From: Kern Sibbald Date: Thu, 18 Sep 2008 19:10:05 +0000 (+0000) Subject: Apply dbcheck patch from Yuri Timofeev . X-Git-Tag: Release-3.0.0~963 X-Git-Url: https://git.sur5r.net/?a=commitdiff_plain;h=f7858a36bc4981a10714ff9e3b247ca9c9fba7c2;p=bacula%2Fbacula Apply dbcheck patch from Yuri Timofeev . It significantly improves the performance of dbcheck for MySQL. git-svn-id: https://bacula.svn.sourceforge.net/svnroot/bacula/trunk@7613 91ce42f0-d328-0410-95d8-f526ca767f89 --- diff --git a/bacula/src/tools/dbcheck.c b/bacula/src/tools/dbcheck.c index 0b1c276252..2c13325892 100644 --- a/bacula/src/tools/dbcheck.c +++ b/bacula/src/tools/dbcheck.c @@ -63,8 +63,6 @@ typedef struct s_name_ctx { int tot_ids; /* total to process */ } NAME_LIST; - - /* Global variables */ static bool fix = false; static bool batch = false; @@ -99,6 +97,17 @@ static void repair_bad_paths(); static void repair_bad_filenames(); static void do_interactive_mode(); static bool yes_no(const char *prompt); +static bool check_idx(const char *col_name); +static bool create_tmp_idx(const char *idx_name, const char *table_name, + const char *col_name); +static bool drop_tmp_idx(const char *idx_name, const char *table_name); +#ifdef HAVE_MYSQL +static int check_idx_handler(void *ctx, int num_fields, char **row); +#endif + + +/* Global variables */ +static const char *idx_tmp_name; static void usage() @@ -135,7 +144,6 @@ int main (int argc, char *argv[]) memset(&id_list, 0, sizeof(id_list)); memset(&name_list, 0, sizeof(name_list)); - while ((ch = getopt(argc, argv, "bc:C:d:fv?")) != -1) { switch (ch) { case 'b': /* batch */ @@ -813,6 +821,15 @@ static void eliminate_orphaned_file_records() static void eliminate_orphaned_path_records() { + idx_tmp_name = NULL; + /* check the existence of the required "one column" index */ + if (!check_idx("PathId")) { + if (yes_no(_("Create temporary index? (yes/no): "))) { + /* create temporary index PathId */ + create_tmp_idx("idxPIchk", "File", "PathId"); + } + } + const char *query = "SELECT DISTINCT Path.PathId,File.PathId FROM Path " "LEFT OUTER JOIN File ON (Path.PathId=File.PathId) " "WHERE File.PathId IS NULL LIMIT 300000"; @@ -847,11 +864,22 @@ static void eliminate_orphaned_path_records() if (!make_id_list(query, &id_list)) { exit(1); } - } + } + /* drop temporary index idx_tmp_name */ + drop_tmp_idx("idxPIchk", "File"); } static void eliminate_orphaned_filename_records() { + idx_tmp_name = NULL; + /* check the existence of the required "one column" index */ + if (!check_idx("FilenameId") ) { + if (yes_no(_("Create temporary index? (yes/no): "))) { + /* create temporary index FilenameId */ + create_tmp_idx("idxFIchk", "File", "FilenameId"); + } + } + const char *query = "SELECT Filename.FilenameId,File.FilenameId FROM Filename " "LEFT OUTER JOIN File ON (Filename.FilenameId=File.FilenameId) " "WHERE File.FilenameId IS NULL LIMIT 300000"; @@ -887,6 +915,9 @@ static void eliminate_orphaned_filename_records() exit(1); } } + /* drop temporary index idx_tmp_name */ + drop_tmp_idx("idxFIchk", "File"); + } static void eliminate_orphaned_fileset_records() @@ -1230,3 +1261,143 @@ static bool yes_no(const char *prompt) } bool python_set_prog(JCR*, char const*) { return false; } + + +/* + * The code below to add indexes is needed only for MySQL, and + * that to improve the performance. + */ + +#ifdef HAVE_MYSQL +#define MAXIDX 100 +typedef struct s_idx_list { + char *key_name; + int count_key; /* how many times the index meets *key_name */ + int count_col; /* how many times meets the desired column name */ +} IDX_LIST; + +static IDX_LIST