Fix Posts List Performance + cursor-based pagination (#3872)

* add token-based pagination + fast subscribed post view * add migrations * fix failing heuristic * revert * output pagination token as next_page, fix off-by-one, restructure * more cleanup * clean * format sql * fix comment * fix tests * e * empty * move last page thing * restructure a bit for readability * rename page_cursor * update for scaled sort * fix * sql format * fix * get rid of macros --------- Co-authored-by: Dessalines <dessalines@users.noreply.github.com>
2024-11-22 01:11:02 +00:00 · 2023-09-18 15:44:48 +02:00 · 2023-09-18 15:44:48 +02:00 · 544d30f0d4
commit 544d30f0d4
parent 6735a98d35
6 changed files with 414 additions and 82 deletions
--- a/crates/api_common/src/post.rs
+++ b/crates/api_common/src/post.rs
@ -5,7 +5,7 @@ use lemmy_db_schema::{
  PostFeatureType,
  SortType,
 };
-use lemmy_db_views::structs::{PostReportView, PostView};
+use lemmy_db_views::structs::{PaginationCursor, PostReportView, PostView};
 use lemmy_db_views_actor::structs::{CommunityModeratorView, CommunityView};
 use serde::{Deserialize, Serialize};
 use serde_with::skip_serializing_none;
@ -70,6 +70,7 @@ pub struct GetPostResponse {
 pub struct GetPosts {
  pub type_: Option<ListingType>,
  pub sort: Option<SortType>,
+  /// DEPRECATED, use page_cursor
  pub page: Option<i64>,
  pub limit: Option<i64>,
  pub community_id: Option<CommunityId>,
@ -78,6 +79,7 @@ pub struct GetPosts {
  pub liked_only: Option<bool>,
  pub disliked_only: Option<bool>,
  pub auth: Option<Sensitive<String>>,
+  pub page_cursor: Option<PaginationCursor>,
 }

 #[derive(Serialize, Deserialize, Debug, Clone)]
@ -86,6 +88,8 @@ pub struct GetPosts {
 /// The post list response.
 pub struct GetPostsResponse {
  pub posts: Vec<PostView>,
+  /// the pagination cursor to use to fetch the next page
+  pub next_page: Option<PaginationCursor>,
 }

 #[derive(Debug, Serialize, Deserialize, Clone, Default)]
--- a/crates/apub/src/api/list_posts.rs
+++ b/crates/apub/src/api/list_posts.rs
@ -11,7 +11,10 @@ use lemmy_api_common::{
  utils::{check_private_instance, local_user_view_from_jwt_opt_new},
 };
 use lemmy_db_schema::source::{community::Community, local_site::LocalSite};
-use lemmy_db_views::{post_view::PostQuery, structs::LocalUserView};
+use lemmy_db_views::{
+  post_view::PostQuery,
+  structs::{LocalUserView, PaginationCursor},
+};
 use lemmy_utils::error::{LemmyError, LemmyErrorExt, LemmyErrorType};

 #[tracing::instrument(skip(context))]
@ -48,6 +51,12 @@ pub async fn list_posts(
    &local_site,
    community_id,
  )?);
+  // parse pagination token
+  let page_after = if let Some(pa) = &data.page_cursor {
+    Some(pa.read(&mut context.pool()).await?)
+  } else {
+    None
+  };

  let posts = PostQuery {
    local_user: local_user_view.as_ref(),
@ -58,6 +67,7 @@ pub async fn list_posts(
    liked_only,
    disliked_only,
    page,
+    page_after,
    limit,
    ..Default::default()
  }
@ -65,5 +75,7 @@ pub async fn list_posts(
  .await
  .with_lemmy_type(LemmyErrorType::CouldntGetPosts)?;

