// SPDX-License-Identifier: GPL-3.0-or-later /** @file db_api.c * @brief This is the file implementing the API to the * logs management database. */ #include "daemon/common.h" #include "db_api.h" #include #include #include "circular_buffer.h" #include "helper.h" #include "lz4.h" #include "parser.h" #define MAIN_DB "main.db" /**< Primary DB with metadata for all the logs managemt collections **/ #define MAIN_COLLECTIONS_TABLE "LogCollections" /*< Table name where logs collections metadata is stored in MAIN_DB **/ #define BLOB_STORE_FILENAME "logs.bin." /*< Filename of BLOBs where logs are stored in **/ #define METADATA_DB_FILENAME "metadata.db" /**< Metadata DB for each log collection **/ #define LOGS_TABLE "Logs" /*< Table name where logs metadata is stored in METADATA_DB_FILENAME **/ #define BLOBS_TABLE "Blobs" /*< Table name where BLOBs metadata is stored in METADATA_DB_FILENAME **/ #define LOGS_MANAG_DB_VERSION 1 static sqlite3 *main_db = NULL; /**< SQLite DB handler for MAIN_DB **/ static char *main_db_dir = NULL; /**< Directory where all the log management databases and log blobs are stored in **/ static char *main_db_path = NULL; /**< Path of MAIN_DB **/ /* -------------------------------------------------------------------------- */ /* Database migrations */ /* -------------------------------------------------------------------------- */ /** * @brief No-op database migration, just to bump up starting version. * @param database Unused * @param name Unused * @return Always 0. */ static int do_migration_noop(sqlite3 *database, const char *name){ UNUSED(database); UNUSED(name); collector_info("Running database migration %s", name); return 0; } typedef struct database_func_migration_list{ char *name; int (*func)(sqlite3 *database, const char *name); } DATABASE_FUNC_MIGRATION_LIST; DATABASE_FUNC_MIGRATION_LIST migration_list_main_db[] = { {.name = MAIN_DB" v0 to v1", .func = do_migration_noop}, // the terminator of this array {.name = NULL, .func = NULL} }; DATABASE_FUNC_MIGRATION_LIST migration_list_metadata_db[] = { {.name = METADATA_DB_FILENAME " v0 to v1", .func = do_migration_noop}, // the terminator of this array {.name = NULL, .func = NULL} }; typedef enum { ERR_TYPE_OTHER, ERR_TYPE_SQLITE, ERR_TYPE_LIBUV, } logs_manag_db_error_t; /** * @brief Logs a database error * @param[in] log_source Log source that caused the error * @param[in] error_type Type of error * @param[in] rc Error code * @param[in] line Line number where the error occurred (__LINE__) * @param[in] file Source file where the error occurred (__FILE__) * @param[in] func Function where the error occurred (__FUNCTION__) */ static void throw_error(const char *const log_source, const logs_manag_db_error_t error_type, const int rc, const int line, const char *const file, const char *const func){ collector_error("[%s]: %s database error: (%d) %s (%s:%s:%d))", log_source ? log_source : "-", error_type == ERR_TYPE_OTHER ? "" : ERR_TYPE_SQLITE ? "SQLite" : "libuv", rc, error_type == ERR_TYPE_OTHER ? "" : ERR_TYPE_SQLITE ? sqlite3_errstr(rc) : uv_strerror(rc), file, func, line); } /** * @brief Get or set user_version of database. * @param db SQLite database to act upon. * @param set_user_version If <= 0, just get user_version. Otherwise, set * user_version first, before returning it. * @return Database user_version or -1 in case of error. */ int db_user_version(sqlite3 *const db, const int set_user_version){ if(unlikely(!db)) return -1; int rc = 0; if(set_user_version <= 0){ sqlite3_stmt *stmt_get_user_version; rc = sqlite3_prepare_v2(db, "PRAGMA user_version;", -1, &stmt_get_user_version, NULL); if (unlikely(SQLITE_OK != rc)) { throw_error(NULL, ERR_TYPE_SQLITE, rc, __LINE__, __FILE__, __FUNCTION__); return -1; } rc = sqlite3_step(stmt_get_user_version); if (unlikely(SQLITE_ROW != rc)) { throw_error(NULL, ERR_TYPE_SQLITE, rc, __LINE__, __FILE__, __FUNCTION__); return -1; } int current_user_version = sqlite3_column_int(stmt_get_user_version, 0); rc = sqlite3_finalize(stmt_get_user_version); if (unlikely(SQLITE_OK != rc)) { throw_error(NULL, ERR_TYPE_SQLITE, rc, __LINE__, __FILE__, __FUNCTION__); return -1; } return current_user_version; } else { char buf[25]; snprintfz(buf, 25, "PRAGMA user_version=%d;", set_user_version); rc = sqlite3_exec(db, buf, NULL, NULL, NULL); if (unlikely(SQLITE_OK!= rc)) { throw_error(NULL, ERR_TYPE_SQLITE, rc, __LINE__, __FILE__, __FUNCTION__); return -1; } return set_user_version; } } static void db_writer_db_mode_none(void *arg){ struct File_info *const p_file_info = (struct File_info *) arg; Circ_buff_item_t *item; while(__atomic_load_n(&p_file_info->state, __ATOMIC_RELAXED) == LOG_SRC_READY){ uv_rwlock_rdlock(&p_file_info->circ_buff->buff_realloc_rwlock); do{ item = circ_buff_read_item(p_file_info->circ_buff);} while(item); circ_buff_read_done(p_file_info->circ_buff); uv_rwlock_rdunlock(&p_file_info->circ_buff->buff_realloc_rwlock); for(int i = 0; i < p_file_info->buff_flush_to_db_interval * 4; i++){ if(__atomic_load_n(&p_file_info->state, __ATOMIC_RELAXED) != LOG_SRC_READY) break; sleep_usec(250 * USEC_PER_MS); } } } #define return_db_writer_db_mode_none(p_file_info, do_mut_unlock){ \ p_file_info->db_mode = LOGS_MANAG_DB_MODE_NONE; \ freez((void *) p_file_info->db_dir); \ p_file_info->db_dir = strdupz(""); \ freez((void *) p_file_info->db_metadata); \ p_file_info->db_metadata = NULL; \ sqlite3_finalize(stmt_logs_insert); \ sqlite3_finalize(stmt_blobs_get_total_filesize); \ sqlite3_finalize(stmt_blobs_update); \ sqlite3_finalize(stmt_blobs_set_zero_filesize); \ sqlite3_finalize(stmt_logs_delete); \ if(do_mut_unlock){ \ uv_mutex_unlock(p_file_info->db_mut); \ uv_rwlock_rdunlock(&p_file_info->circ_buff->buff_realloc_rwlock); \ } \ if(__atomic_load_n(&p_file_info->state, __ATOMIC_RELAXED) == LOG_SRC_READY) \ return fatal_assert(!uv_thread_create( p_file_info->db_writer_thread, \ db_writer_db_mode_none, \ p_file_info)); \ } static void db_writer_db_mode_full(void *arg){ int rc = 0; struct File_info *const p_file_info = (struct File_info *) arg; sqlite3_stmt *stmt_logs_insert = NULL; sqlite3_stmt *stmt_blobs_get_total_filesize = NULL; sqlite3_stmt *stmt_blobs_update = NULL; sqlite3_stmt *stmt_blobs_set_zero_filesize = NULL; sqlite3_stmt *stmt_logs_delete = NULL; /* Prepare LOGS_TABLE INSERT statement */ rc = sqlite3_prepare_v2(p_file_info->db, "INSERT INTO " LOGS_TABLE "(" "FK_BLOB_Id," "BLOB_Offset," "Timestamp," "Msg_compr_size," "Msg_decompr_size," "Num_lines" ") VALUES (?,?,?,?,?,?) ;", -1, &stmt_logs_insert, NULL); if (unlikely(SQLITE_OK != rc)) { throw_error(p_file_info->chartname, ERR_TYPE_SQLITE, rc, __LINE__, __FILE__, __FUNCTION__); return_db_writer_db_mode_none(p_file_info, 0); } /* Prepare BLOBS_TABLE get total filesize statement */ rc = sqlite3_prepare_v2(p_file_info->db, "SELECT SUM(Filesize) FROM " BLOBS_TABLE " ;", -1, &stmt_blobs_get_total_filesize, NULL); if (unlikely(SQLITE_OK != rc)) { throw_error(p_file_info->chartname, ERR_TYPE_SQLITE, rc, __LINE__, __FILE__, __FUNCTION__); return_db_writer_db_mode_none(p_file_info, 