| 1 | /* -*- mode: C++; c-basic-offset: 4; indent-tabs-mode: nil -*- */ |
| 2 | // vim: ft=cpp:expandtab:ts=8:sw=4:softtabstop=4: |
| 3 | #ident "$Id$" |
| 4 | /*====== |
| 5 | This file is part of PerconaFT. |
| 6 | |
| 7 | |
| 8 | Copyright (c) 2006, 2015, Percona and/or its affiliates. All rights reserved. |
| 9 | |
| 10 | PerconaFT is free software: you can redistribute it and/or modify |
| 11 | it under the terms of the GNU General Public License, version 2, |
| 12 | as published by the Free Software Foundation. |
| 13 | |
| 14 | PerconaFT is distributed in the hope that it will be useful, |
| 15 | but WITHOUT ANY WARRANTY; without even the implied warranty of |
| 16 | MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the |
| 17 | GNU General Public License for more details. |
| 18 | |
| 19 | You should have received a copy of the GNU General Public License |
| 20 | along with PerconaFT. If not, see <http://www.gnu.org/licenses/>. |
| 21 | |
| 22 | ---------------------------------------- |
| 23 | |
| 24 | PerconaFT is free software: you can redistribute it and/or modify |
| 25 | it under the terms of the GNU Affero General Public License, version 3, |
| 26 | as published by the Free Software Foundation. |
| 27 | |
| 28 | PerconaFT is distributed in the hope that it will be useful, |
| 29 | but WITHOUT ANY WARRANTY; without even the implied warranty of |
| 30 | MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the |
| 31 | GNU Affero General Public License for more details. |
| 32 | |
| 33 | You should have received a copy of the GNU Affero General Public License |
| 34 | along with PerconaFT. If not, see <http://www.gnu.org/licenses/>. |
| 35 | ======= */ |
| 36 | |
| 37 | #ident "Copyright (c) 2006, 2015, Percona and/or its affiliates. All rights reserved." |
| 38 | |
| 39 | #pragma once |
| 40 | |
| 41 | // This must be first to make the 64-bit file mode work right in Linux |
| 42 | #define _FILE_OFFSET_BITS 64 |
| 43 | |
| 44 | #include <db.h> |
| 45 | |
| 46 | #include "ft/cachetable/cachetable.h" |
| 47 | #include "ft/comparator.h" |
| 48 | #include "ft/msg.h" |
| 49 | #include "util/dbt.h" |
| 50 | |
| 51 | #define OS_PATH_SEPARATOR '/' |
| 52 | |
| 53 | typedef struct ft_handle *FT_HANDLE; |
| 54 | |
| 55 | int toku_open_ft_handle (const char *fname, int is_create, FT_HANDLE *, int nodesize, int basementnodesize, enum toku_compression_method compression_method, CACHETABLE, TOKUTXN, int(*)(DB *,const DBT*,const DBT*)) __attribute__ ((warn_unused_result)); |
| 56 | |
| 57 | // effect: changes the descriptor for the ft of the given handle. |
| 58 | // requires: |
| 59 | // - cannot change descriptor for same ft in two threads in parallel. |
| 60 | // - can only update cmp descriptor immidiately after opening the FIRST ft handle for this ft and before |
| 61 | // ANY operations. to update the cmp descriptor after any operations have already happened, all handles |
| 62 | // and transactions must close and reopen before the change, then you can update the cmp descriptor |
| 63 | void toku_ft_change_descriptor(FT_HANDLE t, const DBT* old_descriptor, const DBT* new_descriptor, bool do_log, TOKUTXN txn, bool update_cmp_descriptor); |
| 64 | uint32_t toku_serialize_descriptor_size(DESCRIPTOR desc); |
| 65 | |
| 66 | void toku_ft_handle_create(FT_HANDLE *ft); |
| 67 | void toku_ft_set_flags(FT_HANDLE, unsigned int flags); |
| 68 | void toku_ft_get_flags(FT_HANDLE, unsigned int *flags); |
| 69 | void toku_ft_handle_set_nodesize(FT_HANDLE, unsigned int nodesize); |