-  Ok(Json(GetPostsResponse { posts }))
+  // if this page wasn't empty, then there is a next page after the last post on this page
+  let next_page = posts.last().map(PaginationCursor::after_post);
+  Ok(Json(GetPostsResponse { posts, next_page }))
 }
--- a/crates/db_views/src/post_view.rs
+++ b/crates/db_views/src/post_view.rs
@ -1,21 +1,26 @@
-use crate::structs::{LocalUserView, PostView};
+use crate::structs::{LocalUserView, PaginationCursor, PostView};
 use diesel::{
  debug_query,
-  dsl::{exists, not, IntervalDsl},
+  dsl::{self, exists, not, IntervalDsl},
+  expression::AsExpression,
  pg::Pg,
  result::Error,
  sql_function,
-  sql_types::{self, Timestamptz},
+  sql_types::{self, SingleValue, SqlType, Timestamptz},
  BoolExpressionMethods,
  BoxableExpression,
+  Expression,
  ExpressionMethods,
  IntoSql,
+  JoinOnDsl,
  NullableExpressionMethods,
+  OptionalExtension,
  PgTextExpressionMethods,
  QueryDsl,
 };
 use diesel_async::RunQueryDsl;
 use lemmy_db_schema::{
+  aggregates::structs::PostAggregates,
  newtypes::{CommunityId, LocalUserId, PersonId, PostId},
  schema::{
    community,
@ -28,12 +33,12 @@ use lemmy_db_schema::{
    person_block,
    person_post_aggregates,
    post,
-    post_aggregates,
+    post_aggregates::{self, newest_comment_time},
    post_like,
    post_read,
    post_saved,
  },
-  utils::{fuzzy_search, limit_and_offset, DbConn, DbPool, ListFn, Queries, ReadFn},
+  utils::{fuzzy_search, get_conn, limit_and_offset, DbConn, DbPool, ListFn, Queries, ReadFn},
  ListingType,
  SortType,
 };
@ -41,6 +46,55 @@ use tracing::debug;

 sql_function!(fn coalesce(x: sql_types::Nullable<sql_types::BigInt>, y: sql_types::BigInt) -> sql_types::BigInt);

+fn order_and_page_filter_desc<Q, C, T>(
+  query: Q,
+  column: C,
+  options: &PostQuery,
+  getter: impl Fn(&PostAggregates) -> T,
+) -> Q
+where
+  Q: diesel::query_dsl::methods::ThenOrderDsl<dsl::Desc<C>, Output = Q>
+    + diesel::query_dsl::methods::ThenOrderDsl<dsl::Asc<C>, Output = Q>
+    + diesel::query_dsl::methods::FilterDsl<dsl::GtEq<C, T>, Output = Q>
+    + diesel::query_dsl::methods::FilterDsl<dsl::LtEq<C, T>, Output = Q>,
+  C: Expression + Copy,
+  C::SqlType: SingleValue + SqlType,
+  T: AsExpression<C::SqlType>,
+{
+  let mut query = query.then_order_by(column.desc());
+  if let Some(before) = &options.page_before_or_equal {
+    query = query.filter(column.ge(getter(&before.0)));
+  }
+  if let Some(after) = &options.page_after {
+    query = query.filter(column.le(getter(&after.0)));
+  }
+  query
+}
+
+fn order_and_page_filter_asc<Q, C, T>(
+  query: Q,
+  column: C,
+  options: &PostQuery,
+  getter: impl Fn(&PostAggregates) -> T,
+) -> Q
+where
+  Q: diesel::query_dsl::methods::ThenOrderDsl<dsl::Asc<C>, Output = Q>
+    + diesel::query_dsl::methods::FilterDsl<dsl::LtEq<C, T>, Output = Q>
+    + diesel::query_dsl::methods::FilterDsl<dsl::GtEq<C, T>, Output = Q>,
+  C: Expression + Copy,
+  C::SqlType: SingleValue + SqlType,
+  T: AsExpression<C::SqlType>,
+{
+  let mut query = query.then_order_by(column.asc());
+  if let Some(before) = &options.page_before_or_equal {
+    query = query.filter(column.le(getter(&before.0)));
+  }
+  if let Some(after) = &options.page_after {
+    query = query.filter(column.ge(getter(&after.0)));
+  }
+  query
+}
+
 fn queries<'a>() -> Queries<
  impl ReadFn<'a, PostView, (PostId, Option<PersonId>, bool)>,
  impl ListFn<'a, PostView, PostQuery<'a>>,
@ -251,13 +305,18 @@ fn queries<'a>() -> Queries<
        .filter(community::removed.eq(false))
        .filter(post::removed.eq(false));
    }
-
-    if options.community_id.is_none() {
-      query = query.then_order_by(post_aggregates::featured_local.desc());
-    } else if let Some(community_id) = options.community_id {
-      query = query
-        .filter(post_aggregates::community_id.eq(community_id))
-        .then_order_by(post_aggregates::featured_community.desc());
+    if options.community_id.is_none() || options.community_id_just_for_prefetch {
+      query = order_and_page_filter_desc(query, post_aggregates::featured_local, &options, |e| {
+        e.featured_local
+      });
+    } else {
+      query =
+        order_and_page_filter_desc(query, post_aggregates::featured_community, &options, |e| {
+          e.featured_community
+        });
+    }
+    if let Some(community_id) = options.community_id {
+      query = query.filter(post_aggregates::community_id.eq(community_id));
    }

