // If the heartbeat stream is broken, send a dummy heartbeat request to re-create the heartbeat stream.
356                    _ = quit_signal.notified() => {
357                        let req = HeartbeatRequest::default();
358                        Some(req)
359                    }
360                };
361                if let Some(req) = req {
362                    metrics::LAST_SENT_HEARTBEAT_ELAPSED
363                        .set(last_sent.elapsed().as_millis() as i64);
364                    // Resets the timer.
365                    last_sent = Instant::now();
366                    debug!("Sending heartbeat request: {:?}", req);
367                    if let Err(e) = tx.send(req).await {
368                        error!(e; "Failed to send heartbeat to metasrv");
369                        match Self::create_streams(
370                            &meta_client,
371                            running.clone(),
372                            handler_executor.clone(),
373                            mailbox.clone(),
374                            None,
375                            quit_signal.clone(),
376                        )
377                        .await
378                        {
379                            Ok((new_tx, new_config)) => {
380                                info!("Reconnected to metasrv, heartbeat config: {}", new_config);
381                                tx = new_tx;
382                                // Update retry_interval from new config
383                                retry_interval = new_config.retry_interval;
384                                // Update region_alive_keeper's heartbeat interval
385                                region_alive_keeper.update_heartbeat_interval(
386                                    new_config.interval.as_millis() as u64,
387                                );
388                                // Triggers to send heartbeat immediately.
389                                sleep.as_mut().reset(Instant::now());
390                            }
391                            Err(e) => {
392                                // Before the META_LEASE_SECS expires,
393                                // any retries are meaningless, it always reads the old meta leader address.
394                                // Triggers to retry after retry_interval from Metasrv config.
395                                sleep.as_mut().reset(Instant::now() + retry_interval);
396                                error!(e; "Failed to reconnect to metasrv!");
397                            }
398                        }
399                    } else {
400                        HEARTBEAT_SENT_COUNT.inc();
401                    }
402                }
403            }
404        });
405
406        Ok(())
407    }
408
409    fn load_region_stats(region_server: &RegionServer) -> Vec<RegionStat> {
410        region_server
411            .reportable_regions()
412            .into_iter()
413            .map(|stat| {
414                let region_stat = region_server
415                    .region_statistic(stat.region_id)
416                    .unwrap_or_default();
417                let mut extensions = HashMap::new();
418                if let Some(serialized) = region_stat.serialize_to_vec() {
419                    extensions.insert(REGION_STATISTIC_KEY.to_string(), serialized);
420                }
421
422                RegionStat {
423                    region_id: stat.region_id.as_u64(),
424                    engine: stat.engine,
425                    role: RegionRole::from(stat.role).into(),
426                    // TODO(weny): w/rcus
427                    rcus: 0,
428                    wcus: 0,
429                    approximate_bytes: region_stat.estimated_disk_size() as i64,
430                    extensions,
431                }
432            })
433            .collect()
434    }
435
436    pub fn close(&self) -> Result<()> {
437        let running = self.running.clone();
438        if running
439            .compare_exchange(true, false, Ordering::AcqRel, Ordering::Acquire)
440            .is_err()
441        {
442            warn!("Call close heartbeat task multiple times");
443        }
444
445        Ok(())
446    }
447}