| 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 | #include <toku_stdint.h> |
| 40 | |
| 41 | #include "ft/serialize/block_table.h" |
| 42 | #include "ft/ft.h" |
| 43 | #include "ft/logger/log-internal.h" |
| 44 | #include "ft/txn/rollback-ct-callbacks.h" |
| 45 | |
| 46 | static void rollback_unpin_remove_callback(CACHEKEY* cachekey, bool for_checkpoint, void* ) { |
| 47 | FT CAST_FROM_VOIDP(ft, extra); |
| 48 | ft->blocktable.free_blocknum(cachekey, ft, for_checkpoint); |
| 49 | } |
| 50 | |
| 51 | void toku_rollback_log_unpin_and_remove(TOKUTXN txn, ROLLBACK_LOG_NODE log) { |
| 52 | int r; |
| 53 | CACHEFILE cf = txn->logger->rollback_cachefile; |
| 54 | FT CAST_FROM_VOIDP(ft, toku_cachefile_get_userdata(cf)); |
| 55 | r = toku_cachetable_unpin_and_remove (cf, log->ct_pair, rollback_unpin_remove_callback, ft); |
| 56 | assert(r == 0); |
| 57 | } |
| 58 | |
| 59 | int |
| 60 | toku_find_xid_by_xid (const TXNID &xid, const TXNID &xidfind) { |
| 61 | if (xid<xidfind) return -1; |
| 62 | if (xid>xidfind) return +1; |
| 63 | return 0; |
| 64 | } |
| 65 | |
| 66 | // TODO: fix this name |
| 67 | // toku_rollback_malloc |
| 68 | void *toku_malloc_in_rollback(ROLLBACK_LOG_NODE log, size_t size) { |
| 69 | return log->rollentry_arena.malloc_from_arena(size); |
| 70 | } |
| 71 | |
| 72 | // TODO: fix this name |
| 73 | // toku_rollback_memdup |
| 74 | void *toku_memdup_in_rollback(ROLLBACK_LOG_NODE log, const void *v, size_t len) { |
| 75 | void *r = toku_malloc_in_rollback(log, len); |
| 76 | memcpy(r, v, len); |
| 77 | return r; |
| 78 | } |
| 79 | |
| 80 | static inline PAIR_ATTR make_rollback_pair_attr(long size) { |
| 81 | PAIR_ATTR result={ |
| 82 | .size = size, |
| 83 | .nonleaf_size = 0, |
| 84 | .leaf_size = 0, |
| 85 | .rollback_size = size, |
| 86 | .cache_pressure_size = 0, |
| 87 | .is_valid = true |
| 88 | }; |
| 89 | return result; |
| 90 | } |
| 91 | |
| 92 | PAIR_ATTR |
| 93 | rollback_memory_size(ROLLBACK_LOG_NODE log) { |
| 94 | size_t size = sizeof(*log); |
| 95 | size += log->rollentry_arena.total_footprint(); |
| 96 | return make_rollback_pair_attr(size); |
| 97 | } |
| 98 | |
| 99 | static void toku_rollback_node_save_ct_pair(CACHEKEY UU(key), void *value_data, PAIR p) { |
| 100 | ROLLBACK_LOG_NODE CAST_FROM_VOIDP(log, value_data); |
| 101 | log->ct_pair = p; |
| 102 | } |
| 103 | |
| 104 | // |
| 105 | // initializes an empty rollback log node |
| 106 | // Does not touch the blocknum, that is the |
| 107 | // responsibility of the caller |
| 108 | // |
| 109 | void rollback_empty_log_init(ROLLBACK_LOG_NODE log) { |
| 110 | // Having a txnid set to TXNID_NONE is how we determine if the |
| 111 | // rollback log node is empty or in use. |
| 112 | log->txnid.parent_id64 = TXNID_NONE; |
| 113 | log->txnid.child_id64 = TXNID_NONE; |
| 114 | |
| 115 | log->layout_version = FT_LAYOUT_VERSION; |
| 116 | log->layout_version_original = FT_LAYOUT_VERSION; |
| 117 | log->layout_version_read_from_disk = FT_LAYOUT_VERSION; |
