AstralRinth/apps/labrinth/src/search/backend/elasticsearch/mod.rs

use crate::database::PgPool;
use crate::database::redis::RedisPool;
use crate::env::ENV;
use crate::models::ids::VersionId;
use crate::routes::ApiError;
use crate::search::backend::{
    SearchIndex, SearchIndexName, combined_search_filters, parse_search_index,
    parse_search_request,
};
use crate::search::indexing::index_local;
use crate::search::{
    ResultSearchProject, SearchBackend, SearchField, SearchRequest,
    SearchResults, TasksCancelFilter, UploadSearchProject,
};
use crate::util::error::Context;
use ariadne::ids::base62_impl::to_base62;
use async_trait::async_trait;
use elasticsearch::auth::Credentials;
use elasticsearch::http::Url;
use elasticsearch::http::request::JsonBody;
use elasticsearch::http::response::Response;
use elasticsearch::http::transport::{
    SingleNodeConnectionPool, TransportBuilder,
};
use elasticsearch::indices::{
    IndicesCreateParts, IndicesDeleteParts, IndicesExistsParts,
    IndicesGetAliasParts, IndicesRefreshParts,
};
use elasticsearch::params::Refresh;
use elasticsearch::tasks::TasksCancelParts;
use elasticsearch::{
    BulkParts, DeleteByQueryParts, Elasticsearch as EsClient, SearchParts,
};
use eyre::eyre;
use regex::Regex;
use reqwest::StatusCode;
use serde::{Deserialize, Serialize};
use serde_json::{Value, json};
use std::borrow::Cow;
use std::collections::HashMap;
use std::sync::LazyLock;
use std::time::Duration;

#[derive(Debug, Clone)]
pub struct ElasticsearchConfig {
    pub url: String,
    pub index_prefix: String,
    pub meta_namespace: String,
    pub username: String,
    pub password: String,
}

impl ElasticsearchConfig {
    pub fn new(meta_namespace: Option<String>) -> Self {
        Self {
            url: ENV.ELASTICSEARCH_URL.clone(),
            index_prefix: ENV.ELASTICSEARCH_INDEX_PREFIX.clone(),
            meta_namespace: meta_namespace.unwrap_or_default(),
            username: ENV.ELASTICSEARCH_USERNAME.clone(),
            password: ENV.ELASTICSEARCH_PASSWORD.clone(),
        }
    }

    pub fn get_index_name(&self, index: &str) -> String {
        if self.meta_namespace.is_empty() {
            format!("{}_{}", self.index_prefix, index)
        } else {
            format!("{}_{}_{}", self.meta_namespace, self.index_prefix, index)
        }
    }
}

pub struct Elasticsearch {
    pub config: ElasticsearchConfig,
    pub client: EsClient,
}

#[derive(Serialize, Deserialize, Debug, Clone, Default)]
pub struct RequestConfig {
    #[serde(default)]
    pub multi_match: MultiMatchConfig,
    #[serde(default)]
    pub sort: Vec<Value>,
}

#[derive(Serialize, Deserialize, Debug, Clone)]
pub struct MultiMatchConfig {
    #[serde(default = "default_multi_match_type", rename = "type")]
    pub ty: String,
    #[serde(default = "default_multi_match_fields")]
    pub fields: Vec<String>,
    #[serde(flatten, default)]
    pub extra: HashMap<String, Value>,
}

impl Default for MultiMatchConfig {
    fn default() -> Self {
        Self {
            ty: default_multi_match_type(),
            fields: default_multi_match_fields(),
            extra: HashMap::new(),
        }
    }
}

fn default_multi_match_type() -> String {
    "bool_prefix".to_string()
}

fn default_multi_match_fields() -> Vec<String> {
    vec![
        "name^8".to_string(),
        "name._2gram^8".to_string(),
        "name._3gram^8".to_string(),
        "slug^8".to_string(),
        "slug._2gram^8".to_string(),
        "slug._3gram^8".to_string(),
        "author^2".to_string(),
        "author._2gram^2".to_string(),
        "author._3gram^2".to_string(),
        "summary^3".to_string(),
        "summary._2gram^3".to_string(),
        "summary._3gram^3".to_string(),
    ]
}

pub struct ElasticsearchFieldSpec {
    pub path: &'static str,
    pub mapping: Value,
}