0); } /* Prepare BLOBS_TABLE UPDATE statement */ rc = sqlite3_prepare_v2(p_file_info->db, "UPDATE " BLOBS_TABLE " SET Filesize = Filesize + ?" " WHERE Id = ? ;", -1, &stmt_blobs_update, NULL); if (unlikely(SQLITE_OK != rc)) { throw_error(p_file_info->chartname, ERR_TYPE_SQLITE, rc, __LINE__, __FILE__, __FUNCTION__); return_db_writer_db_mode_none(p_file_info, 0); } /* Prepare BLOBS_TABLE UPDATE SET zero filesize statement */ rc = sqlite3_prepare_v2(p_file_info->db, "UPDATE " BLOBS_TABLE " SET Filesize = 0" " WHERE Id = ? ;", -1, &stmt_blobs_set_zero_filesize, NULL); if (unlikely(SQLITE_OK != rc)) { throw_error(p_file_info->chartname, ERR_TYPE_SQLITE, rc, __LINE__, __FILE__, __FUNCTION__); return_db_writer_db_mode_none(p_file_info, 0); } /* Prepare LOGS_TABLE DELETE statement */ rc = sqlite3_prepare_v2(p_file_info->db, "DELETE FROM " LOGS_TABLE " WHERE FK_BLOB_Id = ? ;", -1, &stmt_logs_delete, NULL); if (unlikely(SQLITE_OK != rc)) { throw_error(p_file_info->chartname, ERR_TYPE_SQLITE, rc, __LINE__, __FILE__, __FUNCTION__); return_db_writer_db_mode_none(p_file_info, 0); } /* Get initial filesize of logs.bin.0 BLOB */ sqlite3_stmt *stmt_retrieve_filesize_from_id = NULL; if(unlikely( SQLITE_OK != (rc = sqlite3_prepare_v2(p_file_info->db, "SELECT Filesize FROM " BLOBS_TABLE " WHERE Id = ? ;", -1, &stmt_retrieve_filesize_from_id, NULL)) || SQLITE_OK != (rc = sqlite3_bind_int(stmt_retrieve_filesize_from_id, 1, p_file_info->blob_write_handle_offset)) || SQLITE_ROW != (rc = sqlite3_step(stmt_retrieve_filesize_from_id)) )){ throw_error(p_file_info->chartname, ERR_TYPE_SQLITE, rc, __LINE__, __FILE__, __FUNCTION__); return_db_writer_db_mode_none(p_file_info, 0); } int64_t blob_filesize = (int64_t) sqlite3_column_int64(stmt_retrieve_filesize_from_id, 0); rc = sqlite3_finalize(stmt_retrieve_filesize_from_id); if (unlikely(SQLITE_OK != rc)) { throw_error(p_file_info->chartname, ERR_TYPE_SQLITE, rc, __LINE__, __FILE__, __FUNCTION__); return_db_writer_db_mode_none(p_file_info, 0); } struct timespec ts_db_write_start, ts_db_write_end, ts_db_rotate_end; while(__atomic_load_n(&p_file_info->state, __ATOMIC_RELAXED) == LOG_SRC_READY){ clock_gettime(CLOCK_THREAD_CPUTIME_ID, &ts_db_write_start); uv_rwlock_rdlock(&p_file_info->circ_buff->buff_realloc_rwlock); uv_mutex_lock(p_file_info->db_mut); /* --------------------------------------------------------------------- * Read items from circular buffer and store them in disk BLOBs. * After that, SQLite metadata is updated. * ------------------------------------------------------------------ */ Circ_buff_item_t *item = circ_buff_read_item(p_file_info->circ_buff); while (item) { m_assert(TEST_MS_TIMESTAMP_VALID(item->timestamp), "item->timestamp == 0"); m_assert(item->text_compressed_size != 0, "item->text_compressed_size == 0"); m_assert(item->text_size != 0, "item->text_size == 0"); /* Write logs in BLOB */ uv_fs_t write_req; uv_buf_t uv_buf = uv_buf_init((char *) item->text_compressed, (unsigned int) item->text_compressed_size); rc = uv_fs_write( NULL, &write_req, p_file_info->blob_handles[p_file_info->blob_write_handle_offset], &uv_buf, 1, blob_filesize, NULL); // Write synchronously at the end of the BLOB file uv_fs_req_cleanup(&write_req); if(unlikely(rc < 0)){ throw_error(p_file_info->chartname, ERR_TYPE_LIBUV, rc, __LINE__, __FILE__, __FUNCTION__); circ_buff_read_done(p_file_info->circ_buff); return_db_writer_db_mode_none(p_file_info, 1); } /* Ensure data is flushed to BLOB via fdatasync() */ uv_fs_t dsync_req; rc = uv_fs_fdatasync( NULL, &dsync_req, p_file_info->blob_handles[p_file_info->blob_write_handle_offset], NULL); uv_fs_req_cleanup(&dsync_req); if (unlikely(rc)){ throw_error(p_file_info->chartname, ERR_TYPE_LIBUV, rc, __LINE__, __FILE__, __FUNCTION__); circ_buff_read_done(p_file_info->circ_buff); return_db_writer_db_mode_none(p_file_info, 1); } if(unlikely( /* Write metadata of logs in LOGS_TABLE */ SQLITE_OK != (rc = sqlite3_exec(p_file_info->db, "BEGIN TRANSACTION;", NULL, NULL, NULL)) || SQLITE_OK != (rc = sqlite3_bind_int(stmt_logs_insert, 1, p_file_info->blob_write_handle_offset)) || SQLITE_OK != (rc = sqlite3_bind_int64(stmt_logs_insert, 2, (sqlite3_int64) blob_filesize)) || SQLITE_OK != (rc = sqlite3_bind_int64(stmt_logs_insert, 3, (sqlite3_int64) item->timestamp)) || SQLITE_OK != (rc = sqlite3_bind_int64(stmt_logs_insert, 4, (sqlite3_int64) item->text_compressed_size)) || SQLITE_OK != (rc = sqlite3_bind_int64(stmt_logs_insert, 5, (sqlite3_int64)item->text_size)) || SQLITE_OK != (rc = sqlite3_bind_int64(stmt_logs_insert, 6, (sqlite3_int64)item->num_lines)) || SQLITE_DONE != (rc = sqlite3_step(stmt_logs_insert)) || SQLITE_OK != (rc = sqlite3_reset(stmt_logs_insert)) || /* Update metadata of BLOBs filesize in BLOBS_TABLE */ SQLITE_OK != (rc = sqlite3_bind_int64(stmt_blobs_update, 1, (sqlite3_int64)item->text_compressed_size)) || SQLITE_OK != (rc = sqlite3_bind_int(stmt_blobs_update, 2, p_file_info->blob_write_handle_offset)) || SQLITE_DONE != (rc = sqlite3_step(stmt_blobs_update)) || SQLITE_OK != (rc = sqlite3_reset(stmt_blobs_update)) || SQLITE_OK != (rc = sqlite3_exec(p_file_info->db, "END TRANSACTION;", NULL, NULL, NULL)) )) { throw_error(p_file_info->chartname, ERR_TYPE_SQLITE, rc, __LINE__, __FILE__, __FUNCTION__); rc = sqlite3_exec(p_file_info->db, "ROLLBACK;", NULL, NULL, NULL); if (unlikely(SQLITE_OK != rc)) throw_error(p_file_info->chartname, ERR_TYPE_SQLITE, rc, __LINE__, __FILE__, __FUNCTION__); circ_buff_read_done(p_file_info->circ_buff); return_db_writer_db_mode_none(p_file_info, 1); } /* TODO: Should we log it if there is a fatal error in the transaction, * as there will be a mismatch between BLOBs and SQLite metadata? */ /* Increase BLOB offset and read next log message until no more messages in buff */ blob_filesize += (int64_t) item->text_compressed_size; item = circ_buff_read_item(p_file_info->circ_buff); } circ_buff_read_done(p_file_info->circ_buff); clock_gettime(CLOCK_THREAD_CPUTIME_ID, &ts_db_write_end); /* --------------------------------------------------------------------- * If the filesize of the current write-to BLOB is > * p_file_info->blob_max_size, then perform a BLOBs rotation. * ------------------------------------------------------------------ */ if(blob_filesize > p_file_info->blob_max_size){ uv_fs_t rename_req; char old_path[FILENAME_MAX + 1], new_path[FILENAME_MAX + 1]; /* Rotate path of BLOBs */ for(int i = BLOB_MAX_FILES - 1; i >= 0; i--){ snprintfz(old_path, FILENAME_MAX, "%s" BLOB_STORE_FILENAME "%d", p_file_info->db_dir, i); snprintfz(new_path, FILENAME_MAX, "%s" BLOB_STORE_FILENAME "%d", p_file_info->db_dir, i + 1); rc = uv_fs_rename(NULL, &rename_req, old_path, new_path, NULL); uv_fs_req_cleanup(&rename_req); if (unlikely(rc)){ //TODO: This error case needs better handling, as it will result in mismatch with sqlite metadata. // We probably require a WAL or something similar. throw_error(p_file_info->chartname, ERR_TYPE_LIBUV, rc, __LINE__, __FILE__, __FUNCTION__); return_db_writer_db_mode_none(p_file_info, 1); } } /* Replace the maximum number with 