| 118 | log->dirty = true; |
| 119 | log->sequence = 0; |
| 120 | log->previous = make_blocknum(0); |
| 121 | log->oldest_logentry = NULL; |
| 122 | log->newest_logentry = NULL; |
| 123 | log->rollentry_arena.create(0); |
| 124 | log->rollentry_resident_bytecount = 0; |
| 125 | } |
| 126 | |
| 127 | static void rollback_initialize_for_txn( |
| 128 | ROLLBACK_LOG_NODE log, |
| 129 | TOKUTXN txn, |
| 130 | BLOCKNUM previous |
| 131 | ) |
| 132 | { |
| 133 | log->txnid = txn->txnid; |
| 134 | log->sequence = txn->roll_info.num_rollback_nodes++; |
| 135 | log->previous = previous; |
| 136 | log->oldest_logentry = NULL; |
| 137 | log->newest_logentry = NULL; |
| 138 | log->rollentry_arena.create(1024); |
| 139 | log->rollentry_resident_bytecount = 0; |
| 140 | log->dirty = true; |
| 141 | } |
| 142 | |
| 143 | // TODO: fix this name |
| 144 | void make_rollback_log_empty(ROLLBACK_LOG_NODE log) { |
| 145 | log->rollentry_arena.destroy(); |
| 146 | rollback_empty_log_init(log); |
| 147 | } |
| 148 | |
| 149 | // create and pin a new rollback log node. chain it to the other rollback nodes |
| 150 | // by providing a previous blocknum and assigning the new rollback log |
| 151 | // node the next sequence number |
| 152 | static void rollback_log_create ( |
| 153 | TOKUTXN txn, |
| 154 | BLOCKNUM previous, |
| 155 | ROLLBACK_LOG_NODE *result |
| 156 | ) |
| 157 | { |
| 158 | ROLLBACK_LOG_NODE XMALLOC(log); |
| 159 | rollback_empty_log_init(log); |
| 160 | |
| 161 | CACHEFILE cf = txn->logger->rollback_cachefile; |
| 162 | FT CAST_FROM_VOIDP(ft, toku_cachefile_get_userdata(cf)); |
| 163 | rollback_initialize_for_txn(log, txn, previous); |
| 164 | ft->blocktable.allocate_blocknum(&log->blocknum, ft); |
| 165 | const uint32_t hash = toku_cachetable_hash(ft->cf, log->blocknum); |
| 166 | *result = log; |
| 167 | toku_cachetable_put(cf, log->blocknum, hash, |
| 168 | log, rollback_memory_size(log), |
| 169 | get_write_callbacks_for_rollback_log(ft), |
| 170 | toku_rollback_node_save_ct_pair); |
| 171 | txn->roll_info.current_rollback = log->blocknum; |
| 172 | } |
| 173 | |
| 174 | void toku_rollback_log_unpin(TOKUTXN txn, ROLLBACK_LOG_NODE log) { |
| 175 | int r; |
| 176 | CACHEFILE cf = txn->logger->rollback_cachefile; |
| 177 | r = toku_cachetable_unpin( |
| 178 | cf, |
| 179 | log->ct_pair, |
| 180 | (enum cachetable_dirty)log->dirty, |
| 181 | rollback_memory_size(log) |
| 182 | ); |
| 183 | assert(r == 0); |
| 184 | } |
| 185 | |
| 186 | //Requires: log is pinned |
| 187 | // log is current |
| 188 | //After: |
| 189 | // Maybe there is no current after (if it spilled) |
| 190 | void toku_maybe_spill_rollbacks(TOKUTXN txn, ROLLBACK_LOG_NODE log) { |
| 191 | if (log->rollentry_resident_bytecount > txn->logger->write_block_size) { |
| 192 | assert(log->blocknum.b == txn->roll_info.current_rollback.b); |
| 193 | //spill |
| 194 | if (!txn_has_spilled_rollback_logs(txn)) { |
| 195 | //First spilled. Copy to head. |