impl SearchField {
    pub fn elasticsearch_spec(self) -> ElasticsearchFieldSpec {
        match self {
            SearchField::Categories => ElasticsearchFieldSpec {
                path: "categories",
                mapping: json!({ "type": "keyword" }),
            },
            SearchField::ProjectTypes => ElasticsearchFieldSpec {
                path: "project_types",
                mapping: json!({ "type": "keyword" }),
            },
            SearchField::ProjectId => ElasticsearchFieldSpec {
                path: "project_id",
                mapping: json!({ "type": "keyword" }),
            },
            SearchField::OpenSource => ElasticsearchFieldSpec {
                path: "open_source",
                mapping: json!({ "type": "boolean" }),
            },
            SearchField::Environment => ElasticsearchFieldSpec {
                path: "environment",
                mapping: json!({ "type": "keyword" }),
            },
            SearchField::GameVersions => ElasticsearchFieldSpec {
                path: "game_versions",
                mapping: json!({ "type": "keyword" }),
            },
            SearchField::ClientSide => ElasticsearchFieldSpec {
                path: "client_side",
                mapping: json!({ "type": "keyword" }),
            },
            SearchField::ServerSide => ElasticsearchFieldSpec {
                path: "server_side",
                mapping: json!({ "type": "keyword" }),
            },
            SearchField::MinecraftServerRegion => ElasticsearchFieldSpec {
                path: "minecraft_server.region",
                mapping: json!({ "type": "keyword" }),
            },
            SearchField::MinecraftServerLanguages => ElasticsearchFieldSpec {
                path: "minecraft_server.languages",
                mapping: json!({ "type": "keyword" }),
            },
            SearchField::MinecraftJavaServerContentKind => {
                ElasticsearchFieldSpec {
                    path: "minecraft_java_server.content.kind",
                    mapping: json!({ "type": "keyword" }),
                }
            }
            SearchField::MinecraftJavaServerContentSupportedGameVersions => {
                ElasticsearchFieldSpec {
                    path: "minecraft_java_server.content.supported_game_versions",
                    mapping: json!({ "type": "keyword" }),
                }
            }
            SearchField::MinecraftJavaServerPingData => {
                ElasticsearchFieldSpec {
                    path: "minecraft_java_server.ping.data",
                    mapping: json!({ "type": "object" }),
                }
            }
        }
    }
}

static ELASTICSEARCH_PROPERTIES: LazyLock<serde_json::Map<String, Value>> =
    LazyLock::new(|| {
        use strum::IntoEnumIterator;

        let mut properties = serde_json::Map::from_iter([
            ("version_id".to_string(), json!({ "type": "keyword" })),
            (
                "slug".to_string(),
                json!({
                    "type": "search_as_you_type",
                    "fields": { "keyword": { "type": "keyword" } }
                }),
            ),
            (
                "author".to_string(),
                json!({
                    "type": "search_as_you_type",
                    "fields": { "keyword": { "type": "keyword" } }
                }),
            ),
            ("name".to_string(), json!({ "type": "search_as_you_type" })),
            (
                "summary".to_string(),
                json!({ "type": "search_as_you_type" }),
            ),
            (
                "display_categories".to_string(),
                json!({ "type": "keyword" }),
            ),
            ("downloads".to_string(), json!({ "type": "integer" })),
            ("follows".to_string(), json!({ "type": "integer" })),
            ("date_created".to_string(), json!({ "type": "date" })),
            ("created_timestamp".to_string(), json!({ "type": "long" })),
            ("date_modified".to_string(), json!({ "type": "date" })),
            ("modified_timestamp".to_string(), json!({ "type": "long" })),
            (
                "version_published_timestamp".to_string(),
                json!({ "type": "long" }),
            ),
            ("license".to_string(), json!({ "type": "keyword" })),
            ("loaders".to_string(), json!({ "type": "keyword" })),
            ("color".to_string(), json!({ "type": "long" })),
            ("environment".to_string(), json!({ "type": "keyword" })),
            ("mrpack_loaders".to_string(), json!({ "type": "keyword" })),
            (
                "minecraft_server.country".to_string(),
                json!({ "type": "keyword" }),
            ),
        ]);

        for field in SearchField::iter() {
            let spec = field.elasticsearch_spec();
            insert_nested_mapping(&mut properties, spec.path, spec.mapping);
        }

        insert_nested_mapping(
            &mut properties,
            "minecraft_java_server.content.recommended_game_version",
            json!({ "type": "keyword" }),
        );
        insert_nested_mapping(
            &mut properties,
            "minecraft_java_server.verified_plays_2w",
            json!({ "type": "long" }),
        );
        insert_nested_mapping(
            &mut properties,
            "minecraft_java_server.verified_plays_4w",
            json!({ "type": "long" }),
        );
        insert_nested_mapping(
            &mut properties,
            "minecraft_java_server.is_online",
            json!({ "type": "boolean" }),
        );
        insert_nested_mapping(
            &mut properties,
            "minecraft_java_server.ping.data.players_online",
            json!({ "type": "long" }),
        );

        let minecraft_java_server = properties
            .remove("minecraft_java_server")
            .unwrap_or_else(|| json!({ "properties": {} }));

        properties.insert(
            "minecraft_java_server".to_string(),
            json!({
                "type": "object",
                "properties": merge_object_properties(minecraft_java_server)
            }),
        );

        properties
    });

fn merge_object_properties(value: Value) -> Value {
    match value {
        Value::Object(mut obj) => {
            obj.remove("properties").unwrap_or_else(|| json!({}))
        }
        _ => json!({}),
    }
}

fn insert_nested_mapping(
    properties: &mut serde_json::Map<String, Value>,
    path: &str,
    mapping: Value,
) {
    let mut parts = path.split('.');
    let Some(first) = parts.next() else {
        return;
    };

    if let Some(rest) = parts.next() {
        let remaining = std::iter::once(rest)
            .chain(parts)
            .collect::<Vec<_>>()
            .join(".");
        let entry = properties
            .entry(first.to_string())
            .or_insert_with(|| json!({ "type": "object", "properties": {} }));
        let obj = entry.as_object_mut().expect("object mapping");
        let nested = obj
            .entry("properties".to_string())
            .or_insert_with(|| json!({}));
        let nested = nested.as_object_mut().expect("nested properties object");
        insert_nested_mapping(nested, &remaining, mapping);
    } else {
        properties.insert(first.to_string(), mapping);
    }
}

