Spyke

Posts

lemmy_project_pri·Lemmy Project Priorities ObservationsbyRoundSparrow

The brand, hard not to see Twitter to X as similar

In April and May, Lemmy had a great brand... but server crashes and then lemmy.ml closed sign-up.... sort of confusing people and then Lemmy.world became the brand, then it too has had to close home page many times and constant crashes....

The SQL performance issues that were in lemmy for years really - became trouble once data was introduced.

I don't even know how many times people have come to sing-up and heard about Lemmy to find it slow, crashing, errroring.

Social hazing, rock star glamour attitude among the audience who keeps using Lemmy... but these odd trends in 2023 with social media aren't unique to lemmy, are they?

Threads, Twitter to X, Reddit API change, Lemmy crashing and not really caring about the crashes - and Redis or Memcache to mitigate it.... it's all such an odd year. It's as if everyone stirred the pot with social media but no place to really land upon that isn't crashing or in some form of Elon Musk kind of chaos.

I remember the rise of Reddit, the rise of Gmail - and it just felt like the growth was being dealt with. In 2023, Twitter seems to be leading the entire audience into accepting a kind of crash bad experience and 'stick around, no matter how bad it is' that Reddit users seem to have accepted.

I don't want to be a Lemmy developer... I want the people who know Rust Diesel and such to actually make it work - or choose something else that does work. I want Lemmy to actually not crash all the time and at least be where Reddit was in 2008 when there were just a few programmers doing it.

The SQL statements and server crashes in Lemmy speak for them self, just like the chaos of Reddit and Elon Musk Twitter speaks for itself... but a lot of people accept it. The audience is not asking for stability.

Strange times in 2023!

View original on lemmy.ml
0
lemmyfederation·Lemmy FederationbyRoundSparrow

Leveraging Lemmy's code as it is with federation and setting a community is 'local' on multiple servers at once

So this is currently a thought experiment or brainstorming, whatever...

When a community is local, the "home server" for community, that means that the federation is sent out to subscribed servers.

Now it also means that the total subscribe list is only know to that one server... As least I think it works that way, that the subscribe of "home" server is the complete list and everyone else has partial lists.

There is a lot of the structure of a community that offers some data opportunities to Lemmy. For one, a community object can be edited by any of the moderators. and any moderator can Lock a post, Remove a post, etc.

Reddit has a not-often-used Wiki feature, that even basically means more than one person could edit content too. Other than the sidebar of a community, I don't think Lemmy has any concept of multiple people being able to edit a post or a comment. But again, multiple mods can feature or lock a post... so there is some concept of multiple-actors on data.

I think when you get into creating flair / tags and even playlists (multi-community lists), you want multiple people to be able to edit data. Which in Reddit was the Wiki structure.

i know some of it is kind of a waste of time.. because people tend to avoid Wiki and highly favor posting the same repeat questions and content over and over, reposts. But a man can dream, can't he :)

View original on lemmy.ml
1
multipass·Lemmy MultiPassbyRoundSparrow

One Lemmy Instance admin says they will likely defederate those who allow clients to merge communities, multi-reddits

I did get permission to share this private message. For the record, I loath having public Lemmy topics being discussed in private messages. I created this community to get all this OUT in the OPEN.

I already made my decision on NOT changing privacy policy. But now I see that the very idea of having more than one "Subscribed" list for an end-user is raising threats of defederation!

View original on lemmy.ml
4

how to divorce two impl?

Help! I want to divorce ReadFn from ListFN - bypassing the Queries mutual closure behavior so I can better isolate some logic. My need is to get an independent ListFn...

fn queries_<'a>() -> Queries<
  impl ReadFn<'a, PostView, (PostId, Option, bool)>,
  impl ListFn<'a, PostView, PostQuery<'a>>,
> {

Context: https://github.com/LemmyNet/lemmy/blob/main/crates/db_views/src/post_view.rs

For sake of clarity... I'm not wanting to break the whole Queries joined closure marriage project-wide, just this one source file I want to be able to copy/paste this code twice and have just a single closure for ListFn.

Thank you.

View original on lemmy.ml
10

Trump made a mockery of American democracy. Why are Americans shrugging this off? | BY THE MIAMI HERALD EDITORIAL BOARD

Donald Trump faces four indictments, 91 criminal charges and hundreds of years of maximum prison time combined.

This is a former president who — according to the latest grand jury indictment in Fulton County, Georgia — participated in a “criminal enterprise.” Trump and 18 co-defendants are accused of trying “to unlawfully change the outcome of the election” in 2020. Among the 13 felony charges he faces is one count of violating the Georgia RICO (Racketeer Influenced and Corrupt Organizations) Act and two counts of conspiracy to commit forgery.

Most of those charges are related to a fake elector scheme by the Trump campaign in which a slate of “alternate” electors in Georgia would cast electoral votes for Trump instead of Joe Biden. The president of the most powerful democracy in the world allegedly tried to steal an election.