0 in BLOB files. */ snprintfz(old_path, FILENAME_MAX, "%s" BLOB_STORE_FILENAME "%d", p_file_info->db_dir, BLOB_MAX_FILES); snprintfz(new_path, FILENAME_MAX, "%s" BLOB_STORE_FILENAME "%d", p_file_info->db_dir, 0); rc = uv_fs_rename(NULL, &rename_req, old_path, new_path, NULL); uv_fs_req_cleanup(&rename_req); if (unlikely(rc)){ //TODO: This error case needs better handling, as it will result in mismatch with sqlite metadata. // We probably require a WAL or something similar. throw_error(p_file_info->chartname, ERR_TYPE_LIBUV, rc, __LINE__, __FILE__, __FUNCTION__); return_db_writer_db_mode_none(p_file_info, 1); } /* Rotate BLOBS_TABLE Filenames */ rc = sqlite3_exec(p_file_info->db, "UPDATE " BLOBS_TABLE " SET Filename = REPLACE( " " Filename, " " substr(Filename, -1), " " case when " " (cast(substr(Filename, -1) AS INTEGER) < (" LOGS_MANAG_STR(BLOB_MAX_FILES) " - 1)) then " " substr(Filename, -1) + 1 else 0 end);", NULL, NULL, NULL); if (unlikely(rc != SQLITE_OK)) { throw_error(p_file_info->chartname, ERR_TYPE_SQLITE, rc, __LINE__, __FILE__, __FUNCTION__); //TODO: Undo rotation if possible? return_db_writer_db_mode_none(p_file_info, 1); } /* ----------------------------------------------------------------- * (a) Update blob_write_handle_offset, * (b) truncate new write-to BLOB, * (c) update filesize of truncated BLOB in SQLite DB, * (d) delete respective logs in LOGS_TABLE for the truncated BLOB and * (e) reset blob_filesize * -------------------------------------------------------------- */ /* (a) */ p_file_info->blob_write_handle_offset = p_file_info->blob_write_handle_offset == 1 ? BLOB_MAX_FILES : p_file_info->blob_write_handle_offset - 1; /* (b) */ uv_fs_t trunc_req; rc = uv_fs_ftruncate(NULL, &trunc_req, p_file_info->blob_handles[p_file_info->blob_write_handle_offset], 0, NULL); uv_fs_req_cleanup(&trunc_req); if (unlikely(rc)){ //TODO: This error case needs better handling, as it will result in mismatch with sqlite metadata. // We probably require a WAL or something similar. throw_error(p_file_info->chartname, ERR_TYPE_LIBUV, rc, __LINE__, __FILE__, __FUNCTION__); return_db_writer_db_mode_none(p_file_info, 1); } /* (c) */ if(unlikely( SQLITE_OK != (rc = sqlite3_exec(p_file_info->db, "BEGIN TRANSACTION;", NULL, NULL, NULL)) || SQLITE_OK != (rc = sqlite3_bind_int(stmt_blobs_set_zero_filesize, 1, p_file_info->blob_write_handle_offset)) || SQLITE_DONE != (rc = sqlite3_step(stmt_blobs_set_zero_filesize)) || SQLITE_OK != (rc = sqlite3_reset(stmt_blobs_set_zero_filesize)) || /* (d) */ SQLITE_OK != (rc = sqlite3_bind_int(stmt_logs_delete, 1, p_file_info->blob_write_handle_offset)) || SQLITE_DONE != (rc = sqlite3_step(stmt_logs_delete)) || SQLITE_OK != (rc = sqlite3_reset(stmt_logs_delete)) || SQLITE_OK != (rc = sqlite3_exec(p_file_info->db, "END TRANSACTION;", NULL, NULL, NULL)) )) { throw_error(p_file_info->chartname, ERR_TYPE_SQLITE, rc, __LINE__, __FILE__, __FUNCTION__); rc = sqlite3_exec(p_file_info->db, "ROLLBACK;", NULL, NULL, NULL); if (unlikely(SQLITE_OK != rc)) throw_error(p_file_info->chartname, ERR_TYPE_SQLITE, rc, __LINE__, __FILE__, __FUNCTION__); return_db_writer_db_mode_none(p_file_info, 1); } /* (e) */ blob_filesize = 0; } clock_gettime(CLOCK_THREAD_CPUTIME_ID, &ts_db_rotate_end); /* Update database write & rotate timings for this log source */ __atomic_store_n(&p_file_info->db_write_duration, (ts_db_write_end.tv_sec - ts_db_write_start.tv_sec) * NSEC_PER_SEC + (ts_db_write_end.tv_nsec - ts_db_write_start.tv_nsec), __ATOMIC_RELAXED); __atomic_store_n(&p_file_info->db_rotate_duration, (ts_db_rotate_end.tv_sec - ts_db_write_end.tv_sec) * NSEC_PER_SEC + (ts_db_rotate_end.tv_nsec - ts_db_write_end.tv_nsec), __ATOMIC_RELAXED); /* Update total disk usage of all BLOBs for this log source */ rc = sqlite3_step(stmt_blobs_get_total_filesize); if (unlikely(SQLITE_ROW != rc)) { throw_error(p_file_info->chartname, ERR_TYPE_SQLITE, rc, __LINE__, __FILE__, __FUNCTION__); return_db_writer_db_mode_none(p_file_info, 1); } __atomic_store_n(&p_file_info->blob_total_size, sqlite3_column_int64(stmt_blobs_get_total_filesize, 0), __ATOMIC_RELAXED); rc = sqlite3_reset(stmt_blobs_get_total_filesize); if (unlikely(SQLITE_OK != rc)) { throw_error(p_file_info->chartname, ERR_TYPE_SQLITE, rc, __LINE__, __FILE__, __FUNCTION__); return_db_writer_db_mode_none(p_file_info, 1); } // TODO: Can uv_mutex_unlock(p_file_info->db_mut) be moved before if(blob_filesize > p_file_info-> blob_max_size) ? uv_mutex_unlock(p_file_info->db_mut); uv_rwlock_rdunlock(&p_file_info->circ_buff->buff_realloc_rwlock); for(int i = 0; i < p_file_info->buff_flush_to_db_interval * 4; i++){ if(__atomic_load_n(&p_file_info->state, __ATOMIC_RELAXED) != LOG_SRC_READY) break; sleep_usec(250 * USEC_PER_MS); } } return_db_writer_db_mode_none(p_file_info, 0); } inline void db_set_main_dir(char *const dir){ main_db_dir = dir; } int db_init() { int rc = 0; char *err_msg = 0; uv_fs_t mkdir_req; if(unlikely(!main_db_dir || !*main_db_dir)){ rc = -1; collector_error("main_db_dir is unset"); throw_error(NULL, ERR_TYPE_OTHER, rc, __LINE__, __FILE__, __FUNCTION__); goto return_error; } size_t main_db_path_len = strlen(main_db_dir) + sizeof(MAIN_DB) + 1; main_db_path = mallocz(main_db_path_len); snprintfz(main_db_path, main_db_path_len, "%s/" MAIN_DB, main_db_dir); /* Create databases directory if it doesn't exist. */ rc = uv_fs_mkdir(NULL, &mkdir_req, main_db_dir, 0775, NULL); uv_fs_req_cleanup(&mkdir_req); if(rc == 0) collector_info("DB directory created: %s", main_db_dir); else if (rc == UV_EEXIST) collector_info("DB directory %s found", main_db_dir); else { throw_error(NULL, ERR_TYPE_LIBUV, rc, __LINE__, __FILE__, __FUNCTION__); goto return_error; } /* Create or open main db */ rc = sqlite3_open(main_db_path, &main_db); if (unlikely(rc != SQLITE_OK)){ throw_error(MAIN_DB, ERR_TYPE_SQLITE, rc, __LINE__, __FILE__, __FUNCTION__); goto return_error; } /* Configure main database */ rc = sqlite3_exec(main_db, "PRAGMA auto_vacuum = INCREMENTAL;" "PRAGMA synchronous = 1;" "PRAGMA journal_mode = WAL;" "PRAGMA temp_store = MEMORY;" "PRAGMA foreign_keys = ON;", 0, 0, &err_msg); if (unlikely(rc != SQLITE_OK)) { collector_error("Failed to configure database, SQL error: %s\n", err_msg); throw_error(MAIN_DB, ERR_TYPE_SQLITE, rc, __LINE__, __FILE__, __FUNCTION__); goto return_error; } else collector_info("%s configured successfully", MAIN_DB); /* Execute pending main database migrations */ int main_db_ver = db_user_version(main_db, -1); if (likely(LOGS_MANAG_DB_VERSION == main_db_ver)) collector_info("Logs management %s database version is %d (no migration needed)", MAIN_DB, main_db_ver); else { for(int ver = main_db_ver; ver < LOGS_MANAG_DB_VERSION && migration_list_main_db[ver].func; ver++){ rc = (migration_list_main_db[ver].func)(main_db, migration_list_main_db[ver].name); if (unlikely(rc)){ collector_error("Logs management %s database migration from version %d to version %d failed", MAIN_DB, ver, ver + 1); throw_error(MAIN_DB, ERR_TYPE_SQLITE, rc, __LINE__, __FILE__, __FUNCTION__); goto