    if let Some(creator_id) = options.creator_id {
@ -292,12 +351,12 @@ fn queries<'a>() -> Queries<
      }
    }

-    if let Some(url_search) = options.url_search {
+    if let Some(url_search) = &options.url_search {
      query = query.filter(post::url.eq(url_search));
    }

-    if let Some(search_term) = options.search_term {
-      let searcher = fuzzy_search(&search_term);
+    if let Some(search_term) = &options.search_term {
+      let searcher = fuzzy_search(search_term);
      query = query.filter(
        post::name
          .ilike(searcher.clone())
@ -373,70 +432,91 @@ fn queries<'a>() -> Queries<
    }
    let now = diesel::dsl::now.into_sql::<Timestamptz>();

-    query = match options.sort.unwrap_or(SortType::Hot) {
-      SortType::Active => query
-        .then_order_by(post_aggregates::hot_rank_active.desc())
-        .then_order_by(post_aggregates::published.desc()),
-      SortType::Hot => query
-        .then_order_by(post_aggregates::hot_rank.desc())
-        .then_order_by(post_aggregates::published.desc()),
-      SortType::Scaled => query
-        .then_order_by(post_aggregates::scaled_rank.desc())
-        .then_order_by(post_aggregates::published.desc()),
-      SortType::Controversial => query.then_order_by(post_aggregates::controversy_rank.desc()),
-      SortType::New => query.then_order_by(post_aggregates::published.desc()),
-      SortType::Old => query.then_order_by(post_aggregates::published.asc()),
-      SortType::NewComments => query.then_order_by(post_aggregates::newest_comment_time.desc()),
-      SortType::MostComments => query
-        .then_order_by(post_aggregates::comments.desc())
-        .then_order_by(post_aggregates::published.desc()),
-      SortType::TopAll => query
-        .then_order_by(post_aggregates::score.desc())
-        .then_order_by(post_aggregates::published.desc()),
-      SortType::TopYear => query
-        .filter(post_aggregates::published.gt(now - 1.years()))
-        .then_order_by(post_aggregates::score.desc())
-        .then_order_by(post_aggregates::published.desc()),
-      SortType::TopMonth => query
-        .filter(post_aggregates::published.gt(now - 1.months()))
-        .then_order_by(post_aggregates::score.desc())
-        .then_order_by(post_aggregates::published.desc()),
-      SortType::TopWeek => query
-        .filter(post_aggregates::published.gt(now - 1.weeks()))
-        .then_order_by(post_aggregates::score.desc())
-        .then_order_by(post_aggregates::published.desc()),
-      SortType::TopDay => query
-        .filter(post_aggregates::published.gt(now - 1.days()))
-        .then_order_by(post_aggregates::score.desc())
-        .then_order_by(post_aggregates::published.desc()),
-      SortType::TopHour => query
-        .filter(post_aggregates::published.gt(now - 1.hours()))
-        .then_order_by(post_aggregates::score.desc())
-        .then_order_by(post_aggregates::published.desc()),
-      SortType::TopSixHour => query
-        .filter(post_aggregates::published.gt(now - 6.hours()))
-        .then_order_by(post_aggregates::score.desc())
-        .then_order_by(post_aggregates::published.desc()),
-      SortType::TopTwelveHour => query
-        .filter(post_aggregates::published.gt(now - 12.hours()))
-        .then_order_by(post_aggregates::score.desc())
-        .then_order_by(post_aggregates::published.desc()),
-      SortType::TopThreeMonths => query
-        .filter(post_aggregates::published.gt(now - 3.months()))
-        .then_order_by(post_aggregates::score.desc())
-        .then_order_by(post_aggregates::published.desc()),
-      SortType::TopSixMonths => query
-        .filter(post_aggregates::published.gt(now - 6.months()))
-        .then_order_by(post_aggregates::score.desc())
-        .then_order_by(post_aggregates::published.desc()),
-      SortType::TopNineMonths => query
-        .filter(post_aggregates::published.gt(now - 9.months()))
-        .then_order_by(post_aggregates::score.desc())
