/* * Copyright 2023 jacqueline * * SPDX-License-Identifier: GPL-3.0-only */ #include "database.hpp" #include #include #include #include #include #include #include #include #include "esp_log.h" #include "ff.h" #include "freertos/projdefs.h" #include "index.hpp" #include "leveldb/cache.h" #include "leveldb/db.h" #include "leveldb/iterator.h" #include "leveldb/options.h" #include "leveldb/slice.h" #include "leveldb/write_batch.h" #include "env_esp.hpp" #include "file_gatherer.hpp" #include "records.hpp" #include "result.hpp" #include "tag_parser.hpp" #include "tasks.hpp" #include "track.hpp" namespace database { static SingletonEnv sEnv; static const char* kTag = "DB"; static const char kTrackIdKey[] = "next_track_id"; static std::atomic sIsDbOpen(false); static FileGathererImpl sFileGatherer; static TagParserImpl sTagParser; template auto IterateAndParse(leveldb::Iterator* it, std::size_t limit, Parser p) -> void { for (int i = 0; i < limit; i++) { if (!it->Valid()) { delete it; break; } std::invoke(p, it->key(), it->value()); it->Next(); } } auto Database::Open() -> cpp::result { return Open(&sFileGatherer, &sTagParser); } auto Database::Open(IFileGatherer* gatherer, ITagParser* parser) -> cpp::result { // TODO(jacqueline): Why isn't compare_and_exchange_* available? if (sIsDbOpen.exchange(true)) { return cpp::fail(DatabaseError::ALREADY_OPEN); } leveldb::sBackgroundThread.reset( tasks::Worker::Start()); std::shared_ptr worker( tasks::Worker::Start()); return worker ->Dispatch>( [=]() -> cpp::result { leveldb::DB* db; leveldb::Cache* cache = leveldb::NewLRUCache(24 * 1024); leveldb::Options options; options.env = sEnv.env(); options.create_if_missing = true; options.write_buffer_size = 48 * 1024; options.max_file_size = 32; options.block_cache = cache; options.block_size = 512; auto status = leveldb::DB::Open(options, "/.db", &db); if (!status.ok()) { delete cache; ESP_LOGE(kTag, "failed to open db, status %s", status.ToString().c_str()); return cpp::fail(FAILED_TO_OPEN); } ESP_LOGI(kTag, "Database opened successfully"); return new Database(db, cache, gatherer, parser, worker); }) .get(); } auto Database::Destroy() -> void { leveldb::Options options; options.env = sEnv.env(); leveldb::DestroyDB("/.db", options); } Database::Database(leveldb::DB* db, leveldb::Cache* cache, IFileGatherer* file_gatherer, ITagParser* tag_parser, std::shared_ptr worker) : db_(db), cache_(cache), worker_task_(worker), file_gatherer_(file_gatherer), tag_parser_(tag_parser) {} Database::~Database() { // Delete db_ first so that any outstanding background work finishes before // the background task is killed. delete db_; delete cache_; leveldb::sBackgroundThread.reset(); sIsDbOpen.store(false); } auto Database::Update() -> std::future { return worker_task_->Dispatch([&]() -> void { leveldb::ReadOptions read_options; read_options.fill_cache = false; // Stage 0: discard indexes // TODO(jacqueline): I think it should be possible to incrementally update // indexes, but my brain hurts. ESP_LOGI(kTag, "dropping stale indexes"); { leveldb::Iterator* it = db_->NewIterator(read_options); OwningSlice prefix = EncodeAllIndexesPrefix(); it->Seek(prefix.slice); while (it->Valid() && it->key().starts_with(prefix.slice)) { db_->Delete(leveldb::WriteOptions(), it->key()); it->Next(); } } // Stage 1: verify all existing tracks are still valid. ESP_LOGI(kTag, "verifying existing tracks"); { leveldb::Iterator* it = db_->NewIterator(read_options); OwningSlice prefix = EncodeDataPrefix(); it->Seek(prefix.slice); while (it->Valid() && it->key().starts_with(prefix.slice)) { std::optional track = ParseDataValue(it->value()); if (!track) { // The value was malformed. Drop this record. ESP_LOGW(kTag, "dropping malformed metadata"); db_->Delete(leveldb::WriteOptions(), it->key()); it->Next(); continue; } if (track->is_tombstoned()) { ESP_LOGW(kTag, "skipping tombstoned %lx", track->id()); it->Next(); continue; } TrackTags tags{}; if (!tag_parser_->ReadAndParseTags(track->filepath(), &tags) || tags.encoding() == Encoding::kUnsupported) { // We couldn't read the tags for this track. Either they were // malformed, or perhaps the file is missing. Either way, tombstone // this record. ESP_LOGW(kTag, "entombing