impl Elasticsearch {
    fn multi_match_query(
        query_text: &str,
        request_config: &RequestConfig,
    ) -> Value {
        let mut multi_match = serde_json::Map::from_iter([
            ("query".to_string(), Value::String(query_text.to_string())),
            (
                "type".to_string(),
                Value::String(request_config.multi_match.ty.clone()),
            ),
            (
                "fields".to_string(),
                Value::Array(
                    request_config
                        .multi_match
                        .fields
                        .clone()
                        .into_iter()
                        .map(Value::String)
                        .collect(),
                ),
            ),
        ]);

        for (key, value) in &request_config.multi_match.extra {
            if key != "query" {
                multi_match.insert(key.clone(), value.clone());
            }
        }

        json!({ "multi_match": Value::Object(multi_match) })
    }

    fn escape_query_string_value(value: &str) -> String {
        const RESERVED: [char; 21] = [
            '+', '-', '=', '&', '|', '>', '<', '!', '(', ')', '{', '}', '[',
            ']', '^', '"', '~', '*', '?', ':', '\\',
        ];

        let mut escaped = String::with_capacity(value.len());
        for ch in value.chars() {
            if RESERVED.contains(&ch) || ch == '/' {
                escaped.push('\\');
            }
            escaped.push(ch);
        }
        escaped
    }

    fn normalize_meili_filter_syntax(filters: &str) -> String {
        static IN_FILTER_RE: LazyLock<Regex> = LazyLock::new(|| {
            Regex::new(
                r"(?i)\b([a-zA-Z_][a-zA-Z0-9_]*)\s+(NOT\s+)?IN\s*\[([^\]]*)\]",
            )
            .expect("valid regex")
        });

        IN_FILTER_RE
            .replace_all(filters, |captures: &regex::Captures<'_>| {
                let field =
                    captures.get(1).map(|m| m.as_str()).unwrap_or_default();
                let is_not = captures.get(2).is_some();
                let list = captures
                    .get(3)
                    .map(|m| m.as_str())
                    .unwrap_or_default()
                    .split(',')
                    .map(str::trim)
                    .filter(|value| !value.is_empty())
                    .map(Self::escape_query_string_value)
                    .collect::<Vec<_>>();

                if list.is_empty() {
                    captures
                        .get(0)
                        .map(|m| m.as_str())
                        .unwrap_or_default()
                        .to_string()
                } else if is_not {
                    format!("NOT {field}:({})", list.join(" OR "))
                } else {
                    format!("{field}:({})", list.join(" OR "))
                }
            })
            .into_owned()
    }

    fn get_next_index_name(&self, alias_name: &str, next: bool) -> String {
        if next {
            format!("{alias_name}__alt")
        } else {
            format!("{alias_name}__current")
        }
    }

    fn get_index_candidates(&self, alias_name: &str) -> [String; 3] {
        [
            alias_name.to_string(),
            self.get_next_index_name(alias_name, false),
            self.get_next_index_name(alias_name, true),
        ]
    }

    fn parse_condition_query(condition: &str) -> Value {
        let (field, value, negative) =
            if let Some((f, v)) = condition.split_once("!=") {
                (f.trim(), v.trim(), true)
            } else if let Some((f, v)) = condition.split_once(':') {
                (f.trim(), v.trim(), false)
            } else if let Some((f, v)) = condition.split_once('=') {
                (f.trim(), v.trim(), false)
            } else {
                ("", "", false)
            };

        let field = match field {
            "project_type" => "project_types",
            "title" => "name",
            _ => field,
        };
        let clause = match field {
            // Search text fields are analyzed; phrase matching aligns with
            // Meilisearch behavior for quoted/multi-word values.
            "name" | "summary" | "author" | "slug" => json!({
                "match_phrase": {
                    field: {
                        "value": value,
                        "case_insensitive": true
                    }
                }
            }),
            "categories" | "license" | "project_types" | "project_id"
            | "environment" | "game_versions" | "mrpack_loaders"
            | "client_side" | "server_side" => json!({
                "term": {
                    field: {
                        "value": value,
                        "case_insensitive": true
                    }
                }
            }),
            _ => json!({
                "term": {
                    field: value
                }
            }),
        };

        if negative {
            json!({
                "bool": {
                    "must_not": [clause]
                }
            })
        } else {
            clause
        }
    }

    fn facets_filter_clauses(
        facets_json: Option<&str>,
    ) -> Result<Vec<Value>, ApiError> {
        let Some(raw_facets) = facets_json else {
            return Ok(Vec::new());
        };

        let facets = serde_json::from_str::<Vec<Vec<Value>>>(raw_facets)
            .wrap_request_err("failed to parse facets")?;

        let facets = facets
            .into_iter()
            .map(|facet_group| {
                facet_group
                    .into_iter()
                    .map(|facet| {
                        if facet.is_array() {
                            serde_json::from_value::<Vec<String>>(facet)
                                .unwrap_or_default()
                        } else {
                            vec![
                                serde_json::from_value::<String>(facet)
                                    .unwrap_or_default(),
                            ]
                        }
                    })
                    .collect::<Vec<Vec<String>>>()
            })
            .collect::<Vec<Vec<Vec<String>>>>();