| 70 | void toku_ft_handle_get_nodesize(FT_HANDLE, unsigned int *nodesize); |
| 71 | void toku_ft_get_maximum_advised_key_value_lengths(unsigned int *klimit, unsigned int *vlimit); |
| 72 | void toku_ft_handle_set_basementnodesize(FT_HANDLE, unsigned int basementnodesize); |
| 73 | void toku_ft_handle_get_basementnodesize(FT_HANDLE, unsigned int *basementnodesize); |
| 74 | void toku_ft_handle_set_compression_method(FT_HANDLE, enum toku_compression_method); |
| 75 | void toku_ft_handle_get_compression_method(FT_HANDLE, enum toku_compression_method *); |
| 76 | void toku_ft_handle_set_fanout(FT_HANDLE, unsigned int fanout); |
| 77 | void toku_ft_handle_get_fanout(FT_HANDLE, unsigned int *fanout); |
| 78 | int toku_ft_handle_set_memcmp_magic(FT_HANDLE, uint8_t magic); |
| 79 | |
| 80 | void toku_ft_set_bt_compare(FT_HANDLE ft_handle, ft_compare_func cmp_func); |
| 81 | const toku::comparator &toku_ft_get_comparator(FT_HANDLE ft_handle); |
| 82 | |
| 83 | typedef void (*on_redirect_callback)(FT_HANDLE ft_handle, void *); |
| 84 | void toku_ft_set_redirect_callback(FT_HANDLE ft_handle, on_redirect_callback cb, void *); |
| 85 | |
| 86 | // How updates (update/insert/deletes) work: |
| 87 | // There are two flavers of upsertdels: Singleton and broadcast. |
| 88 | // When a singleton upsertdel message arrives it contains a key and an extra DBT. |
| 89 | // |
| 90 | // At the YDB layer, the function looks like |
| 91 | // |
| 92 | // int (*update_function)(DB*, DB_TXN*, const DBT *key, const DBT *old_val, const DBT *extra, |
| 93 | // void (*set_val)(const DBT *new_val, void *set_extra), void *set_extra); |
| 94 | // |
| 95 | // And there are two DB functions |
| 96 | // |
| 97 | // int DB->update(DB *, DB_TXN *, const DBT *key, const DBT *extra); |
| 98 | // Effect: |
| 99 | // If there is a key-value pair visible to the txn with value old_val then the system calls |
| 100 | // update_function(DB, key, old_val, extra, set_val, set_extra) |
| 101 | // where set_val and set_extra are a function and a void* provided by the system. |
| 102 | // The update_function can do one of two things: |
| 103 | // a) call set_val(new_val, set_extra) |
| 104 | // which has the effect of doing DB->put(db, txn, key, new_val, 0) |
| 105 | // overwriting the old value. |
| 106 | // b) Return DB_DELETE (a new return code) |
| 107 | // c) Return 0 (success) without calling set_val, which leaves the old value unchanged. |
| 108 | // If there is no such key-value pair visible to the txn, then the system calls |
| 109 | // update_function(DB, key, NULL, extra, set_val, set_extra) |
| 110 | // and the update_function can do one of the same three things. |
| 111 | // Implementation notes: Update acquires a write lock (just as DB->put |
| 112 | // does). This function works by sending a UPDATE message containing |
| 113 | // the key and extra. |
| 114 | // |
| 115 | // int DB->update_broadcast(DB *, DB_TXN*, const DBT *extra); |
| 116 | // Effect: This has the same effect as building a cursor that walks |
| 117 | // through the DB, calling DB->update() on every key that the cursor |
| 118 | // finds. |
| 119 | // Implementation note: Acquires a write lock on the entire database. |
| 120 | // This function works by sending an BROADCAST-UPDATE message containing |
| 121 | // the key and the extra. |
| 122 | typedef int (*ft_update_func)(DB *db, const DBT *key, const DBT *old_val, const DBT *, |
| 123 | void (*set_val)(const DBT *new_val, void *), |
| 124 | void *); |