return_error; } db_user_version(main_db, ver + 1); } } /* Create new main DB LogCollections table if it doesn't exist */ rc = sqlite3_exec(main_db, "CREATE TABLE IF NOT EXISTS " MAIN_COLLECTIONS_TABLE "(" "Id INTEGER PRIMARY KEY," "Stream_Tag TEXT NOT NULL," "Log_Source_Path TEXT NOT NULL," "Type INTEGER NOT NULL," "DB_Dir TEXT NOT NULL," "UNIQUE(Stream_Tag, DB_Dir) " ");", 0, 0, &err_msg); if (unlikely(SQLITE_OK != rc)) { collector_error("Failed to create table" MAIN_COLLECTIONS_TABLE "SQL error: %s", err_msg); throw_error(MAIN_DB, ERR_TYPE_SQLITE, rc, __LINE__, __FILE__, __FUNCTION__); goto return_error; } sqlite3_stmt *stmt_search_if_log_source_exists = NULL; rc = sqlite3_prepare_v2(main_db, "SELECT COUNT(*), Id, DB_Dir FROM " MAIN_COLLECTIONS_TABLE " WHERE Stream_Tag = ? AND Log_Source_Path = ? AND Type = ? ;", -1, &stmt_search_if_log_source_exists, NULL); if (unlikely(SQLITE_OK != rc)){ throw_error(MAIN_DB, ERR_TYPE_SQLITE, rc, __LINE__, __FILE__, __FUNCTION__); goto return_error; } sqlite3_stmt *stmt_insert_log_collection_metadata = NULL; rc = sqlite3_prepare_v2(main_db, "INSERT INTO " MAIN_COLLECTIONS_TABLE " (Stream_Tag, Log_Source_Path, Type, DB_Dir) VALUES (?,?,?,?) ;", -1, &stmt_insert_log_collection_metadata, NULL); if (unlikely(SQLITE_OK != rc)){ throw_error(MAIN_DB, ERR_TYPE_SQLITE, rc, __LINE__, __FILE__, __FUNCTION__); goto return_error; } for (int i = 0; i < p_file_infos_arr->count; i++) { struct File_info *const p_file_info = p_file_infos_arr->data[i]; if(p_file_info->db_mode == LOGS_MANAG_DB_MODE_NONE){ p_file_info->db_dir = strdupz(""); p_file_info->db_writer_thread = mallocz(sizeof(uv_thread_t)); rc = uv_thread_create(p_file_info->db_writer_thread, db_writer_db_mode_none, p_file_info); if (unlikely(rc)){ throw_error(p_file_info->chartname, ERR_TYPE_LIBUV, rc, __LINE__, __FILE__, __FUNCTION__); goto return_error; } } else if(p_file_info->db_mode == LOGS_MANAG_DB_MODE_FULL){ p_file_info->db_mut = mallocz(sizeof(uv_mutex_t)); rc = uv_mutex_init(p_file_info->db_mut); if (unlikely(rc)) fatal("Failed to initialize uv_mutex_t"); uv_mutex_lock(p_file_info->db_mut); // This error check will be used a lot, so define it here. #define do_sqlite_error_check(p_file_info, rc, rc_expctd) do { \ if(unlikely(rc_expctd != rc)) { \ throw_error(p_file_info->chartname, ERR_TYPE_SQLITE, rc, __LINE__, __FILE__, __FUNCTION__);\ uv_mutex_unlock(p_file_info->db_mut); \ goto return_error; \ } \ } while(0) if(unlikely( SQLITE_OK != (rc = sqlite3_bind_text(stmt_search_if_log_source_exists, 1, p_file_info->stream_guid, -1, NULL)) || SQLITE_OK != (rc = sqlite3_bind_text(stmt_search_if_log_source_exists, 2, p_file_info->filename, -1, NULL)) || SQLITE_OK != (rc = sqlite3_bind_int(stmt_search_if_log_source_exists, 3, p_file_info->log_type)) || /* COUNT(*) query should always return SQLITE_ROW */ SQLITE_ROW != (rc = sqlite3_step(stmt_search_if_log_source_exists)))){ throw_error(p_file_info->chartname, ERR_TYPE_SQLITE, rc, __LINE__, __FILE__, __FUNCTION__); uv_mutex_unlock(p_file_info->db_mut); goto return_error; } const int log_source_occurences = sqlite3_column_int(stmt_search_if_log_source_exists, 0); switch (log_source_occurences) { case 0: { /* Log collection metadata not found in main DB - create a new record */ /* Create directory of collection of logs for the particular * log source (in the form of a UUID) and bind it. */ uuid_t uuid; uuid_generate(uuid); char uuid_str[UUID_STR_LEN]; // ex. "1b4e28ba-2fa1-11d2-883f-0016d3cca427" + "\0" uuid_unparse_lower(uuid, uuid_str); p_file_info->db_dir = mallocz(snprintf(NULL, 0, "%s/%s/", main_db_dir, uuid_str) + 1); sprintf((char *) p_file_info->db_dir, "%s/%s/", main_db_dir, uuid_str); rc = uv_fs_mkdir(NULL, &mkdir_req, p_file_info->db_dir, 0775, NULL); uv_fs_req_cleanup(&mkdir_req); if (unlikely(rc)) { if(errno == EEXIST) collector_error("DB directory %s exists but not found in %s.\n", p_file_info->db_dir, MAIN_DB); throw_error(p_file_info->chartname, ERR_TYPE_LIBUV, rc, __LINE__, __FILE__, __FUNCTION__); uv_mutex_unlock(p_file_info->db_mut); goto return_error; } if(unlikely( SQLITE_OK != (rc = sqlite3_bind_text(stmt_insert_log_collection_metadata, 1, p_file_info->stream_guid, -1, NULL)) || SQLITE_OK != (rc = sqlite3_bind_text(stmt_insert_log_collection_metadata, 2, p_file_info->filename, -1, NULL)) || SQLITE_OK != (rc = sqlite3_bind_int(stmt_insert_log_collection_metadata, 3, p_file_info->log_type)) || SQLITE_OK != (rc = sqlite3_bind_text(stmt_insert_log_collection_metadata, 4, p_file_info->db_dir, -1, NULL)) || SQLITE_DONE != (rc = sqlite3_step(stmt_insert_log_collection_metadata)) || SQLITE_OK != (rc = sqlite3_reset(stmt_insert_log_collection_metadata)))) { throw_error(p_file_info->chartname, ERR_TYPE_SQLITE, rc, __LINE__, __FILE__, __FUNCTION__); uv_mutex_unlock(p_file_info->db_mut); goto return_error; } break; } case 1: { /* File metadata found in DB */ p_file_info->db_dir = mallocz((size_t)sqlite3_column_bytes(stmt_search_if_log_source_exists, 2) + 1); sprintf((char*) p_file_info->db_dir, "%s", sqlite3_column_text(stmt_search_if_log_source_exists, 2)); break; } default: { /* Error, file metadata can exist either 0 or 1 times in DB */ m_assert(0, "Same file stored in DB more than once!"); collector_error("[%s]: Record encountered multiple times in DB " MAIN_COLLECTIONS_TABLE " table \n", p_file_info->filename); throw_error(p_file_info->chartname, ERR_TYPE_OTHER, rc, __LINE__, __FILE__, __FUNCTION__); uv_mutex_unlock(p_file_info->db_mut); goto return_error; } } rc = sqlite3_reset(stmt_search_if_log_source_exists); do_sqlite_error_check(p_file_info, rc, SQLITE_OK); /* Create or open metadata DBs for each log collection */ p_file_info->db_metadata = mallocz(snprintf(NULL, 0, "%s" METADATA_DB_FILENAME, p_file_info->db_dir) + 1); sprintf((char *) p_file_info->db_metadata, "%s" METADATA_DB_FILENAME, p_file_info->db_dir); rc = sqlite3_open(p_file_info->db_metadata, &p_file_info->db); do_sqlite_error_check(p_file_info, rc, SQLITE_OK); /* Configure metadata DB */ rc = sqlite3_exec(p_file_info->db, "PRAGMA auto_vacuum = INCREMENTAL;" "PRAGMA synchronous = 1;" "PRAGMA journal_mode = WAL;" "PRAGMA temp_store = MEMORY;" "PRAGMA foreign_keys = ON;", 0, 0, &err_msg); if (unlikely(rc != SQLITE_OK)) { collector_error("[%s]: Failed to configure database, SQL error: %s", p_file_info->filename, err_msg); throw_error(p_file_info->chartname, ERR_TYPE_SQLITE, rc, __LINE__, __FILE__, __FUNCTION__); uv_mutex_unlock(p_file_info->db_mut); goto return_error; } /* Execute pending metadata database migrations */ collector_info("[%s]: About to execute " METADATA_DB_FILENAME " migrations", p_file_info->chartname); int metadata_db_ver = db_user_version(p_file_info->db, -1); if (likely(LOGS_MANAG_DB_VERSION == metadata_db_ver)) { collector_info( "[%s]: Logs management " METADATA_DB_FILENAME " database version is %d (no migration needed)", p_file_info->chartname, metadata_db_ver); } else { for(int ver = metadata_db_ver; ver < LOGS_MANAG_DB_VERSION && migration_list_metadata_db[ver].func; ver++){ rc = (migration_list_metadata_db[ver].func)(p_file_info->db, migration_list_metadata_db[ver].name); if (unlikely(rc)){ collector_error("[%s]: Logs management " METADATA_DB_FILENAME " database migration from version %d to version %d failed", p_file_info->chartname, ver, ver + 1); throw_error(MAIN_DB, ERR_TYPE_SQLITE, rc, __LINE__, __FILE__, __FUNCTION__); uv_mutex_unlock(p_file_info->db_mut); goto return_error; } db_user_version(p_file_info->db, ver + 1); } } /* ----------------------------------------------------------------- * Create BLOBS_TABLE and LOGS_TABLE if they don't exist. Do it * as a transaction, so that it can all be rolled back if something * goes wrong. * -------------------------------------------------------------- */ { rc = sqlite3_exec(p_file_info->db, "BEGIN TRANSACTION;", NULL, NULL, NULL); do_sqlite_error_check(p_file_info, rc, SQLITE_OK); /* Check if BLOBS_TABLE exists or not */ sqlite3_stmt *stmt_check_if_BLOBS_TABLE_exists = NULL; rc = sqlite3_prepare_v2(p_file_info->db, "SELECT COUNT(*) FROM sqlite_master" " WHERE type='table' AND name='"BLOBS_TABLE"';", -1, &stmt_check_if_BLOBS_TABLE_exists, NULL); do_sqlite_error_check(p_file_info, rc, SQLITE_OK); rc = sqlite3_step(stmt_check_if_BLOBS_TABLE_exists); do_sqlite_error_check(p_file_info, rc, SQLITE_ROW); /* If BLOBS_TABLE doesn't exist, create and populate it */ if(sqlite3_column_int(stmt_check_if_BLOBS_TABLE_exists, 0) == 0){ /* 1. Create it */ rc = sqlite3_exec(p_file_info->db, "CREATE TABLE IF NOT EXISTS " BLOBS_TABLE "(" "Id INTEGER PRIMARY KEY," "Filename TEXT NOT NULL," "Filesize INTEGER NOT NULL" ");", 0, 0, &err_msg); if (unlikely(SQLITE_OK != rc)) { collector_error("[%s]: Failed to create " BLOBS_TABLE ", SQL error: %s", p_file_info->chartname, err_msg); throw_error(p_file_info->chartname, ERR_TYPE_SQLITE, rc, __LINE__, __FILE__, __FUNCTION__); uv_mutex_unlock(p_file_info->db_mut); goto return_error; } else collector_info("[%s]: Table " BLOBS_TABLE " created successfully", p_file_info->chartname); /* 2. Populate it */ sqlite3_stmt *stmt_init_BLOBS_table = NULL; rc = sqlite3_prepare_v2(p_file_info->db, "INSERT INTO " BLOBS_TABLE " (Filename, Filesize) VALUES (?,?) ;", -1, &stmt_init_BLOBS_table, NULL); do_sqlite_error_check(p_file_info, rc, SQLITE_OK); for(int i = 0; i < BLOB_MAX_FILES; i++){ char filename[FILENAME_MAX + 1]; snprintfz(filename, FILENAME_MAX, BLOB_STORE_FILENAME "%d", i); if(unlikely( SQLITE_OK != (rc = sqlite3_bind_text(stmt_init_BLOBS_table, 1, filename, -1, NULL)) || SQLITE_OK != (rc = sqlite3_bind_int64(stmt_init_BLOBS_table, 2, (sqlite3_int64) 0)) || SQLITE_DONE != (rc = sqlite3_step(stmt_init_BLOBS_table)) || SQLITE_OK != (rc = sqlite3_reset(stmt_init_BLOBS_table)))){ throw_error(p_file_info->chartname, ERR_TYPE_SQLITE, rc, __LINE__, __FILE__, __FUNCTION__); uv_mutex_unlock(p_file_info->db_mut); goto return_error; } } rc = sqlite3_finalize(stmt_init_BLOBS_table); do_sqlite_error_check(p_file_info, rc, SQLITE_OK); } rc = sqlite3_finalize(stmt_check_if_BLOBS_TABLE_exists); do_sqlite_error_check(p_file_info, rc, SQLITE_OK); /* If LOGS_TABLE doesn't exist, create it */ rc = sqlite3_exec(p_file_info->db, "CREATE TABLE IF NOT EXISTS " LOGS_TABLE "(" "Id INTEGER PRIMARY KEY," "FK_BLOB_Id INTEGER NOT NULL," "BLOB_Offset INTEGER NOT NULL," "Timestamp INTEGER NOT NULL," "Msg_compr_size INTEGER NOT NULL," "Msg_decompr_size INTEGER NOT NULL," "Num_lines INTEGER NOT NULL," "FOREIGN KEY (FK_BLOB_Id) REFERENCES " BLOBS_TABLE " (Id) ON DELETE CASCADE ON UPDATE CASCADE" ");", 0, 0, &err_msg); if (unlikely(SQLITE_OK != rc)) { collector_error("[%s]: Failed to create " LOGS_TABLE ", SQL error: %s", p_file_info->chartname, err_msg); throw_error(p_file_info->chartname, ERR_TYPE_SQLITE, rc, __LINE__, __FILE__, __FUNCTION__); uv_mutex_unlock(p_file_info->db_mut); goto return_error; } else collector_info("[%s]: Table " LOGS_TABLE " created successfully", p_file_info->chartname); /* Create index on LOGS_TABLE Timestamp * TODO: If this doesn't speed up queries, check SQLITE R*tree * module. Requires benchmarking with/without index. */ rc = sqlite3_exec(p_file_info->db, "CREATE INDEX IF NOT EXISTS logs_timestamps_idx " "ON " LOGS_TABLE "(Timestamp);", 0, 0, &err_msg); if (unlikely(SQLITE_OK != rc)) { collector_error("[%s]: Failed to create logs_timestamps_idx, SQL error: %s", p_file_info->chartname, err_msg); throw_error(p_file_info->chartname, ERR_TYPE_SQLITE, rc, __LINE__, __FILE__, __FUNCTION__); uv_mutex_unlock(p_file_info->db_mut); goto return_error; } else collector_info("[%s]: logs_timestamps_idx created successfully", p_file_info->chartname); rc = sqlite3_exec(p_file_info->db, "END TRANSACTION;", NULL, NULL, NULL); do_sqlite_error_check(p_file_info, rc, SQLITE_OK); } /* ----------------------------------------------------------------- * Remove excess BLOBs beyond BLOB_MAX_FILES (from both DB and disk * storage). * * This is useful if BLOB_MAX_FILES is reduced after an agent * restart (for example, if in the future it is not hardcoded, * but instead it is read from the configuration file). LOGS_TABLE * entries should be deleted automatically (due to ON DELETE CASCADE). * -------------------------------------------------------------- */ { sqlite3_stmt *stmt_get_BLOBS_TABLE_size = NULL; rc = sqlite3_prepare_v2(p_file_info->db, "SELECT MAX(Id) FROM " BLOBS_TABLE ";", -1, &stmt_get_BLOBS_TABLE_size, NULL); do_sqlite_error_check(p_file_info, rc, SQLITE_OK); rc = sqlite3_step(stmt_get_BLOBS_TABLE_size); do_sqlite_error_check(p_file_info, rc, SQLITE_ROW); const int blobs_table_max_id = sqlite3_column_int(stmt_get_BLOBS_TABLE_size, 0); sqlite3_stmt *stmt_retrieve_filename_last_digits = NULL; // This statement retrieves the last digit(s) from the Filename column of BLOBS_TABLE rc = sqlite3_prepare_v2(p_file_info->db, "WITH split(word, str) AS ( SELECT '', (SELECT Filename FROM " BLOBS_TABLE " WHERE Id = ? ) || '.' " "UNION ALL SELECT substr(str, 0, instr(str, '.')), substr(str, instr(str, '.')+1) FROM split WHERE str!='' ) " "SELECT word FROM split WHERE word!