import { Health, Index, MeiliSearch, Stats } from "meilisearch"; import { dbLogger } from "./logger.js"; import config from "@/config/index.js"; import { Note } from "@/models/entities/note.js"; import * as url from "url"; import { ILocalUser } from "@/models/entities/user.js"; import { Followings, Users } from "@/models/index.js"; const logger = dbLogger.createSubLogger("meilisearch", "gray", false); let posts: Index; let client: MeiliSearch; const hasConfig = config.meilisearch && (config.meilisearch.host || config.meilisearch.port || config.meilisearch.apiKey); if (hasConfig) { const host = hasConfig ? config.meilisearch.host ?? "localhost" : ""; const port = hasConfig ? config.meilisearch.port ?? 7700 : 0; const auth = hasConfig ? config.meilisearch.apiKey ?? "" : ""; const ssl = hasConfig ? config.meilisearch.ssl ?? false : false; logger.info("Connecting to MeiliSearch"); client = new MeiliSearch({ host: `${ssl ? "https" : "http"}://${host}:${port}`, apiKey: auth, }); posts = client.index("posts"); posts .updateSearchableAttributes(["text"]) .catch((e) => logger.error(`Setting searchable attr failed, searches won't work: ${e}`), ); posts .updateFilterableAttributes([ "userName", "userHost", "mediaAttachment", "createdAt", "userId", ]) .catch((e) => logger.error( `Setting filterable attr failed, advanced searches won't work: ${e}`, ), ); posts .updateSortableAttributes(["createdAt"]) .catch((e) => logger.error( `Setting sortable attr failed, placeholder searches won't sort properly: ${e}`, ), ); posts .updateStopWords([ "the", "a", "as", "be", "of", "they", "these", "これ", "それ", "あれ", "この", "その", "あの", "ここ", "そこ", "あそこ", "こちら", "どこ", "だれ", "なに", "なん", "何", "私", "貴方", "貴方方", "我々", "私達", "あの人", "あのか", "彼女", "彼", "です", "ありま", "おりま", "います", "は", "が", "の", "に", "を", "で", "え", "から", "まで", "より", "も", "どの", "と", "し", "それで", "しかし", ]) .catch((e) => logger.error( `Failed to set Meilisearch stop words, database size will be larger: ${e}`, ), ); posts .updateRankingRules([ "sort", "words", "typo", "proximity", "attribute", "exactness", ]) .catch((e) => { logger.error("Failed to set ranking rules, sorting won't work properly."); }); logger.info("Connected to MeiliSearch"); } export type MeilisearchNote = { id: string; text: string; userId: string; userHost: string; userName: string; channelId: string; mediaAttachment: string; createdAt: number; }; function timestampToUnix(timestamp: string) { let unix = 0; // Only contains numbers => UNIX timestamp if (/^\d+$/.test(timestamp)) { unix = Number.parseInt(timestamp); } if (unix === 0) { // Try to parse the timestamp as JavaScript Date const date = Date.parse(timestamp); if (isNaN(date)) return 0; unix = date / 1000; } return unix; } export default hasConfig ? { search: async ( query: string, limit: number, offset: number, userCtx: ILocalUser | null, overrideSort: string | null, ) => { /// Advanced search syntax /// from:user => filter by user + optional domain /// has:image/video/audio/text/file => filter by attachment types /// domain:domain.com => filter by domain /// before:Date => show posts made before Date /// after: Date => show posts made after Date /// "text" => get posts with exact text between quotes /// filter:following => show results only from users you follow /// filter:followers => show results only from followers /// order:desc/asc => order results ascending or descending const constructedFilters: string[] = []; let sortRules: string[] = []; const splitSearch = query.split(" "); // Detect search operators and remove them from the actual query const filteredSearchTerms = ( await Promise.all( splitSearch.map(async (term) => { if (term.startsWith("has:")) { const fileType = term.slice(4); constructedFilters.push(`mediaAttachment = "${fileType}"`); return null; } else if (term.startsWith("from:")) { let user = term.slice(5); if (user.length === 0) return null; // Cut off leading @, those aren't saved in the DB if (user.charAt(0) === "@") { user = user.slice(1); } // Determine if we got a webfinger address or a single username if (user.split("@").length > 1) { const splitUser = user.split("@"); const domain = splitUser.pop(); user = splitUser.join("@"); constructedFilters.push( `userName = ${user} AND userHost = ${domain}`, ); } else { constructedFilters.push(`userName = ${user}`); } return null; } else if (term.startsWith("domain:")) { const domain = term.slice(7); constructedFilters.push(`userHost = ${domain}`); return null; } else if (term.startsWith("after:")) { const timestamp = term.slice(6); const unix = timestampToUnix(timestamp); if (unix !