        let mut clauses = Vec::new();
        for or_group in facets {
            let should = or_group
                .into_iter()
                .map(|and_group| {
                    let mut must = Vec::new();
                    let mut must_not = Vec::new();
                    for condition in and_group {
                        let q = Self::parse_condition_query(&condition);
                        if q.get("bool")
                            .and_then(|b| b.get("must_not"))
                            .is_some()
                        {
                            if let Some(parts) =
                                q["bool"]["must_not"].as_array()
                            {
                                must_not.extend(parts.iter().cloned());
                            }
                        } else {
                            must.push(q);
                        }
                    }

                    json!({
                        "bool": {
                            "must": must,
                            "must_not": must_not
                        }
                    })
                })
                .collect::<Vec<_>>();

            clauses.push(json!({
                "bool": {
                    "should": should,
                    "minimum_should_match": 1
                }
            }));
        }

        Ok(clauses)
    }

    pub fn new(meta_namespace: Option<String>) -> eyre::Result<Self> {
        let config = ElasticsearchConfig::new(meta_namespace);
        let url = Url::parse(&config.url)
            .wrap_err("failed to parse Elasticsearch URL")?;
        let mut builder =
            TransportBuilder::new(SingleNodeConnectionPool::new(url));

        let has_basic_username = !config.username.trim().is_empty();
        let has_basic_password = !config.password.trim().is_empty();
        if has_basic_username || has_basic_password {
            if !has_basic_username || !has_basic_password {
                return Err(eyre!(
                    "Elasticsearch basic auth requires both `ELASTICSEARCH_USERNAME` and `ELASTICSEARCH_PASSWORD`"
                ));
            }
            builder = builder.auth(Credentials::Basic(
                config.username.clone(),
                config.password.clone(),
            ));
        }

        let transport = builder
            .build()
            .wrap_err("failed to create Elasticsearch transport")?;
        let client = EsClient::new(transport);

        Ok(Self { config, client })
    }

    fn get_sort_index(
        &self,
        index: &str,
        new_filters: Option<&str>,
        request_config: &RequestConfig,
    ) -> Result<(String, Value), ApiError> {
        let sort = parse_search_index(index, new_filters)?;
        let index_name = match sort.index_name {
            SearchIndexName::Projects => self.config.get_index_name("projects"),
            SearchIndexName::ProjectsFiltered => {
                self.config.get_index_name("projects_filtered")
            }
        };

        let default_sort = match sort.index {
            SearchIndex::Relevance => (
                index_name,
                json!([
                    { "_score": { "order": "desc" } },
                    { "downloads": { "order": "desc" } },
                    { "version_published_timestamp": { "order": "desc" } }
                ]),
            ),
            SearchIndex::Downloads => (
                index_name,
                json!([
                    { "downloads": { "order": "desc" } },
                    { "version_published_timestamp": { "order": "desc" } }
                ]),
            ),
            SearchIndex::Follows => (
                index_name,
                json!([
                    { "follows": { "order": "desc" } },
                    { "version_published_timestamp": { "order": "desc" } }
                ]),
            ),
            SearchIndex::Updated => (
                index_name,
                json!([
                    { "date_modified": { "order": "desc" } },
                    { "version_published_timestamp": { "order": "desc" } }
                ]),
            ),
            SearchIndex::Newest => (
                index_name,
                json!([
                    { "date_created": { "order": "desc" } },
                    { "version_published_timestamp": { "order": "desc" } }
                ]),
            ),
            SearchIndex::MinecraftJavaServerVerifiedPlays2w => (
                index_name,
                json!([
                    {
                        "minecraft_java_server.verified_plays_2w": {
                            "order": "desc"
                        }
                    },
                    {
                        "minecraft_java_server.ping.data.players_online": {
                            "order": "desc"
                        }
                    },
                    { "version_published_timestamp": { "order": "desc" } }
                ]),
            ),
            SearchIndex::MinecraftJavaServerPlayersOnline => (
                index_name,
                json!([
                    {
                        "minecraft_java_server.ping.data.players_online": {
                            "order": "desc"
                        }
                    },
                    { "version_published_timestamp": { "order": "desc" } }
                ]),
            ),
        };

        if request_config.sort.is_empty() {
            Ok(default_sort)
        } else {
            Ok((default_sort.0, Value::Array(request_config.sort.clone())))
        }
    }

    async fn ensure_index(&self, index_name: &str) -> Result<(), ApiError> {
        let exists = self
            .client
            .indices()
            .exists(IndicesExistsParts::Index(&[index_name]))
            .send()
            .await
            .wrap_internal_err(
                "failed to check Elasticsearch index existence",
            )?;

        if exists.status_code().is_success() {
            return Ok(());
        }

        let response = self
            .client
            .indices()
            .create(IndicesCreateParts::Index(index_name))
            .body(json!({
                "mappings": {
                    "dynamic": true,
                    "properties": Value::Object(ELASTICSEARCH_PROPERTIES.clone())
                }
            }))
            .send()
            .await
            .wrap_internal_err("failed to create Elasticsearch index")?;

        if response.status_code().is_success() {
            Ok(())
        } else {
            let body =
                response.json::<Value>().await.unwrap_or_else(|_| json!({}));
            Err(ApiError::Internal(eyre!(
                "failed to create Elasticsearch index `{index_name}`: {body}"
            )))
        }
    }