='' ORDER BY LENGTH(str) LIMIT 1;", -1, &stmt_retrieve_filename_last_digits, NULL); do_sqlite_error_check(p_file_info, rc, SQLITE_OK); sqlite3_stmt *stmt_delete_row_by_id = NULL; rc = sqlite3_prepare_v2(p_file_info->db, "DELETE FROM " BLOBS_TABLE " WHERE Id = ?;", -1, &stmt_delete_row_by_id, NULL); do_sqlite_error_check(p_file_info, rc, SQLITE_OK); for (int id = 1; id <= blobs_table_max_id; id++){ rc = sqlite3_bind_int(stmt_retrieve_filename_last_digits, 1, id); do_sqlite_error_check(p_file_info, rc, SQLITE_OK); rc = sqlite3_step(stmt_retrieve_filename_last_digits); do_sqlite_error_check(p_file_info, rc, SQLITE_ROW); int last_digits = sqlite3_column_int(stmt_retrieve_filename_last_digits, 0); rc = sqlite3_reset(stmt_retrieve_filename_last_digits); do_sqlite_error_check(p_file_info, rc, SQLITE_OK); /* If last_digits > BLOB_MAX_FILES - 1, then some BLOB files * will need to be removed (both from DB BLOBS_TABLE and * also from the disk). */ if(last_digits > BLOB_MAX_FILES - 1){ /* Delete BLOB file from filesystem */ char blob_delete_path[FILENAME_MAX + 1]; snprintfz(blob_delete_path, FILENAME_MAX, "%s" BLOB_STORE_FILENAME "%d", p_file_info->db_dir, last_digits); uv_fs_t unlink_req; rc = uv_fs_unlink(NULL, &unlink_req, blob_delete_path, NULL); uv_fs_req_cleanup(&unlink_req); if (unlikely(rc)) { // TODO: If there is an erro here, the entry won't be deleted from BLOBS_TABLE. What to do? throw_error(p_file_info->chartname, ERR_TYPE_LIBUV, rc, __LINE__, __FILE__, __FUNCTION__); uv_mutex_unlock(p_file_info->db_mut); goto return_error; } do_sqlite_error_check(p_file_info, rc, SQLITE_OK); /* Delete entry from DB BLOBS_TABLE */ rc = sqlite3_bind_int(stmt_delete_row_by_id, 1, id); do_sqlite_error_check(p_file_info, rc, SQLITE_OK); rc = sqlite3_step(stmt_delete_row_by_id); do_sqlite_error_check(p_file_info, rc, SQLITE_DONE); rc = sqlite3_reset(stmt_delete_row_by_id); do_sqlite_error_check(p_file_info, rc, SQLITE_OK); } } rc = sqlite3_finalize(stmt_retrieve_filename_last_digits); do_sqlite_error_check(p_file_info, rc, SQLITE_OK); rc = sqlite3_finalize(stmt_delete_row_by_id); do_sqlite_error_check(p_file_info, rc, SQLITE_OK); /* ------------------------------------------------------------- * BLOBS_TABLE ids after the deletion might not be contiguous. * This needs to be fixed, by having the ids updated. * LOGS_TABLE FKs will be updated automatically * (due to ON UPDATE CASCADE). * ---------------------------------------------------------- */ int old_blobs_table_ids[BLOB_MAX_FILES]; int off = 0; sqlite3_stmt *stmt_retrieve_all_ids = NULL; rc = sqlite3_prepare_v2(p_file_info->db, "SELECT Id FROM " BLOBS_TABLE " ORDER BY Id ASC;", -1, &stmt_retrieve_all_ids, NULL); do_sqlite_error_check(p_file_info, rc, SQLITE_OK); rc = sqlite3_step(stmt_retrieve_all_ids); while(rc == SQLITE_ROW){ old_blobs_table_ids[off++] = sqlite3_column_int(stmt_retrieve_all_ids, 0); rc = sqlite3_step(stmt_retrieve_all_ids); } do_sqlite_error_check(p_file_info, rc, SQLITE_DONE); rc = sqlite3_finalize(stmt_retrieve_all_ids); do_sqlite_error_check(p_file_info, rc, SQLITE_OK); sqlite3_stmt *stmt_update_id = NULL; rc = sqlite3_prepare_v2(p_file_info->db, "UPDATE " BLOBS_TABLE " SET Id = ? WHERE Id = ?;", -1, &stmt_update_id, NULL); do_sqlite_error_check(p_file_info, rc, SQLITE_OK); for (int i = 0; i < BLOB_MAX_FILES; i++){ if(unlikely( SQLITE_OK != (rc = sqlite3_bind_int(stmt_update_id, 1, i + 1)) || SQLITE_OK != (rc = sqlite3_bind_int(stmt_update_id, 2, old_blobs_table_ids[i])) || SQLITE_DONE != (rc = sqlite3_step(stmt_update_id)) || SQLITE_OK != (rc = sqlite3_reset(stmt_update_id)))) { throw_error(p_file_info->chartname, ERR_TYPE_SQLITE, rc, __LINE__, __FILE__, __FUNCTION__); uv_mutex_unlock(p_file_info->db_mut); goto return_error; } } rc = sqlite3_finalize(stmt_update_id); do_sqlite_error_check(p_file_info, rc, SQLITE_OK); } /* ----------------------------------------------------------------- * Traverse BLOBS_TABLE, open logs.bin.X files and store their * file handles in p_file_info array. * -------------------------------------------------------------- */ sqlite3_stmt *stmt_retrieve_metadata_from_id = NULL; rc = sqlite3_prepare_v2(p_file_info->db, "SELECT Filename, Filesize FROM " BLOBS_TABLE " WHERE Id = ? ;", -1, &stmt_retrieve_metadata_from_id, NULL); do_sqlite_error_check(p_file_info, rc, SQLITE_OK); sqlite3_stmt *stmt_retrieve_total_logs_size = NULL; rc = sqlite3_prepare_v2(p_file_info->db, "SELECT SUM(Msg_compr_size) FROM " LOGS_TABLE " WHERE FK_BLOB_Id = ? GROUP BY FK_BLOB_Id ;", -1, &stmt_retrieve_total_logs_size, NULL); do_sqlite_error_check(p_file_info, rc, SQLITE_OK); uv_fs_t open_req; for(int id = 1; id <= BLOB_MAX_FILES; id++){ /* Open BLOB file based on filename stored in BLOBS_TABLE. */ rc = sqlite3_bind_int(stmt_retrieve_metadata_from_id, 1, id); do_sqlite_error_check(p_file_info, rc, SQLITE_OK); rc = sqlite3_step(stmt_retrieve_metadata_from_id); do_sqlite_error_check(p_file_info, rc, SQLITE_ROW); char filename[FILENAME_MAX + 1] = {0}; snprintfz(filename, FILENAME_MAX, "%s%s", p_file_info->db_dir, sqlite3_column_text(stmt_retrieve_metadata_from_id, 0)); rc = uv_fs_open(NULL, &open_req, filename, UV_FS_O_RDWR | UV_FS_O_CREAT | UV_FS_O_APPEND | UV_FS_O_RANDOM, 0644, NULL); if (unlikely(rc < 0)){ uv_fs_req_cleanup(&open_req); throw_error(p_file_info->chartname, ERR_TYPE_LIBUV, rc, __LINE__, __FILE__, __FUNCTION__); uv_mutex_unlock(p_file_info->db_mut); goto return_error; } // open_req.result of a uv_fs_t is the file descriptor in case of the uv_fs_open p_file_info->blob_handles[id] = open_req.result; uv_fs_req_cleanup(&open_req); const int64_t metadata_filesize = (int64_t) sqlite3_column_int64(stmt_retrieve_metadata_from_id, 1); /* ------------------------------------------------------------- * Retrieve total log messages compressed size from LOGS_TABLE * for current FK_BLOB_Id. * Only to assert whether correct - not used elsewhere. * * If no rows are returned, it means it is probably the initial * execution of the program so still valid (except if rc is other * than SQLITE_DONE, which is an error then). * ---------------------------------------------------------- */ rc = sqlite3_bind_int(stmt_retrieve_total_logs_size, 1, id); do_sqlite_error_check(p_file_info, rc, SQLITE_OK); rc = sqlite3_step(stmt_retrieve_total_logs_size); if (SQLITE_ROW == rc){ const int64_t total_logs_filesize = (int64_t) sqlite3_column_int64(stmt_retrieve_total_logs_size, 0); if(unlikely(total_logs_filesize != metadata_filesize)){ throw_error(p_file_info->chartname, ERR_TYPE_OTHER, rc, __LINE__, __FILE__, __FUNCTION__); uv_mutex_unlock(p_file_info->db_mut); goto return_error; } } else do_sqlite_error_check(p_file_info, rc, SQLITE_DONE); /* Get filesize of BLOB file. */ uv_fs_t stat_req; rc = uv_fs_stat(NULL, &stat_req, filename, NULL); if (unlikely(rc)){ uv_fs_req_cleanup(&stat_req); throw_error(p_file_info->chartname, ERR_TYPE_LIBUV, rc, __LINE__, __FILE__, __FUNCTION__); uv_mutex_unlock(p_file_info->db_mut); goto return_error; } const int64_t blob_filesize = (int64_t) stat_req.statbuf.st_size; uv_fs_req_cleanup(&stat_req); do{ /* Case 1: blob_filesize == metadata_filesize (equal, either both zero or not): All good */ if(likely(blob_filesize == metadata_filesize)) break; /* Case 2: blob_filesize == 0 && metadata_filesize > 0: fatal(), however could it mean that * EXT_BLOB_STORE_FILENAME was rotated but the SQLite metadata wasn't updated? So can it * maybe be recovered by un-rotating? Either way, treat as fatal error for now. */ // TODO: Can we avoid fatal()? if(unlikely(blob_filesize == 0 && metadata_filesize > 0)){ collector_error("[%s]: blob_filesize == 0 but metadata_filesize > 0 for '%s'\n", p_file_info->chartname, filename); throw_error(p_file_info->chartname, ERR_TYPE_OTHER, rc, __LINE__, __FILE__, __FUNCTION__); uv_mutex_unlock(p_file_info->db_mut); goto return_error; } /* Case 3: blob_filesize > metadata_filesize: Truncate binary to sqlite filesize, program * crashed or terminated after writing BLOBs to external file but before metadata was updated */ if(unlikely(blob_filesize > metadata_filesize)){ collector_info("[%s]: blob_filesize > metadata_filesize for '%s'. Will attempt to fix it.", p_file_info->chartname, filename); uv_fs_t trunc_req; rc = uv_fs_ftruncate(NULL, &trunc_req, p_file_info->blob_handles[id], metadata_filesize, NULL); uv_fs_req_cleanup(&trunc_req); if(unlikely(rc)) { throw_error(p_file_info->chartname, ERR_TYPE_LIBUV, rc, __LINE__, __FILE__, __FUNCTION__); uv_mutex_unlock(p_file_info->db_mut); goto return_error; } break; } /* Case 4: blob_filesize < metadata_filesize: unrecoverable, * maybe rotation went horrible wrong? * TODO: Delete external BLOB and clear metadata from DB, * start from clean state but the most recent logs. */ if(unlikely(blob_filesize < metadata_filesize)){ collector_info("[%s]: blob_filesize < metadata_filesize for '%s'.", p_file_info->chartname, filename); throw_error(p_file_info->chartname, ERR_TYPE_OTHER, rc, __LINE__, __FILE__, __FUNCTION__); uv_mutex_unlock(p_file_info->db_mut); goto return_error; } /* Case 5: default if none of the above, should never reach here, fatal() */ m_assert(0, "Code should not reach here"); throw_error(p_file_info->chartname, ERR_TYPE_OTHER, rc, __LINE__, __FILE__, __FUNCTION__); uv_mutex_unlock(p_file_info->db_mut); goto return_error; } while(0); /* Initialise blob_write_handle with logs.bin.0 */ if(filename[strlen(filename) - 1] == '0') p_file_info->blob_write_handle_offset = id; rc = sqlite3_reset(stmt_retrieve_total_logs_size); do_sqlite_error_check(p_file_info, rc, SQLITE_OK); rc = sqlite3_reset(stmt_retrieve_metadata_from_id); do_sqlite_error_check(p_file_info, rc, SQLITE_OK); } rc = sqlite3_finalize(stmt_retrieve_metadata_from_id); do_sqlite_error_check(p_file_info, rc, SQLITE_OK); /* Prepare statements to be used in single database queries */ rc = sqlite3_prepare_v2(p_file_info->db, "SELECT Timestamp, Msg_compr_size , Msg_decompr_size, " "BLOB_Offset, " BLOBS_TABLE".Id, Num_lines " "FROM " LOGS_TABLE " INNER JOIN " BLOBS_TABLE " " "ON " LOGS_TABLE ".FK_BLOB_Id = " BLOBS_TABLE ".Id " "WHERE Timestamp >= ? AND Timestamp <= ? " "ORDER BY Timestamp;", -1, &p_file_info->stmt_get_log_msg_metadata_asc, NULL); do_sqlite_error_check(p_file_info, rc, SQLITE_OK); rc = sqlite3_prepare_v2(p_file_info->db, "SELECT Timestamp, Msg_compr_size , Msg_decompr_size, " "BLOB_Offset, " BLOBS_TABLE".Id, Num_lines " "FROM " LOGS_TABLE " INNER JOIN " BLOBS_TABLE " " "ON " LOGS_TABLE ".FK_BLOB_Id = " BLOBS_TABLE ".Id " "WHERE Timestamp <= ? AND Timestamp >= ? " "ORDER BY Timestamp DESC;", -1, &p_file_info->stmt_get_log_msg_metadata_desc, NULL); do_sqlite_error_check(p_file_info, rc, SQLITE_OK); /* DB initialisation finished; release lock */ uv_mutex_unlock(p_file_info->db_mut); /* Create synchronous writer thread, one for each log source */ p_file_info->db_writer_thread = mallocz(sizeof(uv_thread_t)); rc = uv_thread_create(p_file_info->db_writer_thread, db_writer_db_mode_full, p_file_info); if (unlikely(rc)){ throw_error(p_file_info->chartname, ERR_TYPE_LIBUV, rc, __LINE__, __FILE__, __FUNCTION__); goto return_error; } } } rc = sqlite3_finalize(stmt_search_if_log_source_exists); if (unlikely(rc != SQLITE_OK)){ throw_error(MAIN_DB, ERR_TYPE_SQLITE, rc, __LINE__, __FILE__, __FUNCTION__); // TODO: Some additional cleanup required here, e.g. terminate db_writer_thread. goto return_error; } rc = sqlite3_finalize(stmt_insert_log_collection_metadata); if (unlikely(rc != SQLITE_OK)){ throw_error(MAIN_DB, ERR_TYPE_SQLITE, rc, __LINE__, __FILE__, __FUNCTION__); // TODO: Some additional cleanup required here, e.g. terminate db_writer_thread. goto return_error; } return 0; return_error: freez(main_db_path); main_db_path = NULL; sqlite3_close(main_db); // No-op if main_db == NULL sqlite3_free(err_msg); // No-op if err_msg == NULL m_assert(rc != 0, "rc should not be == 0 in case of error"); return rc == 0 ? -1 : rc; } /** * @brief Search database(s) for logs * @details This function searches one or more databases for any results * matching the query parameters. If any results are found, it will decompress * the text of each returned row and add it to the results buffer, up to a * maximum amount of p_query_params->quota bytes (unless timed out). * @todo Make decompress buffer static to reduce mallocs/frees. * @todo Limit number of results returned through SQLite Query to speed up search? */ void db_search(logs_query_params_t *const p_query_params, struct File_info *const p_file_infos[]) { int rc = 0; sqlite3_stmt *stmt_get_log_msg_metadata; sqlite3 *dbt = NULL; // Used only when multiple DBs are searched if(!p_file_infos[1]){ /* Single DB to be searched */ stmt_get_log_msg_metadata = p_query_params->order_by_asc ? p_file_infos[0]->stmt_get_log_msg_metadata_asc : p_file_infos[0]->stmt_get_log_msg_metadata_desc; if(unlikely( SQLITE_OK != (rc = sqlite3_bind_int64(stmt_get_log_msg_metadata, 1, p_query_params->req_from_ts)) || SQLITE_OK != (rc = sqlite3_bind_int64(stmt_get_log_msg_metadata, 2, p_query_params->req_to_ts)) || (SQLITE_ROW != (rc = sqlite3_step(stmt_get_log_msg_metadata)) && (SQLITE_DONE != rc)) )){ throw_error(p_file_infos[0]->chartname, ERR_TYPE_SQLITE, rc, __LINE__, __FILE__, __FUNCTION__); // TODO: If there are errors here, should db_writer_db_mode_full() be terminated? sqlite3_reset(stmt_get_log_msg_metadata); return; } } else { /* Multiple DBs to be searched */ sqlite3_stmt *stmt_attach_db; sqlite3_stmt *stmt_create_tmp_view; int pfi_off = 0; /* Open a new DB connection on the first log source DB and attach other DBs */ if(unlikely( SQLITE_OK != (rc = sqlite3_open_v2(p_file_infos[0]->db_metadata, &dbt, SQLITE_OPEN_READONLY, NULL)) || SQLITE_OK != (rc = sqlite3_prepare_v2(dbt,"ATTACH DATABASE ? AS ? ;", -1, &stmt_attach_db, NULL)) )){ throw_error(p_file_infos[0]->chartname, ERR_TYPE_SQLITE, rc, __LINE__, __FILE__, __FUNCTION__); sqlite3_close_v2(dbt); return; } for(pfi_off = 0; p_file_infos[pfi_off]; pfi_off++){ if(unlikely( SQLITE_OK != (rc = sqlite3_bind_text(stmt_attach_db, 1, p_file_infos[pfi_off]->db_metadata, -1, NULL)) || SQLITE_OK != (rc = sqlite3_bind_int(stmt_attach_db, 2, pfi_off)) || SQLITE_DONE != (rc = sqlite3_step(stmt_attach_db)) || SQLITE_OK != (rc = sqlite3_reset(stmt_attach_db)) )){ throw_error(p_file_infos[pfi_off]->chartname, ERR_TYPE_SQLITE, rc, __LINE__, __FILE__, __FUNCTION__); sqlite3_close_v2(dbt); return; } } /* Create temporary view, then prepare retrieval of metadata from * TMP_VIEW_TABLE statement and execute search. * TODO: Limit number of results returned through SQLite Query to speed up search? */ #define TMP_VIEW_TABLE "compound_view" #define TMP_VIEW_QUERY_PREFIX "CREATE TEMP VIEW " TMP_VIEW_TABLE " AS SELECT * FROM (SELECT * FROM '0'."