| 196 | txn->roll_info.spilled_rollback_head = txn->roll_info.current_rollback; |
| 197 | } |
| 198 | //Unconditionally copy to tail. Old tail does not need to be cached anymore. |
| 199 | txn->roll_info.spilled_rollback_tail = txn->roll_info.current_rollback; |
| 200 | |
| 201 | txn->roll_info.current_rollback = ROLLBACK_NONE; |
| 202 | } |
| 203 | } |
| 204 | |
| 205 | int find_filenum (const FT &h, const FT &hfind); |
| 206 | int find_filenum (const FT &h, const FT &hfind) { |
| 207 | FILENUM fnum = toku_cachefile_filenum(h->cf); |
| 208 | FILENUM fnumfind = toku_cachefile_filenum(hfind->cf); |
| 209 | if (fnum.fileid<fnumfind.fileid) return -1; |
| 210 | if (fnum.fileid>fnumfind.fileid) return +1; |
| 211 | return 0; |
| 212 | } |
| 213 | |
| 214 | //Notify a transaction that it has touched an ft. |
| 215 | void toku_txn_maybe_note_ft (TOKUTXN txn, FT ft) { |
| 216 | toku_txn_lock(txn); |
| 217 | FT ftv; |
| 218 | uint32_t idx; |
| 219 | int r = txn->open_fts.find_zero<FT, find_filenum>(ft, &ftv, &idx); |
| 220 | if (r == 0) { |
| 221 | // already there |
| 222 | assert(ftv == ft); |
| 223 | goto exit; |
| 224 | } |
| 225 | r = txn->open_fts.insert_at(ft, idx); |
| 226 | assert_zero(r); |
| 227 | // TODO(leif): if there's anything that locks the reflock and then |
| 228 | // the txn lock, this may deadlock, because it grabs the reflock. |
| 229 | toku_ft_add_txn_ref(ft); |
| 230 | exit: |
| 231 | toku_txn_unlock(txn); |
| 232 | } |
| 233 | |
| 234 | // Return the number of bytes that went into the rollback data structure (the uncompressed count if there is compression) |
| 235 | int toku_logger_txn_rollback_stats(TOKUTXN txn, struct txn_stat *txn_stat) |
| 236 | { |
| 237 | toku_txn_lock(txn); |
| 238 | txn_stat->rollback_raw_count = txn->roll_info.rollentry_raw_count; |
| 239 | txn_stat->rollback_num_entries = txn->roll_info.num_rollentries; |
| 240 | toku_txn_unlock(txn); |
| 241 | return 0; |
| 242 | } |
| 243 | |
| 244 | void toku_maybe_prefetch_previous_rollback_log(TOKUTXN txn, ROLLBACK_LOG_NODE log) { |
| 245 | //Currently processing 'log'. Prefetch the next (previous) log node. |
| 246 | |
| 247 | BLOCKNUM name = log->previous; |
| 248 | int r = 0; |
| 249 | if (name.b != ROLLBACK_NONE.b) { |
| 250 | CACHEFILE cf = txn->logger->rollback_cachefile; |
| 251 | uint32_t hash = toku_cachetable_hash(cf, name); |
| 252 | FT CAST_FROM_VOIDP(h, toku_cachefile_get_userdata(cf)); |
| 253 | bool doing_prefetch = false; |
| 254 | r = toku_cachefile_prefetch(cf, name, hash, |
| 255 | get_write_callbacks_for_rollback_log(h), |
| 256 | toku_rollback_fetch_callback, |
| 257 | toku_rollback_pf_req_callback, |
| 258 | toku_rollback_pf_callback, |
| 259 | h, |
| 260 | &doing_prefetch); |
| 261 | assert(r == 0); |
| 262 | } |
| 263 | } |
| 264 | |
| 265 | void toku_rollback_verify_contents(ROLLBACK_LOG_NODE log, |
| 266 | TXNID_PAIR txnid, uint64_t sequence) |
| 267 | { |
| 268 | assert(log->txnid.parent_id64 == txnid.parent_id64); |
| 269 | assert(log->txnid.child_id64 == txnid.child_id64); |