| 125 | void toku_ft_set_update(FT_HANDLE ft_h, ft_update_func update_fun); |
| 126 | |
| 127 | int toku_ft_handle_open(FT_HANDLE, const char *fname_in_env, |
| 128 | int is_create, int only_create, CACHETABLE ct, TOKUTXN txn) __attribute__ ((warn_unused_result)); |
| 129 | int toku_ft_handle_open_recovery(FT_HANDLE, const char *fname_in_env, int is_create, int only_create, CACHETABLE ct, TOKUTXN txn, |
| 130 | FILENUM use_filenum, LSN max_acceptable_lsn) __attribute__ ((warn_unused_result)); |
| 131 | |
| 132 | // clone an ft handle. the cloned handle has a new dict_id but refers to the same fractal tree |
| 133 | int toku_ft_handle_clone(FT_HANDLE *cloned_ft_handle, FT_HANDLE ft_handle, TOKUTXN txn); |
| 134 | |
| 135 | // close an ft handle during normal operation. the underlying ft may or may not close, |
| 136 | // depending if there are still references. an lsn for this close will come from the logger. |
| 137 | void toku_ft_handle_close(FT_HANDLE ft_handle); |
| 138 | // close an ft handle during recovery. the underlying ft must close, and will use the given lsn. |
| 139 | void toku_ft_handle_close_recovery(FT_HANDLE ft_handle, LSN oplsn); |
| 140 | |
| 141 | // At the ydb layer, a DICTIONARY_ID uniquely identifies an open dictionary. |
| 142 | // With the introduction of the loader (ticket 2216), it is possible for the file that holds |
| 143 | // an open dictionary to change, so these are now separate and independent unique identifiers (see FILENUM) |
| 144 | struct DICTIONARY_ID { |
| 145 | uint64_t dictid; |
| 146 | }; |
| 147 | static const DICTIONARY_ID DICTIONARY_ID_NONE = { .dictid = 0 }; |
| 148 | |
| 149 | int |
| 150 | toku_ft_handle_open_with_dict_id( |
| 151 | FT_HANDLE ft_h, |
| 152 | const char *fname_in_env, |
| 153 | int is_create, |
| 154 | int only_create, |
| 155 | CACHETABLE cachetable, |
| 156 | TOKUTXN txn, |
| 157 | DICTIONARY_ID use_dictionary_id |
| 158 | ) __attribute__ ((warn_unused_result)); |
| 159 | |
| 160 | // Effect: Insert a key and data pair into an ft |
| 161 | void toku_ft_insert (FT_HANDLE ft_h, DBT *k, DBT *v, TOKUTXN txn); |
| 162 | |
| 163 | // Returns: 0 if the key was inserted, DB_KEYEXIST if the key already exists |
| 164 | int toku_ft_insert_unique(FT_HANDLE ft, DBT *k, DBT *v, TOKUTXN txn, bool do_logging); |
| 165 | |
| 166 | // Effect: Optimize the ft |
| 167 | void toku_ft_optimize (FT_HANDLE ft_h); |
| 168 | |
| 169 | // Effect: Insert a key and data pair into an ft if the oplsn is newer than the ft's lsn. This function is called during recovery. |
| 170 | void toku_ft_maybe_insert (FT_HANDLE ft_h, DBT *k, DBT *v, TOKUTXN txn, bool oplsn_valid, LSN oplsn, bool do_logging, enum ft_msg_type type); |
| 171 | |
| 172 | // Effect: Send an update message into an ft. This function is called |
| 173 | // during recovery. |
| 174 | void toku_ft_maybe_update(FT_HANDLE ft_h, const DBT *key, const DBT *, TOKUTXN txn, bool oplsn_valid, LSN oplsn, bool do_logging); |
| 175 | |
| 176 | // Effect: Send a broadcasting update message into an ft. This function |
| 177 | // is called during recovery. |
| 178 | void toku_ft_maybe_update_broadcast(FT_HANDLE ft_h, const DBT *, TOKUTXN txn, bool oplsn_valid, LSN oplsn, bool do_logging, bool is_resetting_op); |
| 179 | |
| 180 | void toku_ft_load_recovery(TOKUTXN txn, FILENUM old_filenum, char const * new_iname, int do_fsync, int do_log, LSN *load_lsn); |
| 181 | void toku_ft_load(FT_HANDLE ft_h, TOKUTXN txn, char const * new_iname, int do_fsync, LSN *get_lsn); |
