sui_indexer_alt_framework/pipeline/concurrent/committer.rs
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215
// Copyright (c) Mysten Labs, Inc.
// SPDX-License-Identifier: Apache-2.0
use std::{sync::Arc, time::Duration};
use backoff::ExponentialBackoff;
use tokio::{sync::mpsc, task::JoinHandle};
use tokio_stream::wrappers::ReceiverStream;
use tokio_util::sync::CancellationToken;
use tracing::{debug, error, info, warn};
use crate::{
metrics::{CheckpointLagMetricReporter, IndexerMetrics},
pipeline::{Break, CommitterConfig, WatermarkPart},
store::Store,
task::TrySpawnStreamExt,
};
use super::{BatchedRows, Handler};
/// If the committer needs to retry a commit, it will wait this long initially.
const INITIAL_RETRY_INTERVAL: Duration = Duration::from_millis(100);
/// If the committer needs to retry a commit, it will wait at most this long between retries.
const MAX_RETRY_INTERVAL: Duration = Duration::from_secs(1);
/// The committer task is responsible for writing batches of rows to the database. It receives
/// batches on `rx` and writes them out to the `db` concurrently (`config.write_concurrency`
/// controls the degree of fan-out).
///
/// The writing of each batch will be repeatedly retried on an exponential back-off until it
/// succeeds. Once the write succeeds, the [WatermarkPart]s for that batch are sent on `tx` to the
/// watermark task, as long as `skip_watermark` is not true.
///
/// This task will shutdown via its `cancel`lation token, or if its receiver or sender channels are
/// closed.
pub(super) fn committer<H: Handler + 'static>(
config: CommitterConfig,
skip_watermark: bool,
rx: mpsc::Receiver<BatchedRows<H>>,
tx: mpsc::Sender<Vec<WatermarkPart>>,
db: H::Store,
metrics: Arc<IndexerMetrics>,
cancel: CancellationToken,
) -> JoinHandle<()> {
tokio::spawn(async move {
info!(pipeline = H::NAME, "Starting committer");
let checkpoint_lag_reporter = CheckpointLagMetricReporter::new_for_pipeline::<H>(
&metrics.partially_committed_checkpoint_timestamp_lag,
&metrics.latest_partially_committed_checkpoint_timestamp_lag_ms,
&metrics.latest_partially_committed_checkpoint,
);
match ReceiverStream::new(rx)
.try_for_each_spawned(
config.write_concurrency,
|BatchedRows { values, watermark }| {
let values = Arc::new(values);
let tx = tx.clone();
let db = db.clone();
let metrics = metrics.clone();
let cancel = cancel.clone();
let checkpoint_lag_reporter = checkpoint_lag_reporter.clone();
// Repeatedly try to get a connection to the DB and write the batch. Use an
// exponential backoff in case the failure is due to contention over the DB
// connection pool.
let backoff = ExponentialBackoff {
initial_interval: INITIAL_RETRY_INTERVAL,
current_interval: INITIAL_RETRY_INTERVAL,
max_interval: MAX_RETRY_INTERVAL,
max_elapsed_time: None,
..Default::default()
};
let highest_checkpoint = watermark.iter().map(|w| w.checkpoint()).max();
let highest_checkpoint_timestamp =
watermark.iter().map(|w| w.timestamp_ms()).max();
use backoff::Error as BE;
let commit = move || {
let values = values.clone();
let db = db.clone();
let metrics = metrics.clone();
let checkpoint_lag_reporter = checkpoint_lag_reporter.clone();
async move {
if values.is_empty() {
return Ok(());
}
metrics
.total_committer_batches_attempted
.with_label_values(&[H::NAME])
.inc();
let guard = metrics
.committer_commit_latency
.with_label_values(&[H::NAME])
.start_timer();
let mut conn = db.connect().await.map_err(|e| {
warn!(
pipeline = H::NAME,
"Committed failed to get connection for DB"
);
metrics
.total_committer_batches_failed
.with_label_values(&[H::NAME])
.inc();
BE::transient(Break::Err(e))
})?;
let affected = H::commit(values.as_slice(), &mut conn).await;
let elapsed = guard.stop_and_record();
match affected {
Ok(affected) => {
debug!(
pipeline = H::NAME,
elapsed_ms = elapsed * 1000.0,
affected,
committed = values.len(),
"Wrote batch",
);
checkpoint_lag_reporter.report_lag(
// unwrap is safe because we would have returned if values is empty.
highest_checkpoint.unwrap(),
highest_checkpoint_timestamp.unwrap(),
);
metrics
.total_committer_batches_succeeded
.with_label_values(&[H::NAME])
.inc();
metrics
.total_committer_rows_committed
.with_label_values(&[H::NAME])
.inc_by(values.len() as u64);
metrics
.total_committer_rows_affected
.with_label_values(&[H::NAME])
.inc_by(affected as u64);
metrics
.committer_tx_rows
.with_label_values(&[H::NAME])
.observe(affected as f64);
Ok(())
}
Err(e) => {
warn!(
pipeline = H::NAME,
elapsed_ms = elapsed * 1000.0,
committed = values.len(),
"Error writing batch: {e}",
);
metrics
.total_committer_batches_failed
.with_label_values(&[H::NAME])
.inc();
Err(BE::transient(Break::Err(e)))
}
}
}
};
async move {
tokio::select! {
_ = cancel.cancelled() => {
return Err(Break::Cancel);
}
// Double check that the commit actually went through, (this backoff should
// not produce any permanent errors, but if it does, we need to shutdown
// the pipeline).
commit = backoff::future::retry(backoff, commit) => {
let () = commit?;
}
};
if !skip_watermark && tx.send(watermark).await.is_err() {
info!(pipeline = H::NAME, "Watermark closed channel");
return Err(Break::Cancel);
}
Ok(())
}
},
)
.await
{
Ok(()) => {
info!(pipeline = H::NAME, "Batches done, stopping committer");
}
Err(Break::Cancel) => {
info!(pipeline = H::NAME, "Shutdown received, stopping committer");
}
Err(Break::Err(e)) => {
error!(pipeline = H::NAME, "Error from committer: {e}");
cancel.cancel();
}
}
})
}