    async fn delete_index_if_exists(
        &self,
        index_name: &str,
    ) -> Result<(), ApiError> {
        let delete = self
            .client
            .indices()
            .delete(IndicesDeleteParts::Index(&[index_name]))
            .send()
            .await
            .wrap_internal_err("failed to delete Elasticsearch index")?;

        let success_or_not_found = delete.status_code().is_success()
            || delete.status_code() == StatusCode::NOT_FOUND;

        if !success_or_not_found {
            let body =
                delete.json::<Value>().await.unwrap_or_else(|_| json!({}));
            return Err(ApiError::Internal(eyre!(
                "failed to delete Elasticsearch index `{index_name}`: {body}"
            )));
        }

        Ok(())
    }

    async fn get_alias_target(
        &self,
        alias_name: &str,
    ) -> Result<Option<String>, ApiError> {
        let response = self
            .client
            .indices()
            .get_alias(IndicesGetAliasParts::Name(&[alias_name]))
            .send()
            .await
            .wrap_internal_err("failed to get Elasticsearch alias")?;

        if response.status_code() == StatusCode::NOT_FOUND {
            return Ok(None);
        }

        if !response.status_code().is_success() {
            let body =
                response.json::<Value>().await.unwrap_or_else(|_| json!({}));
            return Err(ApiError::Internal(eyre!(
                "failed to get Elasticsearch alias `{alias_name}`: {body}"
            )));
        }

        let body = response.json::<Value>().await.wrap_internal_err(
            "failed to parse Elasticsearch alias response",
        )?;
        Ok(body
            .as_object()
            .and_then(|x| x.keys().next().cloned())
            .filter(|x| !x.is_empty()))
    }

    async fn index_exists(&self, index_name: &str) -> Result<bool, ApiError> {
        let exists = self
            .client
            .indices()
            .exists(IndicesExistsParts::Index(&[index_name]))
            .send()
            .await
            .wrap_internal_err(
                "failed to check Elasticsearch index existence",
            )?;
        Ok(exists.status_code().is_success())
    }

    async fn swap_alias(
        &self,
        alias_name: &str,
        next_index: &str,
        current_index: Option<&str>,
    ) -> Result<(), ApiError> {
        let mut actions = vec![json!({
            "add": {
                "index": next_index,
                "alias": alias_name
            }
        })];
        if let Some(current_index) = current_index {
            actions.push(json!({
                "remove": {
                    "index": current_index,
                    "alias": alias_name
                }
            }));
        }

        let response = self
            .client
            .indices()
            .update_aliases()
            .body(json!({ "actions": actions }))
            .send()
            .await
            .wrap_internal_err("failed to update Elasticsearch aliases")?;

        if !response.status_code().is_success() {
            let body =
                response.json::<Value>().await.unwrap_or_else(|_| json!({}));
            return Err(ApiError::Internal(eyre!(
                "failed to swap Elasticsearch alias `{alias_name}`: {body}"
            )));
        }

        Ok(())
    }

    async fn bulk_index_documents(
        &self,
        index_name: &str,
        docs: &[crate::search::UploadSearchProject],
    ) -> Result<(), ApiError> {
        if docs.is_empty() {
            return Ok(());
        }

        let mut body: Vec<JsonBody<Value>> = Vec::with_capacity(docs.len() * 2);
        for doc in docs {
            body.push(json!({"index": {"_id": doc.version_id}}).into());
            body.push(
                serde_json::to_value(doc)
                    .wrap_internal_err("failed to serialize document for Elasticsearch bulk index")?
                    .into(),
            );
        }

        let response = self
            .client
            .bulk(BulkParts::Index(index_name))
            .refresh(Refresh::WaitFor)
            .body(body)
            .send()
            .await
            .wrap_internal_err(
                "failed to request bulk index Elasticsearch documents",
            )?
            .error_for_status_code()
            .wrap_internal_err(
                "failed to bulk index Elasticsearch documents",
            )?;

        self.ensure_no_errors(response, "bulk index").await
    }

    async fn ensure_no_errors(
        &self,
        resp: Response,
        action: &str,
    ) -> Result<(), ApiError> {
        let body = resp
            .json::<Value>()
            .await
            .wrap_internal_err("failed to parse Elasticsearch response")?;
        if body.get("errors").and_then(Value::as_bool).unwrap_or(false) {
            return Err(ApiError::Internal(eyre!(
                "Elasticsearch `{action}` reported partial failures: {body}"
            )));
        }

        Ok(())
    }

    fn meili_like_filters(info: &SearchRequest) -> Option<Cow<'_, str>> {
        let raw = combined_search_filters(info)?;