-        .then_order_by(post_aggregates::published.desc()),
+    {
+      use post_aggregates::{
+        comments,
+        controversy_rank,
+        hot_rank,
+        hot_rank_active,
+        published,
+        scaled_rank,
+        score,
+      };
+      match options.sort.as_ref().unwrap_or(&SortType::Hot) {
+        SortType::Active => {
+          query =
+            order_and_page_filter_desc(query, hot_rank_active, &options, |e| e.hot_rank_active);
+          query = order_and_page_filter_desc(query, published, &options, |e| e.published);
+        }
+        SortType::Hot => {
+          query = order_and_page_filter_desc(query, hot_rank, &options, |e| e.hot_rank);
+          query = order_and_page_filter_desc(query, published, &options, |e| e.published);
+        }
+        SortType::Scaled => {
+          query = order_and_page_filter_desc(query, scaled_rank, &options, |e| e.scaled_rank);
+          query = order_and_page_filter_desc(query, published, &options, |e| e.published);
+        }
+        SortType::Controversial => {
+          query =
+            order_and_page_filter_desc(query, controversy_rank, &options, |e| e.controversy_rank);
+          query = order_and_page_filter_desc(query, published, &options, |e| e.published);
+        }
+        SortType::New => {
+          query = order_and_page_filter_desc(query, published, &options, |e| e.published)
+        }
+        SortType::Old => {
+          query = order_and_page_filter_asc(query, published, &options, |e| e.published)
+        }
+        SortType::NewComments => {
+          query = order_and_page_filter_desc(query, newest_comment_time, &options, |e| {
+            e.newest_comment_time
+          })
+        }
+        SortType::MostComments => {
+          query = order_and_page_filter_desc(query, comments, &options, |e| e.comments);
+          query = order_and_page_filter_desc(query, published, &options, |e| e.published);
+        }
+        SortType::TopAll => {
+          query = order_and_page_filter_desc(query, score, &options, |e| e.score);
+          query = order_and_page_filter_desc(query, published, &options, |e| e.published);
+        }
+        o @ (SortType::TopYear
+        | SortType::TopMonth
+        | SortType::TopWeek
+        | SortType::TopDay
+        | SortType::TopHour
+        | SortType::TopSixHour
+        | SortType::TopTwelveHour
+        | SortType::TopThreeMonths
+        | SortType::TopSixMonths
+        | SortType::TopNineMonths) => {
+          let interval = match o {
+            SortType::TopYear => 1.years(),
+            SortType::TopMonth => 1.months(),
+            SortType::TopWeek => 1.weeks(),
+            SortType::TopDay => 1.days(),
+            SortType::TopHour => 1.hours(),
+            SortType::TopSixHour => 6.hours(),
+            SortType::TopTwelveHour => 12.hours(),
+            SortType::TopThreeMonths => 3.months(),
+            SortType::TopSixMonths => 6.months(),
+            SortType::TopNineMonths => 9.months(),
+            _ => return Err(Error::NotFound),
+          };
+          query = query.filter(post_aggregates::published.gt(now - interval));
+          query = order_and_page_filter_desc(query, score, &options, |e| e.score);
+          query = order_and_page_filter_desc(query, published, &options, |e| e.published);
+        }
+      }
    };

-    let (limit, offset) = limit_and_offset(options.page, options.limit)?;
-
+    let (limit, mut offset) = limit_and_offset(options.page, options.limit)?;
+    if options.page_after.is_some() {
+      // always skip exactly one post because that's the last post of the previous page
+      // fixing the where clause is more difficult because we'd have to change only the last order-by-where clause
+      // e.g. WHERE (featured_local<=, hot_rank<=, published<=) to WHERE (<=, <=, <)
+      offset = 1;
+    }
    query = query.limit(limit).offset(offset);

    debug!("Post View Query: {:?}", debug_query::<Pg, _>(&query));
@ -468,12 +548,42 @@ impl PostView {
  }
 }