idx_list[MAXIDX]; + +/* + * Called here with each table index to be added to the list + */ +static int check_idx_handler(void *ctx, int num_fields, char **row) +{ + /* Table | Non_unique | Key_name | Seq_in_index | Column_name |... + * File | 0 | PRIMARY | 1 | FileId |... + */ + char *name, *key_name, *col_name; + int i, len; + int found = false; + name = (char *)ctx; + key_name = row[2]; + col_name = row[4]; + for(i = 0; (idx_list[i].key_name != NULL) && (i < MAXIDX); i++) { + if (strcasecmp(idx_list[i].key_name, key_name) == 0 ) { + idx_list[i].count_key++; + found = true; + if (strcasecmp(col_name, name) == 0) { + idx_list[i].count_col++; + } + break; + } + } + /* if the new Key_name, add it to the list */ + if (!found) { + len = strlen(key_name) + 1; + idx_list[i].key_name = (char *)malloc(len); + bstrncpy(idx_list[i].key_name, key_name, len); + idx_list[i].count_key = 1; + if (strcasecmp(col_name, name) == 0) { + idx_list[i].count_col = 1; + } else { + idx_list[i].count_col = 0; + } + } + return 0; +} +#endif + +/* + * Return TRUE if "one column" index over *col_name exists + */ +static bool check_idx(const char *col_name) +{ +#ifdef HAVE_MYSQL + int i; + int found = false; + + memset(&idx_list, 0, sizeof(idx_list)); + char *query = "SHOW INDEX FROM File"; + if (!db_sql_query(db, query, check_idx_handler, (void *)col_name)) { + printf("%s\n", db_strerror(db)); + } + + for(i = 0; (idx_list[i].key_name != NULL) && (i < MAXIDX) ; i++) { + /* NOTE : if (idx_list[i].count_key > 1) then index idx_list[i].key_name is "multiple-column" index */ + if ((idx_list[i].count_key == 1) && (idx_list[i].count_col == 1)) { + /* "one column" index over *col_name found */ + found = true; + } + } + if (found) { + if (verbose) { + printf(_("Ok. Index over the %s column already exists and dbcheck will work faster.\n"), col_name); + } + } else { + printf(_("Note. Index over the %s column not found, that can greatly slow down dbcheck.\n"), col_name); + } + + return found; +#else + return true; +#endif +} + +/* + * Create temporary one-column index + */ +static bool create_tmp_idx(const char *idx_name, const char *table_name, + const char *col_name) +{ + idx_tmp_name = NULL; + printf(_("Create temporary index... This may take some time!\n")); + bsnprintf(buf, sizeof(buf), "CREATE INDEX %s ON %s (%s)", idx_name, table_name, col_name); + if (verbose) { + printf("%s\n", buf); + } + if (db_sql_query(db, buf, NULL, NULL)) { + idx_tmp_name = idx_name; + if (verbose) { + printf(_("Temporary index created.\n")); + } + } else { + printf("%s\n", db_strerror(db)); + return false; + } + return true; +} + +/* + * Drop temporary index + */ +static bool drop_tmp_idx(const char *idx_name, const char *table_name) +{ + if (idx_tmp_name != NULL) { + printf(_("Drop temporary index.\n")); + bsnprintf(buf, sizeof(buf), "DROP INDEX %s ON %s", idx_name, table_name); + if (verbose) { + printf("%s\n", buf); + } + if (!db_sql_query(db, buf, NULL, NULL)) { + printf("%s\n", db_strerror(db)); + return false; + } else { + if (verbose) { + printf(_("Temporary index %s deleted.\n"), idx_tmp_name); + } + } + } + idx_tmp_name = NULL; + return true; +} diff --git a/bacula/technotes-2.5 b/bacula/technotes-2.5 index bddefefa65..2aad7e15d7 100644 --- a/bacula/technotes-2.5 +++ b/bacula/technotes-2.5 @@ -19,6 +19,9 @@ remove reader/writer in FOPTS???? General: 18Sep08 +kes Apply dbcheck patch from Yuri Timofeev . + It significantly improves the performance of dbcheck for + MySQL. kes Fix bad debug code call in src/filed/accurate.c 17Sep08 kes Change two Jmsgs in accurate to Dmsg to reduce unnecessary