Batch hot rank updates (#3175)

This commit is contained in:
Sander Saarend 2023-06-27 11:13:51 +03:00 committed by GitHub
parent f5209fffc1
commit 211e76dc27
No known key found for this signature in database
GPG key ID: 4AEE18F83AFDEB23
5 changed files with 92 additions and 66 deletions

1
Cargo.lock generated
View file

@ -2786,6 +2786,7 @@ dependencies = [
"activitypub_federation", "activitypub_federation",
"actix-cors", "actix-cors",
"actix-web", "actix-web",
"chrono",
"clokwerk", "clokwerk",
"console-subscriber", "console-subscriber",
"diesel", "diesel",

View file

@ -147,4 +147,4 @@ rustls = { workspace = true }
futures-util = { workspace = true } futures-util = { workspace = true }
tokio-postgres = { workspace = true } tokio-postgres = { workspace = true }
tokio-postgres-rustls = { workspace = true } tokio-postgres-rustls = { workspace = true }
chrono = { workspace = true }

View file

@ -0,0 +1,2 @@
drop index idx_comment_aggregates_published;
drop index idx_community_aggregates_published;

View file

@ -0,0 +1,4 @@
-- Add indexes on published column (needed for hot_rank updates)
create index idx_community_aggregates_published on community_aggregates (published desc);
create index idx_comment_aggregates_published on comment_aggregates (published desc);