-#[derive(Default)]
+impl PaginationCursor {
+  // get cursor for page that starts immediately after the given post
+  pub fn after_post(view: &PostView) -> PaginationCursor {
+    // hex encoding to prevent ossification
+    PaginationCursor(format!("P{:x}", view.counts.post_id.0))
+  }
+  pub async fn read(&self, pool: &mut DbPool<'_>) -> Result<PaginationCursorData, Error> {
+    Ok(PaginationCursorData(
+      PostAggregates::read(
+        pool,
+        PostId(
+          self
+            .0
+            .get(1..)
+            .and_then(|e| i32::from_str_radix(e, 16).ok())
+            .ok_or_else(|| Error::QueryBuilderError("Could not parse pagination token".into()))?,
+        ),
+      )
+      .await?,
+    ))
+  }
+}
+
+// currently we use a postaggregates struct as the pagination token.
+// we only use some of the properties of the post aggregates, depending on which sort type we page by
+#[derive(Clone)]
+pub struct PaginationCursorData(PostAggregates);
+
+#[derive(Default, Clone)]
 pub struct PostQuery<'a> {
  pub listing_type: Option<ListingType>,
  pub sort: Option<SortType>,
  pub creator_id: Option<PersonId>,
  pub community_id: Option<CommunityId>,
+  // if true, the query should be handled as if community_id was not given except adding the literal filter
+  pub community_id_just_for_prefetch: bool,
  pub local_user: Option<&'a LocalUserView>,
  pub search_term: Option<String>,
  pub url_search: Option<String>,
@ -484,11 +594,97 @@ pub struct PostQuery<'a> {
  pub is_profile_view: bool,
  pub page: Option<i64>,
  pub limit: Option<i64>,
+  pub page_after: Option<PaginationCursorData>,
+  pub page_before_or_equal: Option<PaginationCursorData>,
 }

 impl<'a> PostQuery<'a> {
+  async fn prefetch_upper_bound_for_page_before(
+    &self,
+    pool: &mut DbPool<'_>,
+  ) -> Result<Option<PostQuery<'a>>, Error> {
+    // first get one page for the most popular community to get an upper bound for the the page end for the real query
+    // the reason this is needed is that when fetching posts for a single community PostgreSQL can optimize
+    // the query to use an index on e.g. (=, >=, >=, >=) and fetch only LIMIT rows
+    // but for the followed-communities query it has to query the index on (IN, >=, >=, >=)
+    // which it currently can't do at all (as of PG 16). see the discussion here:
+    // https://github.com/LemmyNet/lemmy/issues/2877#issuecomment-1673597190
+    //
+    // the results are correct no matter which community we fetch these for, since it basically covers the "worst case" of the whole page consisting of posts from one community
+    // but using the largest community decreases the pagination-frame so make the real query more efficient.
+    use lemmy_db_schema::schema::{
+      community_aggregates::dsl::{community_aggregates, community_id, users_active_month},
+      community_follower::dsl::{
+        community_follower,
+        community_id as follower_community_id,
+        person_id,
+      },
+    };
+    let (limit, offset) = limit_and_offset(self.page, self.limit)?;
+    if offset != 0 {
+      return Err(Error::QueryBuilderError(
+        "legacy pagination cannot be combined with v2 pagination".into(),
+      ));
+    }
+    let self_person_id = self
+      .local_user
+      .expect("part of the above if")
+      .local_user
+      .person_id;
+    let largest_subscribed = {
+      let conn = &mut get_conn(pool).await?;
+      community_follower
+        .filter(person_id.eq(self_person_id))
+        .inner_join(community_aggregates.on(community_id.eq(follower_community_id)))
+        .order_by(users_active_month.desc())
+        .select(community_id)
+        .limit(1)
+        .get_result::<CommunityId>(conn)
+        .await
+        .optional()?
+    };
+    let Some(largest_subscribed) = largest_subscribed else {
+      // nothing subscribed to? no posts
+      return Ok(None);
+    };
+
+    let mut v = queries()
+      .list(
+        pool,
+        PostQuery {
+          community_id: Some(largest_subscribed),
+          community_id_just_for_prefetch: true,
+          ..self.clone()
+        },
+      )
+      .await?;
+    // take last element of array. if this query returned less than LIMIT elements,
+    // the heuristic is invalid since we can't guarantee the full query will return >= LIMIT results (return original query)
+    if (v.len() as i64) < limit {
+      Ok(Some(self.clone()))
+    } else {
+      let page_before_or_equal = Some(PaginationCursorData(v.pop().expect("else case").counts));
+      Ok(Some(PostQuery {
+        page_before_or_equal,
+        ..self.clone()
+      }))
+    }
+  }
+
  pub async fn list(self, pool: &mut DbPool<'_>) -> Result<Vec<PostView>, Error> {
-    queries().list(pool, self).await
+    if self.listing_type == Some(ListingType::Subscribed)
+      && self.community_id.is_none()
+      && self.local_user.is_some()
+      && self.page_before_or_equal.is_none()
+    {
+      if let Some(query) = self.prefetch_upper_bound_for_page_before(pool).await? {
+        queries().list(pool, query).await
+      } else {
+        Ok(vec![])
+      }
+    } else {
+      queries().list(pool, self).await
+    }
  }
 }