        Some(Self::normalize_meili_filter_syntax(&raw).into())
    }
}

#[async_trait]
impl SearchBackend for Elasticsearch {
    async fn search_for_project_raw(
        &self,
        info: &SearchRequest,
    ) -> Result<SearchResults, ApiError> {
        let parsed = parse_search_request(info)?;
        let request_config = &info.elasticsearch_config;
        let (index_name, sort) = self.get_sort_index(
            parsed.index,
            info.new_filters.as_deref(),
            request_config,
        )?;
        let include_metadata = info.show_metadata;

        let mut must = Vec::new();
        let query_text = parsed.query.trim();
        if query_text.is_empty() {
            must.push(json!({"match_all": {}}));
        } else {
            must.push(Self::multi_match_query(query_text, request_config));
        }

        let mut filter = Self::facets_filter_clauses(info.facets.as_deref())?;
        if let Some(filter_string) = Self::meili_like_filters(info)
            && !filter_string.trim().is_empty()
        {
            filter.push(json!({
                "query_string": {
                    "query": filter_string,
                    "default_operator": "AND",
                    "lenient": true
                }
            }));
        }

        let response = self
            .client
            .search(SearchParts::Index(&[index_name.as_str()]))
            .from(parsed.offset as i64)
            .size(parsed.hits_per_page as i64)
            .track_total_hits(true)
            .body(json!({
                "query": {
                    "bool": {
                        "must": must,
                        "filter": filter
                    }
                },
                "collapse": {
                    "field": "project_id"
                },
                "sort": sort,
                "track_scores": include_metadata,
                "explain": include_metadata,
                "profile": include_metadata,
                "aggs": {
                    "unique_projects": {
                        "cardinality": {
                            "field": "project_id",
                        }
                    }
                }
            }))
            .send()
            .await
            .wrap_internal_err("failed to execute Elasticsearch search")?;

        if let Err(err) = response.error_for_status_code_ref() {
            let err = eyre!(err);
            let err = match response.json::<Value>().await {
                Ok(json) => err.wrap_err(eyre!(
                    "search request failed: {}",
                    serde_json::to_string_pretty(&json).unwrap()
                )),
                Err(_) => err.wrap_err("search request failed"),
            };
            return Err(ApiError::Internal(err));
        }

        let response_body = response.json::<Value>().await.wrap_internal_err(
            "failed to parse Elasticsearch search response",
        )?;

        let hits = response_body["hits"]["hits"]
            .as_array()
            .cloned()
            .unwrap_or_default()
            .into_iter()
            .map(|hit| -> Result<Option<ResultSearchProject>, ApiError> {
                let Some(source) = hit.get("_source").cloned() else {
                    return Ok(None);
                };

                let metadata = include_metadata.then(|| {
                    let mut metadata = serde_json::Map::new();

                    if let Some(score) = hit.get("_score") {
                        metadata.insert("_score".to_string(), score.clone());
                    }
                    if let Some(sort) = hit.get("sort") {
                        metadata.insert("sort".to_string(), sort.clone());
                    }
                    if let Some(explanation) = hit.get("_explanation") {
                        metadata.insert(
                            "_explanation".to_string(),
                            explanation.clone(),
                        );
                    }

                    Value::Object(metadata)
                });

                serde_json::from_value::<UploadSearchProject>(source)
                    .wrap_internal_err(
                        "failed to deserialize Elasticsearch hit",
                    )
                    .map(|project| {
                        let mut result: ResultSearchProject = project.into();
                        result.search_metadata = metadata;
                        Some(result)
                    })
            })
            .collect::<Result<Vec<_>, ApiError>>()?
            .into_iter()
            .flatten()
            .collect::<Vec<_>>();

        let total_hits = response_body.get("aggregations")
            .and_then(|aggs| aggs.get("unique_projects"))
            .and_then(|unique| unique.get("value"))
            .and_then(Value::as_u64)
            .map(|v| v as usize)
            .wrap_internal_err("missing `aggregations.unique_projects.value` in Elasticsearch response")?;

        Ok(SearchResults {
            hits,
            page: parsed.page,
            hits_per_page: parsed.hits_per_page,
            total_hits,
        })
    }

    async fn index_projects(
        &self,
        ro_pool: PgPool,
        redis: RedisPool,
    ) -> eyre::Result<()> {
        let projects_alias = self.config.get_index_name("projects");
        let filtered_alias = self.config.get_index_name("projects_filtered");

        let projects_current = self.get_alias_target(&projects_alias).await?;
        let filtered_current = self.get_alias_target(&filtered_alias).await?;
        let projects_legacy_current = projects_current.is_none()
            && self.index_exists(&projects_alias).await?;
        let filtered_legacy_current = filtered_current.is_none()
            && self.index_exists(&filtered_alias).await?;

        let projects_next = if projects_current
            .as_deref()
            .is_some_and(|x| x.ends_with("__alt"))
        {
            self.get_next_index_name(&projects_alias, false)
        } else {
            self.get_next_index_name(&projects_alias, true)
        };
        let filtered_next = if filtered_current
            .as_deref()
            .is_some_and(|x| x.ends_with("__alt"))
        {
            self.get_next_index_name(&filtered_alias, false)
        } else {
            self.get_next_index_name(&filtered_alias, true)
        };

        self.delete_index_if_exists(&projects_next).await?;
        self.delete_index_if_exists(&filtered_next).await?;
        self.ensure_index(&projects_next).await?;
        self.ensure_index(&filtered_next).await?;
        let mut cursor = 0_i64;

        loop {
            let (uploads, next_cursor) = index_local(
                &ro_pool,
                &redis,
                cursor,
                ENV.SEARCH_INDEX_CHUNK_SIZE,
            )
            .await
            .wrap_internal_err("failed to index local")?;
            if uploads.is_empty() {
                break;
            }