| 182 | void toku_ft_hot_index_recovery(TOKUTXN txn, FILENUMS filenums, int do_fsync, int do_log, LSN *hot_index_lsn); |
| 183 | void toku_ft_hot_index(FT_HANDLE ft_h, TOKUTXN txn, FILENUMS filenums, int do_fsync, LSN *lsn); |
| 184 | |
| 185 | void toku_ft_log_put_multiple (TOKUTXN txn, FT_HANDLE src_ft, FT_HANDLE *fts, uint32_t num_fts, const DBT *key, const DBT *val); |
| 186 | void toku_ft_log_put (TOKUTXN txn, FT_HANDLE ft_h, const DBT *key, const DBT *val); |
| 187 | void toku_ft_log_del_multiple (TOKUTXN txn, FT_HANDLE src_ft, FT_HANDLE *fts, uint32_t num_fts, const DBT *key, const DBT *val); |
| 188 | void toku_ft_log_del (TOKUTXN txn, FT_HANDLE ft_h, const DBT *key); |
| 189 | |
| 190 | // Effect: Delete a key from an ft |
| 191 | void toku_ft_delete (FT_HANDLE ft_h, DBT *k, TOKUTXN txn); |
| 192 | |
| 193 | // Effect: Delete a key from an ft if the oplsn is newer than the ft lsn. This function is called during recovery. |
| 194 | void toku_ft_maybe_delete (FT_HANDLE ft_h, DBT *k, TOKUTXN txn, bool oplsn_valid, LSN oplsn, bool do_logging); |
| 195 | |
| 196 | TXNID toku_ft_get_oldest_referenced_xid_estimate(FT_HANDLE ft_h); |
| 197 | struct txn_manager *toku_ft_get_txn_manager(FT_HANDLE ft_h); |
| 198 | |
| 199 | struct txn_gc_info; |
| 200 | void toku_ft_send_insert(FT_HANDLE ft_h, DBT *key, DBT *val, XIDS xids, enum ft_msg_type type, txn_gc_info *gc_info); |
| 201 | void toku_ft_send_delete(FT_HANDLE ft_h, DBT *key, XIDS xids, txn_gc_info *gc_info); |
| 202 | void toku_ft_send_commit_any(FT_HANDLE ft_h, DBT *key, XIDS xids, txn_gc_info *gc_info); |
| 203 | |
| 204 | int toku_close_ft_handle_nolsn (FT_HANDLE, char **error_string) __attribute__ ((warn_unused_result)); |
| 205 | |
| 206 | int toku_dump_ft (FILE *,FT_HANDLE ft_h) __attribute__ ((warn_unused_result)); |
| 207 | |
| 208 | extern int toku_ft_debug_mode; |
| 209 | int toku_verify_ft (FT_HANDLE ft_h) __attribute__ ((warn_unused_result)); |
| 210 | int toku_verify_ft_with_progress (FT_HANDLE ft_h, int (*progress_callback)(void *, float progress), void *, int verbose, int keep_going) __attribute__ ((warn_unused_result)); |
| 211 | |
| 212 | int toku_ft_recount_rows( |
| 213 | FT_HANDLE ft, |
| 214 | int (*progress_callback)( |
| 215 | uint64_t count, |
| 216 | uint64_t deleted, |
| 217 | void* ), |
| 218 | void* ); |
| 219 | |
| 220 | |
| 221 | DICTIONARY_ID toku_ft_get_dictionary_id(FT_HANDLE); |
| 222 | |
| 223 | enum ft_flags { |
| 224 | //TOKU_DB_DUP = (1<<0), //Obsolete #2862 |
| 225 | //TOKU_DB_DUPSORT = (1<<1), //Obsolete #2862 |
| 226 | TOKU_DB_KEYCMP_BUILTIN = (1<<2), |
| 227 | TOKU_DB_VALCMP_BUILTIN_13 = (1<<3), |
| 228 | }; |
| 229 | |
| 230 | void toku_ft_keyrange(FT_HANDLE ft_h, DBT *key, uint64_t *less, uint64_t *equal, uint64_t *greater); |
| 231 | void toku_ft_keysrange(FT_HANDLE ft_h, DBT* key_left, DBT* key_right, uint64_t *less_p, uint64_t* equal_left_p, uint64_t* middle_p, uint64_t* equal_right_p, uint64_t* greater_p, bool* middle_3_exact_p); |
| 232 | |
| 233 | int toku_ft_get_key_after_bytes(FT_HANDLE ft_h, const DBT *start_key, uint64_t skip_len, void (*callback)(const DBT *end_key, uint64_t actually_skipped, void *), void *); |
| 234 | |
| 235 | struct ftstat64_s { |
| 236 | uint64_t nkeys; /* estimate how many unique keys (even when flattened this may be an estimate) */ |
| 237 | uint64_t ndata; /* estimate the number of pairs (exact when flattened and committed) */ |
| 238 | uint64_t dsize; /* estimate the sum of the sizes of the pairs (exact when flattened and committed) */ |
| 239 | uint64_t fsize; /* the size of the underlying file */ |