== 0) constructedFilters.push(`createdAt > ${unix}`); return null; } else if (term.startsWith("before:")) { const timestamp = term.slice(7); const unix = timestampToUnix(timestamp); if (unix !== 0) constructedFilters.push(`createdAt < ${unix}`); return null; } else if (term.startsWith("filter:following")) { // Check if we got a context user if (userCtx) { // Fetch user follows from DB const followedUsers = await Followings.find({ where: { followerId: userCtx.id, }, select: { followeeId: true, }, }); const followIDs = followedUsers.map( (user) => user.followeeId, ); if (followIDs.length === 0) return null; constructedFilters.push(`userId IN [${followIDs.join(",")}]`); } else { logger.warn( "search filtered to follows called without user context", ); } return null; } else if (term.startsWith("filter:followers")) { // Check if we got a context user if (userCtx) { // Fetch users follows from DB const followedUsers = await Followings.find({ where: { followeeId: userCtx.id, }, select: { followerId: true, }, }); const followIDs = followedUsers.map( (user) => user.followerId, ); if (followIDs.length === 0) return null; constructedFilters.push(`userId IN [${followIDs.join(",")}]`); } else { logger.warn( "search filtered to followers called without user context", ); } return null; } else if (term.startsWith("order:desc")) { sortRules.push("createdAt:desc"); return null; } else if (term.startsWith("order:asc")) { sortRules.push("createdAt:asc"); return null; } return term; }), ) ).filter((term) => term !== null); // An empty search term with defined filters means we have a placeholder search => https://www.meilisearch.com/docs/reference/api/search#placeholder-search // These have to be ordered manually, otherwise the *oldest* posts are returned first, which we don't want // If the user has defined a sort rule, don't mess with it if ( filteredSearchTerms.length === 0 && constructedFilters.length > 0 && sortRules.length === 0 ) { sortRules.push("createdAt:desc"); } // More than one sorting rule doesn't make sense. We only keep the first one, otherwise weird stuff may happen. if (sortRules.length > 1) { sortRules = [sortRules[0]]; } // An override sort takes precedence, user sorting is ignored here if (overrideSort) { sortRules = [overrideSort]; } logger.info(`Searching for ${filteredSearchTerms.join(" ")}`); logger.info(`Limit: ${limit}`); logger.info(`Offset: ${offset}`); logger.info(`Filters: ${constructedFilters}`); logger.info(`Ordering: ${sortRules}`); return posts.search(filteredSearchTerms.join(" "), { limit: limit, offset: offset, filter: constructedFilters, sort: sortRules, }); }, ingestNote: async (ingestNotes: Note | Note[]) => { if (ingestNotes instanceof Note) { ingestNotes = [ingestNotes]; } const indexingBatch: MeilisearchNote[] = []; for (const note of ingestNotes) { if (note.user === undefined) { note.user = await Users.findOne({ where: { id: note.userId, }, }); } let attachmentType = ""; if (note.attachedFileTypes.length > 0) { attachmentType = note.attachedFileTypes[0].split("/")[0]; switch (attachmentType) { case "image": case "video": case "audio": case "text": break; default: attachmentType = "file"; break; } } indexingBatch.push({ id: note.id.toString(), text: note.text ? note.text : "", userId: note.userId, userHost: note.userHost !== "" ? note.userHost : url.parse(config.host).host, channelId: note.channelId ? note.channelId : "", mediaAttachment: attachmentType, userName: note.user?.username ?? "UNKNOWN", createdAt: note.createdAt.getTime() / 1000, // division by 1000 is necessary because Node returns in ms-accuracy }); } return posts .addDocuments(indexingBatch, { primaryKey: "id", }) .then(() => logger.info(`sent ${indexingBatch.length} posts for indexing`), ); }, serverStats: async () => { const health: Health = await client.health(); const stats: Stats = await client.getStats(); return { health: health.status, size: stats.databaseSize, indexed_count: stats.indexes["posts"].numberOfDocuments, }; }, deleteNotes: async (note: Note | Note[] | string | string[]) => { if (note instanceof Note) { note = [note]; } if (typeof note === "string") { note = [note]; } const deletionBatch = note .map((n) => { if (n instanceof Note) { return n.id; } if (n.length > 0) return n; logger.error( `Failed to delete note from Meilisearch, invalid post ID: ${JSON.stringify( n, )}`, ); throw new Error( `Invalid note ID passed to meilisearch deleteNote: ${JSON.stringify( n, )}`, ); }) .filter((el) => el !== null); await posts.deleteDocuments(deletionBatch as string[]).then(() => { logger.info( `submitted ${deletionBatch.length} large batch for deletion`, ); }); }, } : null;