View file

@ -1,28 +1,21 @@
use chrono::NaiveDateTime;
use clokwerk::{Scheduler, TimeUnits as CTimeUnits}; use clokwerk::{Scheduler, TimeUnits as CTimeUnits};
use diesel::{ use diesel::{
dsl::{now, IntervalDsl}, dsl::{now, IntervalDsl},
sql_types::{Integer, Timestamp},
Connection, Connection,
ExpressionMethods, ExpressionMethods,
NullableExpressionMethods, NullableExpressionMethods,
QueryDsl, QueryDsl,
QueryableByName,
}; };
// Import week days and WeekDay // Import week days and WeekDay
use diesel::{sql_query, PgConnection, RunQueryDsl}; use diesel::{sql_query, PgConnection, RunQueryDsl};
use lemmy_api_common::context::LemmyContext; use lemmy_api_common::context::LemmyContext;
use lemmy_db_schema::{ use lemmy_db_schema::{
schema::{ schema::{activity, comment, community_person_ban, instance, person, post},
activity,
comment,
comment_aggregates,
community_aggregates,
community_person_ban,
instance,
person,
post,
post_aggregates,
},
source::instance::{Instance, InstanceForm}, source::instance::{Instance, InstanceForm},
utils::{functions::hot_rank, naive_now, DELETED_REPLACEMENT_TEXT}, utils::{naive_now, DELETED_REPLACEMENT_TEXT},
}; };
use lemmy_routes::nodeinfo::NodeInfo; use lemmy_routes::nodeinfo::NodeInfo;
use lemmy_utils::{error::LemmyError, REQWEST_TIMEOUT}; use lemmy_utils::{error::LemmyError, REQWEST_TIMEOUT};
@ -49,9 +42,9 @@ pub fn setup(
update_banned_when_expired(&mut conn); update_banned_when_expired(&mut conn);
}); });
// Update hot ranks every 5 minutes // Update hot ranks every 15 minutes
let url = db_url.clone(); let url = db_url.clone();
scheduler.every(CTimeUnits::minutes(5)).run(move || { scheduler.every(CTimeUnits::minutes(15)).run(move || {
let mut conn = PgConnection::establish(&url).expect("could not establish connection"); let mut conn = PgConnection::establish(&url).expect("could not establish connection");
update_hot_ranks(&mut conn, true); update_hot_ranks(&mut conn, true);
}); });
@ -100,66 +93,92 @@ fn startup_jobs(db_url: &str) {
} }
/// Update the hot_rank columns for the aggregates tables /// Update the hot_rank columns for the aggregates tables
/// Runs in batches until all necessary rows are updated once
fn update_hot_ranks(conn: &mut PgConnection, last_week_only: bool) { fn update_hot_ranks(conn: &mut PgConnection, last_week_only: bool) {
let mut post_update = diesel::update(post_aggregates::table).into_boxed(); let process_start_time = if last_week_only {
let mut comment_update = diesel::update(comment_aggregates::table).into_boxed();
let mut community_update = diesel::update(community_aggregates::table).into_boxed();
// Only update for the last week of content
if last_week_only {
info!("Updating hot ranks for last week..."); info!("Updating hot ranks for last week...");
let last_week = now - diesel::dsl::IntervalDsl::weeks(1); naive_now() - chrono::Duration::days(7)
post_update = post_update.filter(post_aggregates::published.gt(last_week));
comment_update = comment_update.filter(comment_aggregates::published.gt(last_week));
community_update = community_update.filter(community_aggregates::published.gt(last_week));
} else { } else {
info!("Updating hot ranks for all history..."); info!("Updating hot ranks for all history...");
} NaiveDateTime::from_timestamp_opt(0, 0).expect("0 timestamp creation")
};
match post_update process_hot_ranks_in_batches(
.set(( conn,
post_aggregates::hot_rank.eq(hot_rank(post_aggregates::score, post_aggregates::published)), "post_aggregates",
post_aggregates::hot_rank_active.eq(hot_rank( "SET hot_rank = hot_rank(a.score, a.published),
post_aggregates::score, hot_rank_active = hot_rank(a.score, a.newest_comment_time_necro)",
post_aggregates::newest_comment_time_necro, process_start_time,
)), );
process_hot_ranks_in_batches(
conn,
"comment_aggregates",
"SET hot_rank = hot_rank(a.score, a.published)",
process_start_time,
);
process_hot_ranks_in_batches(
conn,
"community_aggregates",
"SET hot_rank = hot_rank(a.subscribers, a.published)",
process_start_time,
);
info!("Finished hot ranks update!");
}
#[derive(QueryableByName)]
struct HotRanksUpdateResult {
#[diesel(sql_type = Timestamp)]
published: NaiveDateTime,
}
/// Runs the hot rank update query in batches until all rows after `process_start_time` have been
/// processed.
/// In `set_clause`, "a" will refer to the current aggregates table.
/// Locked rows are skipped in order to prevent deadlocks (they will likely get updated on the next
/// run)
fn process_hot_ranks_in_batches(
conn: &mut PgConnection,
table_name: &str,
set_clause: &str,
process_start_time: NaiveDateTime,
) {
let update_batch_size = 1000; // Bigger batches than this tend to cause seq scans
let mut previous_batch_result = Some(process_start_time);
while let Some(previous_batch_last_published) = previous_batch_result {
// Raw `sql_query` is used as a performance optimization - Diesel does not support doing this
// in a single query (neither as a CTE, nor using a subquery)
let result = sql_query(format!(
r#"WITH batch AS (SELECT a.id
FROM {aggregates_table} a
WHERE a.published > $1
ORDER BY a.published
LIMIT $2
FOR UPDATE SKIP LOCKED)
UPDATE {aggregates_table} a {set_clause}
FROM batch WHERE a.id = batch.id RETURNING a.published;
"#,
aggregates_table = table_name,
set_clause = set_clause
)) ))
.execute(conn) .bind::<Timestamp, _>(previous_batch_last_published)
{ .bind::<Integer, _>(update_batch_size)
Ok(_) => {} .get_results::<HotRanksUpdateResult>(conn);
Err(e) => {
error!("Failed to update post_aggregates hot_ranks: {}", e)
}
}
match comment_update match result {
.set(comment_aggregates::hot_rank.eq(hot_rank( Ok(updated_rows) => previous_batch_result = updated_rows.last().map(|row| row.published),
comment_aggregates::score, Err(e) => {
comment_aggregates::published, error!("Failed to update {} hot_ranks: {}", table_name, e);
))) break;
.execute(conn) }
{
Ok(_) => {}
Err(e) => {
error!("Failed to update comment_aggregates hot_ranks: {}", e)
}
}
match community_update
.set(community_aggregates::hot_rank.eq(hot_rank(
community_aggregates::subscribers,
community_aggregates::published,
)))
.execute(conn)
{
Ok(_) => {
info!("Done.");
}
Err(e) => {
error!("Failed to update community_aggregates hot_ranks: {}", e)
} }
} }
info!(
"Finished process_hot_ranks_in_batches execution for {}",
table_name
);
} }
/// Clear old activities (this table gets very large) /// Clear old activities (this table gets very large)