            self.bulk_index_documents(&projects_next, &uploads).await?;
            self.bulk_index_documents(&filtered_next, &uploads).await?;
            cursor = next_cursor;
        }

        let indices = [projects_next.as_str(), filtered_next.as_str()];
        self.client
            .indices()
            .refresh(IndicesRefreshParts::Index(&indices))
            .send()
            .await
            .wrap_internal_err("failed to refresh Elasticsearch indexes")?;

        if projects_legacy_current {
            self.delete_index_if_exists(&projects_alias).await?;
        }
        if filtered_legacy_current {
            self.delete_index_if_exists(&filtered_alias).await?;
        }

        self.swap_alias(
            &projects_alias,
            &projects_next,
            projects_current.as_deref(),
        )
        .await?;
        self.swap_alias(
            &filtered_alias,
            &filtered_next,
            filtered_current.as_deref(),
        )
        .await?;

        if let Some(index) = projects_current {
            self.delete_index_if_exists(&index).await?;
        }
        if let Some(index) = filtered_current {
            self.delete_index_if_exists(&index).await?;
        }

        Ok(())
    }

    async fn remove_documents(&self, ids: &[VersionId]) -> eyre::Result<()> {
        if ids.is_empty() {
            return Ok(());
        }

        let ids_base62 =
            ids.iter().map(|id| to_base62(id.0)).collect::<Vec<_>>();
        for alias_name in [
            self.config.get_index_name("projects"),
            self.config.get_index_name("projects_filtered"),
        ] {
            let index_names = self.get_index_candidates(&alias_name);
            for index_name in index_names {
                let response = self
                    .client
                    .delete_by_query(DeleteByQueryParts::Index(&[
                        index_name.as_str()
                    ]))
                    .refresh(true)
                    .body(json!({
                        "query": {
                            "terms": {
                                "version_id": ids_base62
                            }
                        }
                    }))
                    .send()
                    .await
                    .wrap_internal_err(
                        "failed to delete Elasticsearch documents by query",
                    )?;
                let status = response.status_code();
                if status == StatusCode::NOT_FOUND {
                    continue;
                }
                if !status.is_success() {
                    let body = response
                        .json::<Value>()
                        .await
                        .unwrap_or_else(|_| json!({}));
                    return Err(eyre!(
                        "failed to delete documents from index `{index_name}`: {body}"
                    ));
                }
            }
        }

        Ok(())
    }

    async fn tasks(&self) -> eyre::Result<Value> {
        #[derive(Serialize)]
        struct ElasticTask {
            uid: u64,
            status: &'static str,
            duration: Option<Duration>,
            enqueued_at: Option<u64>,
        }

        #[derive(Serialize)]
        struct TaskList {
            by_instance: HashMap<String, Vec<ElasticTask>>,
        }

        let response = self
            .client
            .tasks()
            .list()
            .detailed(true)
            .group_by(elasticsearch::params::GroupBy::Nodes)
            .send()
            .await
            .wrap_internal_err("failed to list Elasticsearch tasks")?;

        let body = response
            .json::<Value>()
            .await
            .wrap_internal_err("failed to parse Elasticsearch task response")?;

        let by_instance = body["nodes"]
            .as_object()
            .map(|nodes| {
                nodes
                    .iter()
                    .map(|(node_id, node_value)| {
                        let tasks = node_value["tasks"]
                            .as_object()
                            .map(|tasks| {
                                tasks
                                    .iter()
                                    .map(|(task_id, task)| {
                                        let uid = task_id
                                            .rsplit(':')
                                            .next()
                                            .and_then(|v| v.parse::<u64>().ok())
                                            .unwrap_or_default();
                                        let nanos =
                                            task["running_time_in_nanos"]
                                                .as_u64();
                                        ElasticTask {
                                            uid,
                                            status: "processing",
                                            duration: nanos
                                                .map(Duration::from_nanos),
                                            enqueued_at: task
                                                .get("start_time_in_millis")
                                                .and_then(Value::as_u64),
                                        }
                                    })
                                    .collect::<Vec<_>>()
                            })
                            .unwrap_or_default();
                        (node_id.clone(), tasks)
                    })
                    .collect::<HashMap<_, _>>()
            })
            .unwrap_or_default();

        let response = serde_json::to_value(TaskList { by_instance })
            .wrap_internal_err("failed to serialize Elasticsearch tasks")?;
        Ok(response)
    }

    async fn tasks_cancel(
        &self,
        filter: &TasksCancelFilter,
    ) -> eyre::Result<()> {
        match filter {
            TasksCancelFilter::All | TasksCancelFilter::AllEnqueued => {
                let response = self
                    .client
                    .tasks()
                    .cancel(TasksCancelParts::None)
                    .wait_for_completion(true)
                    .send()
                    .await
                    .wrap_internal_err(
                        "failed to cancel Elasticsearch tasks",
                    )?;
                if !response.status_code().is_success() {
                    let body = response
                        .json::<Value>()
                        .await
                        .unwrap_or_else(|_| json!({}));
                    return Err(eyre!(
                        "failed to cancel Elasticsearch tasks: {body}"
                    ));
                }
            }
            TasksCancelFilter::Indexes { indexes } => {
                let response = self
                    .client
                    .tasks()
                    .list()
                    .detailed(true)
                    .group_by(elasticsearch::params::GroupBy::None)
                    .send()
                    .await
                    .wrap_internal_err("failed to list Elasticsearch tasks")?;