missing #%lx", track->id()); dbPutTrackData(track->Entomb()); it->Next(); continue; } // At this point, we know that the track still exists in its original // location. All that's left to do is update any metadata about it. uint64_t new_hash = tags.Hash(); if (new_hash != track->tags_hash()) { // This track's tags have changed. Since the filepath is exactly the // same, we assume this is a legitimate correction. Update the // database. ESP_LOGI(kTag, "updating hash (%llx -> %llx)", track->tags_hash(), new_hash); dbPutTrackData(track->UpdateHash(new_hash)); dbPutHash(new_hash, track->id()); } dbCreateIndexesForTrack({*track, tags}); it->Next(); } delete it; } // Stage 2: search for newly added files. ESP_LOGI(kTag, "scanning for new tracks"); file_gatherer_->FindFiles("", [&](const std::string& path) { TrackTags tags; if (!tag_parser_->ReadAndParseTags(path, &tags) || tags.encoding() == Encoding::kUnsupported) { // No parseable tags; skip this fiile. return; } // Check for any existing record with the same hash. uint64_t hash = tags.Hash(); OwningSlice key = EncodeHashKey(hash); std::optional existing_hash; std::string raw_entry; if (db_->Get(leveldb::ReadOptions(), key.slice, &raw_entry).ok()) { existing_hash = ParseHashValue(raw_entry); } if (!existing_hash) { // We've never met this track before! Or we have, but the entry is // malformed. Either way, record this as a new track. TrackId id = dbMintNewTrackId(); ESP_LOGI(kTag, "recording new 0x%lx", id); TrackData data(id, path, hash); dbPutTrackData(data); dbPutHash(hash, id); dbCreateIndexesForTrack({data, tags}); return; } std::optional existing_data = dbGetTrackData(*existing_hash); if (!existing_data) { // We found a hash that matches, but there's no data record? Weird. TrackData new_data(*existing_hash, path, hash); dbPutTrackData(new_data); dbCreateIndexesForTrack({*existing_data, tags}); return; } if (existing_data->is_tombstoned()) { ESP_LOGI(kTag, "exhuming track %lu", existing_data->id()); dbPutTrackData(existing_data->Exhume(path)); dbCreateIndexesForTrack({*existing_data, tags}); } else if (existing_data->filepath() != path) { ESP_LOGW(kTag, "tag hash collision"); } }); }); } auto Database::GetTrackPath(TrackId id) -> std::future> { return worker_task_->Dispatch>( [=, this]() -> std::optional { auto track_data = dbGetTrackData(id); if (track_data) { return track_data->filepath(); } return {}; }); } auto Database::GetTrack(TrackId id) -> std::future> { return worker_task_->Dispatch>( [=, this]() -> std::optional { std::optional data = dbGetTrackData(id); if (!data || data->is_tombstoned()) { return {}; } TrackTags tags; if (!tag_parser_->ReadAndParseTags(data->filepath(), &tags)) { return {}; } return Track(*data, tags); }); } auto Database::GetBulkTracks(std::vector ids) -> std::future>> { return worker_task_->Dispatch>>( [=, this]() -> std::vector> { std::map id_to_track{}; // Sort the list of ids so that we can retrieve them all in a single // iteration through the database, without re-seeking. std::vector sorted_ids = ids; std::sort(sorted_ids.begin(), sorted_ids.end()); leveldb::Iterator* it = db_->NewIterator(leveldb::ReadOptions{}); for (const TrackId& id : sorted_ids) { OwningSlice key = EncodeDataKey(id); it->Seek(key.slice); if (!it->Valid() || it->key() != key.slice) { // This id wasn't found at all. Skip it. continue; } std::optional track = ParseRecord(it->key(), it->value()); if (track) { id_to_track.insert({id, *track}); } } // We've fetched all of the ids in the request, so now just put them // back into the order they were asked for in. std::vector> results; for (const TrackId& id : ids) { if (id_to_track.contains(id)) { results.push_back(id_to_track.at(id)); } else { // This lookup failed. results.push_back({}); } } return results; }); } auto Database::GetIndexes() -> std::vector { // TODO(jacqueline): This probably needs to be async? When we have runtime // configurable indexes, they will need to come from somewhere. return { kAllTracks, kAlbumsByArtist, kTracksByGenre, }; } auto Database::GetTracksByIndex(const IndexInfo& index, std::size_t page_size) -> std::future*> { return worker_task_->Dispatch*>( [=, this]() -> Result* { IndexKey::Header header{ .id = index.id, .depth = 0, .components_hash = 0, }; OwningSlice prefix = EncodeIndexPrefix(header); Continuation c{.iterator = nullptr, .prefix = prefix.data, .start_key = prefix.data, .forward = true, .was_prev_forward = true, .page_size = page_size}; return dbGetPage(c); }); } auto Database::GetTracks(std::size_t page_size) -> std::future*> { return worker_task_->Dispatch*>([=, this]() -> Result* { Continuation c{.iterator = nullptr, .prefix = EncodeDataPrefix().data, .start_key = EncodeDataPrefix().data, .forward = true, .was_prev_forward = true, .page_size = page_size}; return dbGetPage(c); }); } auto Database::GetDump(std::size_t page_size) -> std::future*> { return worker_task_->Dispatch*>( [=, this]() -> Result* { Continuation c{.iterator = nullptr, .prefix = "", .start_key = "", .forward = true, .was_prev_forward = true, .page_size = page_size}; return dbGetPage(c); }); } template auto Database::GetPage(Continuation* c) -> std::future*> { Continuation copy = *c; return worker_task_->Dispatch*>( [=, this]() -> Result* { return dbGetPage(copy); }); } template auto Database::GetPage(Continuation* c) -> std::future*>; template auto Database::GetPage(Continuation* c) -> std::future*>; template auto Database::GetPage(Continuation* c) -> std::future*>; auto Database::dbMintNewTrackId() -> TrackId { TrackId next_id = 1; std::string val; auto status = db_->Get(leveldb::ReadOptions(), kTrackIdKey, &val); if (status.ok()) { next_id = BytesToTrackId(val).value_or(next_id); } else if (!status.IsNotFound()) { // TODO(jacqueline): Handle this more. ESP_LOGE(kTag, "failed to get next track id"); } if (!db_->Put(leveldb::WriteOptions(), kTrackIdKey, TrackIdToBytes(next_id + 1).slice) .ok()) { ESP_LOGE(kTag, "failed to write next track id"); } return next_id; } auto Database::dbEntomb(TrackId id, uint64_t hash) -> void { OwningSlice key = EncodeHashKey(hash); OwningSlice val = EncodeHashValue(id); if (!db_->Put(leveldb::WriteOptions(), key.slice, val.slice).ok()) { ESP_LOGE(kTag, "failed to entomb #%llx (id #%lx)", hash, id); } } auto Database::dbPutTrackData(const TrackData& s) -> void { OwningSlice key = EncodeDataKey(s.id()); OwningSlice val = EncodeDataValue(s); if (!db_->Put(leveldb::WriteOptions(), key.slice, val.slice).ok()) { ESP_LOGE(kTag, "failed to write data for #%lx", s.id()); } } auto Database::dbGetTrackData(TrackId id) -> std::optional { OwningSlice key = EncodeDataKey(id); std::string raw_val; if (!db_->Get(leveldb::ReadOptions(), key.slice, &raw_val).ok()) { ESP_LOGW(kTag, "no key found for #%lx", id); return {}; } return ParseDataValue(raw_val); } auto Database::dbPutHash(const uint64_t& hash, TrackId i) -> void { OwningSlice key = EncodeHashKey(hash); OwningSlice val = EncodeHashValue(i); if (!db_->Put(leveldb::WriteOptions(), key.slice, val.slice).ok()) { ESP_LOGE(kTag, "failed to write hash for #%lx", i); } } auto Database::dbGetHash(const uint64_t& hash) -> std::optional { OwningSlice key = EncodeHashKey(hash); std::string raw_val; if (!db_->Get(leveldb::ReadOptions(), key.slice, &raw_val).ok()) { ESP_LOGW(kTag, "no key found for hash #%llx", hash); return {}; } return ParseHashValue(raw_val); } auto Database::dbCreateIndexesForTrack(Track track) -> void { for (const IndexInfo& index : GetIndexes()) { leveldb::WriteBatch writes; if (Index(index, track, &writes)) { db_->Write(leveldb::WriteOptions(), &writes); } } } template auto Database::dbGetPage(const Continuation& c) -> Result* { // Work out our starting point. Sometimes this will already done. leveldb::Iterator* it = nullptr; if (c.iterator != nullptr) { it = c.iterator->release(); } if (it == nullptr) { it = db_->NewIterator(leveldb::ReadOptions()); it->Seek(c.start_key); } // Fix off-by-one if we just changed direction. if (c.forward != c.was_prev_forward) { if (c.forward) { it->Next(); } else { it->Prev(); } } // Grab results. std::optional first_key; std::vector records; while (records.size() < c.page_size && it->Valid()) { if (!it->key().starts_with(c.prefix)) { break; } if (!first_key) { first_key = it->key().ToString(); } std::optional parsed = ParseRecord(it->key(), it->value()); if (parsed) { records.push_back(*parsed); } if (c.forward) { it->Next(); } else { it->Prev(); } } std::unique_ptr iterator(it); if (iterator != nullptr) { if (!iterator->Valid() || !it->key().starts_with(c.prefix)) { iterator.reset(); } } // Put results into canonical order if we were iterating backwards. if (!c.forward) { std::reverse(records.begin(), records.end()); } // Work out the new continuations. std::optional> next_page; if (c.forward) { if (iterator != nullptr) { // We were going forward, and now we want the next page. Re-use the // existing iterator, and point the start key at it. std::string key = iterator->key().ToString(); next_page = Continuation{ .iterator = std::make_shared>( std::move(iterator)), .prefix = c.prefix, .start_key = key, .forward = true, .was_prev_forward = true, .page_size = c.page_size, }; } // No iterator means we ran out of results in this direction. } else { // We were going backwards, and now we want the next page. This is a // reversal, to set the start key to the first record we saw and mark that // it's off by one. next_page = Continuation{ .iterator = nullptr, .prefix = c.prefix, .start_key = *first_key, .forward = true, .was_prev_forward = false, .page_size = c.page_size, }; } std::optional> prev_page; if (c.forward) { // We were going forwards, and now we want the previous page. Set the search // key to the first result we saw, and mark that it's off by one. prev_page = Continuation{ .iterator = nullptr, .prefix = c.prefix, .start_key = *first_key, .forward = false, .was_prev_forward = true, .page_size = c.page_size, }; } else { if (iterator != nullptr) { // We were going backwards, and we still want to go backwards. The // iterator is still valid. std::string key = iterator->key().ToString(); prev_page = Continuation{ .iterator = std::make_shared>( std::move(iterator)), .prefix = c.prefix, .start_key = key, .forward = false, .was_prev_forward = false, .page_size = c.page_size, }; } // No iterator means we ran out of results in this direction. } return new Result(std::move(records), next_page, prev_page); } template auto Database::dbGetPage(const Continuation& c) -> Result*; template auto Database::dbGetPage( const Continuation& c) -> Result*; template <> auto Database::ParseRecord(const leveldb::Slice& key, const leveldb::Slice& val) -> std::optional { std::optional data = ParseIndexKey(key); if (!data) { return {}; } std::optional title; if (!val.empty()) { title = val.ToString(); } return IndexRecord(*data, title, data->track); } template <> auto Database::ParseRecord(const leveldb::Slice& key, const leveldb::Slice& val) -> std::optional { std::optional data = ParseDataValue(val); if (!data || data->is_tombstoned()) { return {}; } TrackTags tags; if (!tag_parser_->ReadAndParseTags(data->filepath(), &tags)) { return {}; } return Track(*data, tags); } template <> auto Database::ParseRecord(const leveldb::Slice& key, const leveldb::Slice& val) -> std::optional { std::ostringstream stream; stream << "key: "; if (key.size() < 3 || key.data()[1] != '\0') { stream << key.ToString().c_str(); } else { std::string str = key.ToString(); for (size_t i = 0; i < str.size(); i++) { if (i == 0) { stream << str[i]; } else if (i == 1) { stream << " / 0x"; } else { stream << std::hex << std::setfill('0') << std::setw(2) << static_cast(str[i]); } } } if (!val.empty()) { stream << "\tval: 0x"; std::string str = val.ToString(); for (int i = 0; i < val.size(); i++) { stream << std::hex << std::setfill('0') << std::setw(2) << static_cast(str[i]); } } return stream.str(); } IndexRecord::IndexRecord(const IndexKey& key, std::optional title, std::optional track) : key_(key), override_text_(title), track_(track) {} auto IndexRecord::text() const -> std::optional { if (override_text_) { return override_text_; } return key_.item; } auto IndexRecord::track() const -> std::optional { return track_; } auto IndexRecord::Expand(std::size_t page_size) const -> std::optional> { if (track_) { return {}; } IndexKey::Header new_header = ExpandHeader(key_.header, key_.item); OwningSlice new_prefix = EncodeIndexPrefix(new_header); return Continuation{ .iterator = nullptr, .prefix = new_prefix.data, .start_key = new_prefix.data, .forward = true, .was_prev_forward = true, .page_size = page_size, }; } } // namespace database