--- a/crates/db_views/src/structs.rs
+++ b/crates/db_views/src/structs.rs
@ -89,6 +89,13 @@ pub struct PostReportView {
  pub resolver: Option<Person>,
 }

+/// currently this is just a wrapper around post id, but should be seen as opaque from the client's perspective
+/// stringified since we might want to use arbitrary info later, with a P prepended to prevent ossification
+/// (api users love to make assumptions (e.g. parse stuff that looks like numbers as numbers) about apis that aren't part of the spec
+#[derive(Serialize, Deserialize, Debug, Clone)]
+#[cfg_attr(feature = "full", derive(ts_rs::TS))]
+pub struct PaginationCursor(pub(crate) String);
+
 #[skip_serializing_none]
 #[derive(Debug, PartialEq, Serialize, Deserialize, Clone)]
 #[cfg_attr(feature = "full", derive(TS, Queryable))]
--- a/migrations/2023-09-07-215546_post-queries-efficient/down.sql
+++ b/migrations/2023-09-07-215546_post-queries-efficient/down.sql
@ -0,0 +1,54 @@
+DROP INDEX idx_post_aggregates_featured_community_active;
+
+DROP INDEX idx_post_aggregates_featured_community_controversy;
+
+DROP INDEX idx_post_aggregates_featured_community_hot;
+
+DROP INDEX idx_post_aggregates_featured_community_scaled;
+
+DROP INDEX idx_post_aggregates_featured_community_most_comments;
+
+DROP INDEX idx_post_aggregates_featured_community_newest_comment_time;
+
+DROP INDEX idx_post_aggregates_featured_community_newest_comment_time_necro;
+
+DROP INDEX idx_post_aggregates_featured_community_published;
+
+DROP INDEX idx_post_aggregates_featured_community_score;
+
+CREATE INDEX idx_post_aggregates_featured_community_active ON post_aggregates (featured_community DESC, hot_rank_active DESC, published DESC);
+
+CREATE INDEX idx_post_aggregates_featured_community_controversy ON post_aggregates (featured_community DESC, controversy_rank DESC);
+
+CREATE INDEX idx_post_aggregates_featured_community_hot ON post_aggregates (featured_community DESC, hot_rank DESC, published DESC);
+
+CREATE INDEX idx_post_aggregates_featured_community_scaled ON post_aggregates (featured_community DESC, scaled_rank DESC, published DESC);
+
+CREATE INDEX idx_post_aggregates_featured_community_most_comments ON post_aggregates (featured_community DESC, comments DESC, published DESC);
+
+CREATE INDEX idx_post_aggregates_featured_community_newest_comment_time ON post_aggregates (featured_community DESC, newest_comment_time DESC);
+
+CREATE INDEX idx_post_aggregates_featured_community_newest_comment_time_necro ON post_aggregates (featured_community DESC, newest_comment_time_necro DESC);
+
+CREATE INDEX idx_post_aggregates_featured_community_published ON post_aggregates (featured_community DESC, published DESC);
+
+CREATE INDEX idx_post_aggregates_featured_community_score ON post_aggregates (featured_community DESC, score DESC, published DESC);
+
+DROP INDEX idx_post_aggregates_community_active;
+
+DROP INDEX idx_post_aggregates_community_controversy;
+
+DROP INDEX idx_post_aggregates_community_hot;
+
+DROP INDEX idx_post_aggregates_community_scaled;
+
+DROP INDEX idx_post_aggregates_community_most_comments;
+
+DROP INDEX idx_post_aggregates_community_newest_comment_time;
+
+DROP INDEX idx_post_aggregates_community_newest_comment_time_necro;
+
+DROP INDEX idx_post_aggregates_community_published;
+
+DROP INDEX idx_post_aggregates_community_score;
+
--- a/migrations/2023-09-07-215546_post-queries-efficient/up.sql
+++ b/migrations/2023-09-07-215546_post-queries-efficient/up.sql
@ -0,0 +1,59 @@
+-- these indices are used for single-community filtering and for the followed-communities (front-page) query