| 240 | uint64_t ffree; /* Number of free bytes in the underlying file */ |
| 241 | uint64_t create_time_sec; /* creation time in seconds. */ |
| 242 | uint64_t modify_time_sec; /* time of last serialization, in seconds. */ |
| 243 | uint64_t verify_time_sec; /* time of last verification, in seconds */ |
| 244 | }; |
| 245 | |
| 246 | void toku_ft_handle_stat64 (FT_HANDLE, TOKUTXN, struct ftstat64_s *stat); |
| 247 | |
| 248 | struct ftinfo64 { |
| 249 | uint64_t num_blocks_allocated; // number of blocks in the blocktable |
| 250 | uint64_t num_blocks_in_use; // number of blocks in use by most recent checkpoint |
| 251 | uint64_t size_allocated; // sum of sizes of blocks in blocktable |
| 252 | uint64_t size_in_use; // sum of sizes of blocks in use by most recent checkpoint |
| 253 | }; |
| 254 | |
| 255 | void toku_ft_handle_get_fractal_tree_info64(FT_HANDLE, struct ftinfo64 *); |
| 256 | |
| 257 | int toku_ft_handle_iterate_fractal_tree_block_map(FT_HANDLE, int (*)(uint64_t,int64_t,int64_t,int64_t,int64_t,void*), void *); |
| 258 | |
| 259 | int toku_ft_layer_init(void) __attribute__ ((warn_unused_result)); |
| 260 | void toku_ft_open_close_lock(void); |
| 261 | void toku_ft_open_close_unlock(void); |
| 262 | void toku_ft_layer_destroy(void); |
| 263 | void toku_ft_serialize_layer_init(void); |
| 264 | void toku_ft_serialize_layer_destroy(void); |
| 265 | |
| 266 | void toku_maybe_truncate_file (int fd, uint64_t size_used, uint64_t expected_size, uint64_t *new_size); |
| 267 | // Effect: truncate file if overallocated by at least 32MiB |
| 268 | |
| 269 | void toku_maybe_preallocate_in_file (int fd, int64_t size, int64_t expected_size, int64_t *new_size); |
| 270 | // Effect: make the file bigger by either doubling it or growing by 16MiB whichever is less, until it is at least size |
| 271 | // Return 0 on success, otherwise an error number. |
| 272 | |
| 273 | int toku_ft_get_fragmentation(FT_HANDLE ft_h, TOKU_DB_FRAGMENTATION report) __attribute__ ((warn_unused_result)); |
| 274 | |
| 275 | bool toku_ft_is_empty_fast (FT_HANDLE ft_h) __attribute__ ((warn_unused_result)); |
| 276 | // Effect: Return true if there are no messages or leaf entries in the tree. If so, it's empty. If there are messages or leaf entries, we say it's not empty |
| 277 | // even though if we were to optimize the tree it might turn out that they are empty. |
| 278 | |
| 279 | int toku_ft_strerror_r(int error, char *buf, size_t buflen); |
| 280 | // Effect: LIke the XSI-compliant strerorr_r, extended to db_strerror(). |
| 281 | // If error>=0 then the result is to do strerror_r(error, buf, buflen), that is fill buf with a descriptive error message. |
| 282 | // If error<0 then return a PerconaFT-specific error code. For unknown cases, we return -1 and set errno=EINVAL, even for cases that *should* be known. (Not all DB errors are known by this function which is a bug.) |
| 283 | |
| 284 | extern bool garbage_collection_debug; |
| 285 | |
| 286 | // This is a poor place to put global options like these. |
| 287 | void toku_ft_set_direct_io(bool direct_io_on); |
| 288 | void toku_ft_set_compress_buffers_before_eviction(bool compress_buffers); |
| 289 | |
| 290 | void toku_note_deserialized_basement_node(bool fixed_key_size); |
| 291 | |
| 292 | // Creates all directories for the path if necessary, |
| 293 | // returns true if all dirs are created successfully or |
| 294 | // all dirs exist, false otherwise. |
| 295 | bool toku_create_subdirs_if_needed(const char* path); |
| 296 | |