We can’t say it often enough: This is serious. Americans cannot shrug this off or normalize it, no matter how many times Trump gets indicted. Yet it feels like business as usual. Not only is Trump favored to win the GOP presidential nomination, he’s also neck and neck with President Biden in the 2024 general election, according to a July poll by the New York Times/Siena Poll.

MORE THAN A CULT

Trump’s support cannot only be explained as the product of the cult-like power he has over his MAGA base, which accounts for roughly 40% of Republican voters who believe those indictments are nothing but a conspiracy against him.

more: https://www.miamiherald.com/opinion/editorials/article278265068.html

https://www.miamiherald.com/opinion/editorials/article278265068.htmlOpen linkView original on lemmy.ml
573

I need help with Lemmy code, Diesel object - can I remove these SQL joins that aren't needed?

I'm trying to wrangle in and get 'back to basics' with Lemmy's Diesel code and at every turn I run into not understanding the complexity of the Rust code.

You may want to do a GitHub checkout of this branch if you want to see what I'm attempting: https://github.com/LemmyNet/lemmy/pull/3865

I'm basing my experiments off the code in that pull request, which links to this branch on this repository: https://github.com/dullbananas/lemmy/tree/post-view-same-joins

Right now Lemmy's Diesel code spins up many SQL table joins and for an anonymous user it just passes a -1 user id to all the joins - and it really makes for difficult to read SQL statements. So I decided to experiment with removing as much logic as I could to get the bare-bones behavior on generating the desired SQL statement....

I copied/pasted the queries function/method and gave it a new name, kept removing as much as I could see that referenced the user being logged-in vs. anonymous, and got to this point:

fn queries_anonymous<'a>() -> Queries<
  impl ReadFn<'a, PostView, (PostId, Option, bool)>,
  impl ListFn<'a, PostView, PostQuery<'a>>,
> {
  let is_creator_banned_from_community = exists(
    community_person_ban::table.filter(
      post_aggregates::community_id
        .eq(community_person_ban::community_id)
        .and(community_person_ban::person_id.eq(post_aggregates::creator_id)),
    ),
  );

// how do we eliminate these next 3 assignments, this is anonymous user, not needed

  let is_saved = |person_id_join| {
    exists(
      post_saved::table.filter(
        post_aggregates::post_id
          .eq(post_saved::post_id)
          .and(post_saved::person_id.eq(person_id_join)),
      ),
    )
  };

  let is_read = |person_id_join| {
    exists(
      post_read::table.filter(
        post_aggregates::post_id
          .eq(post_read::post_id)
          .and(post_read::person_id.eq(person_id_join)),
      ),
    )
  };

  let is_creator_blocked = |person_id_join| {
    exists(
      person_block::table.filter(
        post_aggregates::creator_id
          .eq(person_block::target_id)
          .and(person_block::person_id.eq(person_id_join)),
      ),
    )
  };

  let all_joins = move |query: post_aggregates::BoxedQuery<'a, Pg>, my_person_id: Option| {
    // The left join below will return None in this case
    let person_id_join = my_person_id.unwrap_or(PersonId(-1));

    query
      .inner_join(person::table)
      .inner_join(community::table)
      .inner_join(post::table)
// how do we eliminate these next 3 joins that are user/person references?
      .left_join(
        community_follower::table.on(
          post_aggregates::community_id
            .eq(community_follower::community_id)
        ),
      )
      .left_join(
        community_moderator::table.on(
          post::community_id
            .eq(community_moderator::community_id)
        ),
      )
      .left_join(
        post_like::table.on(
          post_aggregates::post_id
            .eq(post_like::post_id)
        ),
      )
      .left_join(
        person_post_aggregates::table.on(
          post_aggregates::post_id
            .eq(person_post_aggregates::post_id)
        ),
      )
      .select((
        post::all_columns,
        person::all_columns,
        community::all_columns,
        is_creator_banned_from_community,
        post_aggregates::all_columns,
        CommunityFollower::select_subscribed_type(),
// how do we eliminate these next 3 for anonymous?
        is_saved(person_id_join),
        is_read(person_id_join),
        is_creator_blocked(person_id_join),
        post_like::score.nullable(),
        coalesce(
          post_aggregates::comments.nullable() - person_post_aggregates::read_comments.nullable(),
          post_aggregates::comments,
        ),
      ))
  };

  let read =
    move |mut conn: DbConn<'a>,
          (post_id, my_person_id, is_mod_or_admin): (PostId, Option, bool)| async move {

      let mut query = all_joins(
        post_aggregates::table
          .filter(post_aggregates::post_id.eq(post_id))
          .into_boxed(),
        my_person_id,
      );

        query = query
          .filter(community::removed.eq(false))
          .filter(post::removed.eq(false))
          ;

      query.first::(&mut conn).await
    };

  let list = move |mut conn: DbConn<'a>, options: PostQuery<'a>| async move {
    let person_id = options.local_user.map(|l| l.person.id);

    let mut query = all_joins(post_aggregates::table.into_boxed(), person_id);


      query = query
        .filter(community::deleted.eq(false))
        .filter(post::deleted.eq(false));