                let body = response.json::<Value>().await.wrap_internal_err(
                    "failed to parse Elasticsearch tasks list",
                )?;
                let tasks =
                    body["tasks"].as_object().cloned().unwrap_or_default();

                for (task_id, task) in tasks {
                    let description =
                        task["description"].as_str().unwrap_or_default();
                    if indexes.iter().any(|index| description.contains(index)) {
                        let response = self
                            .client
                            .tasks()
                            .cancel(TasksCancelParts::TaskId(&task_id))
                            .wait_for_completion(true)
                            .send()
                            .await
                            .wrap_internal_err(
                                "failed to cancel Elasticsearch task by id",
                            )?;
                        if !response.status_code().is_success() {
                            let body = response
                                .json::<Value>()
                                .await
                                .unwrap_or_else(|_| json!({}));
                            return Err(eyre!(
                                "failed to cancel Elasticsearch task `{task_id}`: {body}"
                            ));
                        }
                    }
                }
            }
        }

        Ok(())
    }
}

#[cfg(test)]
mod tests {
    use super::Elasticsearch;
    use crate::search::{
        SearchRequest,
        backend::{elasticsearch::RequestConfig, typesense},
    };
    use serde_json::json;

    #[test]
    fn search_regression_not_in_filter_list_query_string() {
        // failing case:
        // http://localhost:8000/v2/search?facets=%5B%5B%22client_side%3Aoptional%22%2C%22client_side%3Arequired%22%5D%2C%5B%22project_type%3Amod%22%5D%2C%5B%22versions%3A1.8.9%22%2C%22versions%3A1.12.2%22%2C%22versions%3A1.17.1%22%2C%22versions%3A1.18.2%22%2C%22versions%3A1.19%22%2C%22versions%3A1.19.2%22%2C%22versions%3A1.19.3%22%2C%22versions%3A1.19.4%22%2C%22versions%3A1.20%22%2C%22versions%3A1.20.1%22%2C%22versions%3A1.20.2%22%2C%22versions%3A1.20.4%22%2C%22versions%3A1.20.6%22%2C%22versions%3A1.21%22%2C%22versions%3A1.21.1%22%2C%22versions%3A1.21.3%22%2C%22versions%3A1.21.4%22%2C%22versions%3A1.21.5%22%2C%22versions%3A1.21.7%22%2C%22versions%3A1.21.8%22%2C%22versions%3A1.21.10%22%2C%22versions%3A1.21.11%22%5D%5D&filters=(project_id%20NOT%20IN%20[P7dR8mSH,%20hvFnDODi,%20XaIYsn4W,%20xIEuGYOS,%20kqJFAPU9,%20H8CaAYZC,%203llatzyE,%20JyKlunuD])&index=relevance&limit=20&offset=0

        let facets = "[[\"client_side:optional\",\"client_side:required\"],[\"project_type:mod\"],[\"versions:1.8.9\",\"versions:1.12.2\",\"versions:1.17.1\",\"versions:1.18.2\",\"versions:1.19\",\"versions:1.19.2\",\"versions:1.19.3\",\"versions:1.19.4\",\"versions:1.20\",\"versions:1.20.1\",\"versions:1.20.2\",\"versions:1.20.4\",\"versions:1.20.6\",\"versions:1.21\",\"versions:1.21.1\",\"versions:1.21.3\",\"versions:1.21.4\",\"versions:1.21.5\",\"versions:1.21.7\",\"versions:1.21.8\",\"versions:1.21.10\",\"versions:1.21.11\"]]";
        let filter_query = "(project_id NOT IN [P7dR8mSH, hvFnDODi, XaIYsn4W, xIEuGYOS, kqJFAPU9, H8CaAYZC, 3llatzyE, JyKlunuD])";

        let info = SearchRequest {
            query: None,
            offset: Some("0".to_string()),
            index: Some("relevance".to_string()),
            limit: Some("20".to_string()),
            show_metadata: false,
            elasticsearch_config: RequestConfig::default(),
            typesense_config: typesense::RequestConfig::default(),
            new_filters: None,
            facets: Some(facets.to_string()),
            filters: Some(filter_query.to_string()),
            version: None,
        };

        let mut filter =
            Elasticsearch::facets_filter_clauses(info.facets.as_deref())
                .expect("facets should parse");
        let filter_string = Elasticsearch::meili_like_filters(&info)
            .expect("expected filter string");
        filter.push(json!({
            "query_string": {
                "query": filter_string,
                "default_operator": "AND",
                "lenient": true
            }
        }));

        let query = filter
            .last()
            .and_then(|x| x.get("query_string"))
            .and_then(|x| x.get("query"))
            .and_then(|x| x.as_str())
            .expect("expected query_string.query");

        let expected = "(NOT project_id:(P7dR8mSH OR hvFnDODi OR XaIYsn4W OR xIEuGYOS OR kqJFAPU9 OR H8CaAYZC OR 3llatzyE OR JyKlunuD))";
        assert_eq!(query, expected);
        assert!(
            !query.contains("NOT IN ["),
            "error case: Elasticsearch query_string cannot parse Meilisearch-style `NOT IN [..]` filters"
        );
    }
}