+-- basically one index per Sort
+-- index name is truncated to 63 chars so abbreviate a bit
+CREATE INDEX idx_post_aggregates_community_active ON post_aggregates (community_id, featured_local DESC, hot_rank_active DESC, published DESC);
+
+CREATE INDEX idx_post_aggregates_community_controversy ON post_aggregates (community_id, featured_local DESC, controversy_rank DESC);
+
+CREATE INDEX idx_post_aggregates_community_hot ON post_aggregates (community_id, featured_local DESC, hot_rank DESC, published DESC);
+
+CREATE INDEX idx_post_aggregates_community_scaled ON post_aggregates (community_id, featured_local DESC, scaled_rank DESC, published DESC);
+
+CREATE INDEX idx_post_aggregates_community_most_comments ON post_aggregates (community_id, featured_local DESC, comments DESC, published DESC);
+
+CREATE INDEX idx_post_aggregates_community_newest_comment_time ON post_aggregates (community_id, featured_local DESC, newest_comment_time DESC);
+
+CREATE INDEX idx_post_aggregates_community_newest_comment_time_necro ON post_aggregates (community_id, featured_local DESC, newest_comment_time_necro DESC);
+
+CREATE INDEX idx_post_aggregates_community_published ON post_aggregates (community_id, featured_local DESC, published DESC);
+
+CREATE INDEX idx_post_aggregates_community_score ON post_aggregates (community_id, featured_local DESC, score DESC, published DESC);
+
+-- these indices are used for "per-community" filtering
+-- these indices weren't really useful because whenever the query filters by featured_community it also filters by community_id, so prepend that to all these indexes
+DROP INDEX idx_post_aggregates_featured_community_active;
+
+DROP INDEX idx_post_aggregates_featured_community_controversy;
+
+DROP INDEX idx_post_aggregates_featured_community_hot;
+
+DROP INDEX idx_post_aggregates_featured_community_scaled;
+
+DROP INDEX idx_post_aggregates_featured_community_most_comments;
+
+DROP INDEX idx_post_aggregates_featured_community_newest_comment_time;
+
+DROP INDEX idx_post_aggregates_featured_community_newest_comment_time_necro;
+
+DROP INDEX idx_post_aggregates_featured_community_published;
+
+DROP INDEX idx_post_aggregates_featured_community_score;
+
+CREATE INDEX idx_post_aggregates_featured_community_active ON post_aggregates (community_id, featured_community DESC, hot_rank_active DESC, published DESC);
+
+CREATE INDEX idx_post_aggregates_featured_community_controversy ON post_aggregates (community_id, featured_community DESC, controversy_rank DESC);
+
+CREATE INDEX idx_post_aggregates_featured_community_hot ON post_aggregates (community_id, featured_community DESC, hot_rank DESC, published DESC);
+
+CREATE INDEX idx_post_aggregates_featured_community_scaled ON post_aggregates (community_id, featured_community DESC, scaled_rank DESC, published DESC);
+
+CREATE INDEX idx_post_aggregates_featured_community_most_comments ON post_aggregates (community_id, featured_community DESC, comments DESC, published DESC);
+
+CREATE INDEX idx_post_aggregates_featured_community_newest_comment_time ON post_aggregates (community_id, featured_community DESC, newest_comment_time DESC);
+
+CREATE INDEX idx_post_aggregates_featured_community_newest_comment_time_necro ON post_aggregates (community_id, featured_community DESC, newest_comment_time_necro DESC);
+
+CREATE INDEX idx_post_aggregates_featured_community_published ON post_aggregates (community_id, featured_community DESC, published DESC);
+
+CREATE INDEX idx_post_aggregates_featured_community_score ON post_aggregates (community_id, featured_community DESC, score DESC, published DESC);
+