    // every SELECT has to labor away on removed filtering
      query = query
        .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 let Some(creator_id) = options.creator_id {
      query = query.filter(post_aggregates::creator_id.eq(creator_id));
    }


    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);
      query = query.filter(
        post::name
          .ilike(searcher.clone())
          .or(post::body.ilike(searcher)),
      );
    }

      query = query
        .filter(post::nsfw.eq(false))
        .filter(community::nsfw.eq(false));


    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::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()),
    };

    let (limit, offset) = limit_and_offset(options.page, options.limit)?;

    query = query.limit(limit).offset(offset);

    debug!("Post View Query: {:?}", debug_query::(&query));

    query.load::(&mut conn).await
  };

  Queries::new(read, list)
}

This compiles, but I can not progress further. There are 3 joins more that aren't really needed for an anonymous user... but the interdependent Rust object structures I can't unravel enough to remove them from the code.

For example, Lemmy allows you to "save" a post, but an anonymous user doesn't have that ability - so how can I remove the JOIN + select related to that while still satisfying the object requirements? I even tried creating a variation of PostViewTuple object without one of the bool fields, but it all cascades into 50 lines of compiler errors. Thank you.

View original on lemmy.ml
7

Can I accomplish this in a single SQL statement?

SELECT id
    FROM my_table
    WHERE id IN (
     SELECT id
     FROM my_table
     WHERE criteria_a = 19
     ORDER BY create_when DESC
     LIMIT 1000
  );

This is the pattern I am looking for, but I need the criteria_a to be repeated for every value of criteria_a with the important focus being the LIMIT 1000 for any single value of criteria_a. There is no need to put a total LIMIT on the query, just to limit to the 1000 per criteria_a with the specific ORDER BY at that point. Put another way...

SELECT id
    FROM my_table
    WHERE id IN (
          SELECT id
		 FROM my_table
		 WHERE criteria_a = 19
		 ORDER BY create_when DESC
		 LIMIT 1000
	)
       OR id IN (
	  SELECT id
		 FROM my_table
		 WHERE criteria_a = 20
		 ORDER BY create_when DESC
		 LIMIT 1000
     );

Where I desire 2000 total rows. I could turn this into programming code (even a PostgreSQL FUNCTION) that loops over every value of criteria_a and replaces 19 in the example.

I don't care of it is a JOIN or an IN, I'm more stuck on how to repeat the inner SELECT with the LIMIT 1000 based on sort and criteria_a. Can I do it without looping and/or UNION? Thank you.

View original on lemmy.ml
7

Lemmy server mass update of comment reply (child) count with PostgreSQL ltree structure

lemmy_server PostgreSQL table for comment does not keep parent comment id directly, it uses a path field of ltree type.

by default, every comment has a path of it's own primary key id.

comment id 101, path = "0.101"
comment id 102, path = "0.102"
comment id 103, path = "0.101.103"
comment id 104, path = "0.101.103.104"

comment 103 is a reply to comment 101, 104 is a reply to 103.

A second table named comment_aggregates has a count field with comment_id column linking to comment table id key. On each new comment reply, lemmy_server issues an update statement to update the counts on every parent in the tree. Rust code issues this to PostgreSQL:

        if let Some(parent_id) = parent_id {
          let top_parent = format!("0.{}", parent_id);
          let update_child_count_stmt = format!(
            "
update comment_aggregates ca set child_count = c.child_count
from (
  select c.id, c.path, count(c2.id) as child_count from comment c
  join comment c2 on c2.path <@ c.path and c2.path != c.path
  and c.path <@ '{top_parent}'
  group by c.id
) as c
where ca.comment_id = c.id"
          );
      sql_query(update_child_count_stmt).execute(conn).await?;
    }
    

I've been playing with doing bulk INSERT of thousands of comments at once to test SELECT query performance.

So far, this is the only SQL statement I have found that does a mass UPDATE of child_count from path for the entire comment table:

UPDATE
    comment_aggregates ca
SET
    child_count = c2.child_count
FROM (
    SELECT
        c.id,
        c.path,
        count(c2.id) AS child_count
    FROM
        comment c
    LEFT JOIN comment c2 ON c2.path <@ c.path
        AND c2.path != c.path
GROUP BY
    c.id) AS c2
WHERE
    ca.comment_id = c2.id;

There are 1 to 2 millions comments stored on lemmy.ml and lemmy.world - this rebuild of child_count can take hours, and may not complete at all. Even on 100,000 rows in a test system, it's a harsh UPDATE statement to execute. EDIT: I found my API connection to production server was timing out and the run-time on the total rebuild isn't as bad as I thought. With my testing system I'm also finding it is taking under 19 seconds with 312684 comments. The query does seem to execute and run normal, not stuck.

Anyone have suggestions on how to improve this and help make Lemmy PostgreSQL servers more efficient?

EDIT: lemmy 0.18.3 and 0.18.4 are munging the less-than and greater-than signs in these code blocks.

View original on lemmy.ml
21