| 270 | assert(log->sequence == sequence); |
| 271 | } |
| 272 | |
| 273 | void toku_get_and_pin_rollback_log(TOKUTXN txn, BLOCKNUM blocknum, ROLLBACK_LOG_NODE *log) { |
| 274 | void * value; |
| 275 | CACHEFILE cf = txn->logger->rollback_cachefile; |
| 276 | FT CAST_FROM_VOIDP(h, toku_cachefile_get_userdata(cf)); |
| 277 | uint32_t hash = toku_cachetable_hash(cf, blocknum); |
| 278 | int r = toku_cachetable_get_and_pin_with_dep_pairs(cf, blocknum, hash, |
| 279 | &value, NULL, |
| 280 | get_write_callbacks_for_rollback_log(h), |
| 281 | toku_rollback_fetch_callback, |
| 282 | toku_rollback_pf_req_callback, |
| 283 | toku_rollback_pf_callback, |
| 284 | PL_WRITE_CHEAP, // lock_type |
| 285 | h, |
| 286 | 0, NULL, NULL |
| 287 | ); |
| 288 | assert(r == 0); |
| 289 | ROLLBACK_LOG_NODE CAST_FROM_VOIDP(pinned_log, value); |
| 290 | assert(pinned_log->blocknum.b == blocknum.b); |
| 291 | *log = pinned_log; |
| 292 | } |
| 293 | |
| 294 | void toku_get_and_pin_rollback_log_for_new_entry (TOKUTXN txn, ROLLBACK_LOG_NODE *log) { |
| 295 | ROLLBACK_LOG_NODE pinned_log = NULL; |
| 296 | invariant(txn->state == TOKUTXN_LIVE || txn->state == TOKUTXN_PREPARING); // hot indexing may call this function for prepared transactions |
| 297 | if (txn_has_current_rollback_log(txn)) { |
| 298 | toku_get_and_pin_rollback_log(txn, txn->roll_info.current_rollback, &pinned_log); |
| 299 | toku_rollback_verify_contents(pinned_log, txn->txnid, txn->roll_info.num_rollback_nodes - 1); |
| 300 | } else { |
| 301 | // For each transaction, we try to acquire the first rollback log |
| 302 | // from the rollback log node cache, so that we avoid |
| 303 | // putting something new into the cachetable. However, |
| 304 | // if transaction has spilled rollbacks, that means we |
| 305 | // have already done a lot of work for this transaction, |
| 306 | // and subsequent rollback log nodes are created |
| 307 | // and put into the cachetable. The idea is for |
| 308 | // transactions that don't do a lot of work to (hopefully) |
| 309 | // get a rollback log node from a cache, as opposed to |
| 310 | // taking the more expensive route of creating a new one. |
| 311 | if (!txn_has_spilled_rollback_logs(txn)) { |
| 312 | txn->logger->rollback_cache.get_rollback_log_node(txn, &pinned_log); |
| 313 | if (pinned_log != NULL) { |
| 314 | rollback_initialize_for_txn( |
| 315 | pinned_log, |
| 316 | txn, |
| 317 | txn->roll_info.spilled_rollback_tail |
| 318 | ); |
| 319 | txn->roll_info.current_rollback = pinned_log->blocknum; |
| 320 | } |
| 321 | } |
| 322 | if (pinned_log == NULL) { |
| 323 | rollback_log_create(txn, txn->roll_info.spilled_rollback_tail, &pinned_log); |
| 324 | } |
| 325 | } |
| 326 | assert(pinned_log->txnid.parent_id64 == txn->txnid.parent_id64); |
| 327 | assert(pinned_log->txnid.child_id64 == txn->txnid.child_id64); |
| 328 | assert(pinned_log->blocknum.b != ROLLBACK_NONE.b); |
| 329 | *log = pinned_log; |
| 330 | } |
| 331 | |