\ LOGS_TABLE " INNER JOIN (VALUES(0)) ORDER BY Timestamp) " #define TMP_VIEW_QUERY_BODY_1 "UNION ALL SELECT * FROM (SELECT * FROM '" #define TMP_VIEW_QUERY_BODY_2 "'." LOGS_TABLE " INNER JOIN (VALUES(" #define TMP_VIEW_QUERY_BODY_3 ")) ORDER BY Timestamp) " #define TMP_VIEW_QUERY_POSTFIX "ORDER BY Timestamp;" char tmp_view_query[sizeof(TMP_VIEW_QUERY_PREFIX) + ( sizeof(TMP_VIEW_QUERY_BODY_1) + sizeof(TMP_VIEW_QUERY_BODY_2) + sizeof(TMP_VIEW_QUERY_BODY_3) + 4 ) * (LOGS_MANAG_MAX_COMPOUND_QUERY_SOURCES - 1) + sizeof(TMP_VIEW_QUERY_POSTFIX) + 50 /* +50 bytes to play it safe */] = TMP_VIEW_QUERY_PREFIX; int pos = sizeof(TMP_VIEW_QUERY_PREFIX) - 1; for(pfi_off = 1; p_file_infos[pfi_off]; pfi_off++){ // Skip p_file_infos[0] int n = snprintf(&tmp_view_query[pos], sizeof(tmp_view_query) - pos, "%s%d%s%d%s", TMP_VIEW_QUERY_BODY_1, pfi_off, TMP_VIEW_QUERY_BODY_2, pfi_off, TMP_VIEW_QUERY_BODY_3); if (n < 0 || n >= (int) sizeof(tmp_view_query) - pos){ throw_error(p_file_infos[pfi_off]->chartname, ERR_TYPE_OTHER, n, __LINE__, __FILE__, __FUNCTION__); sqlite3_close_v2(dbt); return; } pos += n; } snprintf(&tmp_view_query[pos], sizeof(tmp_view_query) - pos, "%s", TMP_VIEW_QUERY_POSTFIX); if(unlikely( SQLITE_OK != (rc = sqlite3_prepare_v2(dbt, tmp_view_query, -1, &stmt_create_tmp_view, NULL)) || SQLITE_DONE != (rc = sqlite3_step(stmt_create_tmp_view)) || SQLITE_OK != (rc = sqlite3_prepare_v2(dbt, p_query_params->order_by_asc ? "SELECT Timestamp, Msg_compr_size , Msg_decompr_size, " "BLOB_Offset, FK_BLOB_Id, Num_lines, column1 " "FROM " TMP_VIEW_TABLE " " "WHERE Timestamp >= ? AND Timestamp <= ?;" : /* TODO: The following can also be done by defining * a descending order tmp_view_query, which will * probably be faster. Needs to be measured. */ "SELECT Timestamp, Msg_compr_size , Msg_decompr_size, " "BLOB_Offset, FK_BLOB_Id, Num_lines, column1 " "FROM " TMP_VIEW_TABLE " " "WHERE Timestamp <= ? AND Timestamp >= ? ORDER BY Timestamp DESC;", -1, &stmt_get_log_msg_metadata, NULL)) || SQLITE_OK != (rc = sqlite3_bind_int64(stmt_get_log_msg_metadata, 1, (sqlite3_int64)p_query_params->req_from_ts)) || SQLITE_OK != (rc = sqlite3_bind_int64(stmt_get_log_msg_metadata, 2, (sqlite3_int64)p_query_params->req_to_ts)) || (SQLITE_ROW != (rc = sqlite3_step(stmt_get_log_msg_metadata)) && (SQLITE_DONE != rc)) )){ throw_error(p_file_infos[0]->chartname, ERR_TYPE_SQLITE, rc, __LINE__, __FILE__, __FUNCTION__); sqlite3_close_v2(dbt); return; } } Circ_buff_item_t tmp_itm = {0}; BUFFER *const res_buff = p_query_params->results_buff; logs_query_res_hdr_t res_hdr = { // results header .timestamp = p_query_params->act_to_ts, .text_size = 0, .matches = 0, .log_source = "", .log_type = "", .basename = "", .filename = "", .chartname ="" }; size_t text_compressed_size_max = 0; while (rc == SQLITE_ROW) { /* Retrieve metadata from DB */ tmp_itm.timestamp = (msec_t)sqlite3_column_int64(stmt_get_log_msg_metadata, 0); tmp_itm.text_compressed_size = (size_t)sqlite3_column_int64(stmt_get_log_msg_metadata, 1); tmp_itm.text_size = (size_t)sqlite3_column_int64(stmt_get_log_msg_metadata, 2); int64_t blob_offset = (int64_t) sqlite3_column_int64(stmt_get_log_msg_metadata, 3); int blob_handles_offset = sqlite3_column_int(stmt_get_log_msg_metadata, 4); unsigned long num_lines = (unsigned long) sqlite3_column_int64(stmt_get_log_msg_metadata, 5); int db_off = p_file_infos[1] ? sqlite3_column_int(stmt_get_log_msg_metadata, 6) : 0; /* If exceeding quota or timeout is reached and new timestamp * is different than previous, terminate query. */ if((res_buff->len >= p_query_params->quota || terminate_logs_manag_query(p_query_params)) && tmp_itm.timestamp != res_hdr.timestamp){ p_query_params->act_to_ts = res_hdr.timestamp; break; } res_hdr.timestamp = tmp_itm.timestamp; snprintfz(res_hdr.log_source, sizeof(res_hdr.log_source), "%s", log_src_t_str[p_file_infos[db_off]->log_source]); snprintfz(res_hdr.log_type, sizeof(res_hdr.log_type), "%s", log_src_type_t_str[p_file_infos[db_off]->log_type]); snprintfz(res_hdr.basename, sizeof(res_hdr.basename), "%s", p_file_infos[db_off]->file_basename); snprintfz(res_hdr.filename, sizeof(res_hdr.filename), "%s", p_file_infos[db_off]->filename); snprintfz(res_hdr.chartname, sizeof(res_hdr.chartname), "%s", p_file_infos[db_off]->chartname); /* Retrieve compressed log messages from BLOB file */ if(tmp_itm.text_compressed_size > text_compressed_size_max){ text_compressed_size_max = tmp_itm.text_compressed_size; tmp_itm.text_compressed = reallocz(tmp_itm.text_compressed, text_compressed_size_max); } uv_fs_t read_req; uv_buf_t uv_buf = uv_buf_init(tmp_itm.text_compressed, tmp_itm.text_compressed_size); rc = uv_fs_read(NULL, &read_req, p_file_infos[db_off]->blob_handles[blob_handles_offset], &uv_buf, 1, blob_offset, NULL); uv_fs_req_cleanup(&read_req); if (unlikely(rc < 0)){ throw_error(NULL, ERR_TYPE_LIBUV, rc, __LINE__, __FILE__, __FUNCTION__); break; } /* Append retrieved results to BUFFER. * In the case of search_keyword(), less than sizeof(res_hdr) + tmp_itm.text_size *space may be required, but go for worst case scenario for now */ buffer_increase(res_buff, sizeof(res_hdr) + tmp_itm.text_size); if(!p_query_params->keyword || !*p_query_params->keyword || !strcmp(p_query_params->keyword, " ")){ rc = LZ4_decompress_safe(tmp_itm.text_compressed, &res_buff->buffer[res_buff->len + sizeof(res_hdr)], tmp_itm.text_compressed_size, tmp_itm.text_size); if(unlikely(rc < 0)){ throw_error(p_file_infos[db_off]->chartname, ERR_TYPE_OTHER, rc, __LINE__, __FILE__, __FUNCTION__); break; } res_hdr.matches = num_lines; res_hdr.text_size = tmp_itm.text_size; } else { tmp_itm.data = mallocz(tmp_itm.text_size); rc = LZ4_decompress_safe(tmp_itm.text_compressed, tmp_itm.data, tmp_itm.text_compressed_size, tmp_itm.text_size); if(unlikely(rc < 0)){ freez(tmp_itm.data); throw_error(p_file_infos[db_off]->chartname, ERR_TYPE_OTHER, rc, __LINE__, __FILE__, __FUNCTION__); break; } res_hdr.matches = search_keyword( tmp_itm.data, tmp_itm.text_size, &res_buff->buffer[res_buff->len + sizeof(res_hdr)], &res_hdr.text_size, p_query_params->keyword, NULL, p_query_params->ignore_case); freez(tmp_itm.data); m_assert( (res_hdr.matches > 0 && res_hdr.text_size > 0) || (res_hdr.matches == 0 && res_hdr.text_size == 0), "res_hdr.matches and res_hdr.text_size must both be > 0 or == 0."); if(unlikely(res_hdr.matches < 0)){ /* res_hdr.matches < 0 - error during keyword search */ throw_error(p_file_infos[db_off]->chartname, ERR_TYPE_LIBUV, rc, __LINE__, __FILE__, __FUNCTION__); break; } } if(res_hdr.text_size){ res_buff->buffer[res_buff->len + sizeof(res_hdr) + res_hdr.text_size - 1] = '\n'; // replace '\0' with '\n' memcpy(&res_buff->buffer[res_buff->len], &res_hdr, sizeof(res_hdr)); res_buff->len += sizeof(res_hdr) + res_hdr.text_size; p_query_params->num_lines += res_hdr.matches; } m_assert(TEST_MS_TIMESTAMP_VALID(res_hdr.timestamp), "res_hdr.timestamp is invalid"); rc = sqlite3_step(stmt_get_log_msg_metadata); if (unlikely(rc != SQLITE_ROW && rc != SQLITE_DONE)){ throw_error(p_file_infos[db_off]->chartname, ERR_TYPE_SQLITE, rc, __LINE__, __FILE__, __FUNCTION__); // TODO: If there are errors here, should db_writer_db_mode_full() be terminated? break; } } if(tmp_itm.text_compressed) freez(tmp_itm.text_compressed); if(p_file_infos[1]) rc = sqlite3_close_v2(dbt); else rc = sqlite3_reset(stmt_get_log_msg_metadata); if (unlikely(SQLITE_OK != rc)) throw_error(p_file_infos[0]->chartname, ERR_TYPE_SQLITE, rc, __LINE__, __FILE__, __FUNCTION__); }