| 1 | /* Copyright (C) 2006 MySQL AB & MySQL Finland AB & TCX DataKonsult AB |
| 2 | |
| 3 | This program is free software; you can redistribute it and/or modify |
| 4 | it under the terms of the GNU General Public License as published by |
| 5 | the Free Software Foundation; version 2 of the License. |
| 6 | |
| 7 | This program is distributed in the hope that it will be useful, |
| 8 | but WITHOUT ANY WARRANTY; without even the implied warranty of |
| 9 | MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the |
| 10 | GNU General Public License for more details. |
| 11 | |
| 12 | You should have received a copy of the GNU General Public License |
| 13 | along with this program; if not, write to the Free Software |
| 14 | Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02111-1301 USA */ |
| 15 | |
| 16 | /* |
| 17 | Preload indexes into key cache |
| 18 | */ |
| 19 | |
| 20 | #include "maria_def.h" |
| 21 | |
| 22 | |
| 23 | /* |
| 24 | Preload pages of the index file for a table into the key cache |
| 25 | |
| 26 | SYNOPSIS |
| 27 | maria_preload() |
| 28 | info open table |
| 29 | map map of indexes to preload into key cache |
| 30 | ignore_leaves only non-leaves pages are to be preloaded |
| 31 | |
| 32 | RETURN VALUE |
| 33 | 0 if a success. error code - otherwise. |
| 34 | |
| 35 | NOTES. |
| 36 | At present pages for all indexes are preloaded. |
| 37 | In future only pages for indexes specified in the key_map parameter |
| 38 | of the table will be preloaded. |
| 39 | We don't yet use preload_buff_size (we read page after page). |
| 40 | */ |
| 41 | |
| 42 | int maria_preload(MARIA_HA *info, ulonglong key_map, my_bool ignore_leaves) |
| 43 | { |
| 44 | ulong block_length= 0; |
| 45 | uchar *buff; |
| 46 | MARIA_SHARE* share= info->s; |
| 47 | uint keynr; |
| 48 | my_off_t key_file_length= share->state.state.key_file_length; |
| 49 | pgcache_page_no_t page_no, page_no_max; |
| 50 | PAGECACHE_BLOCK_LINK *page_link; |
| 51 | DBUG_ENTER("maria_preload" ); |
| 52 | |
| 53 | if (!share->state.header.keys || !maria_is_any_key_active(key_map) || |
| 54 | (key_file_length == share->base.keystart)) |
| 55 | DBUG_RETURN(0); |
| 56 | |
| 57 | block_length= share->pagecache->block_size; |
| 58 | |
| 59 | if (!(buff= (uchar *) my_malloc(block_length, MYF(MY_WME)))) |
| 60 | DBUG_RETURN(my_errno= HA_ERR_OUT_OF_MEM); |
| 61 | |
| 62 | if (flush_pagecache_blocks(share->pagecache, &share->kfile, FLUSH_RELEASE)) |
| 63 | goto err; |
| 64 | |
| 65 | /* |
| 66 | Currently when we come here all other open instances of the table have |
| 67 | been closed, and we flushed all pages of our own instance, so there |
| 68 | cannot be any page of this table in the cache. Thus my_pread() would be |
| 69 | safe. But in the future, we will allow more concurrency during |
| 70 | preloading, so we use pagecache_read() instead of my_pread() because we |
| 71 | observed that on some Linux, concurrent pread() and pwrite() (which |
| 72 | could be from a page eviction by another thread) to the same page can |
| 73 | make pread() see an half-written page. |
| 74 | In this future, we should find a way to read state.key_file_length |
| 75 | reliably, handle concurrent shrinks (delete_all_rows()) etc. |
| 76 | */ |
| 77 | for ((page_no= share->base.keystart / block_length), |
| 78 | (page_no_max= key_file_length / block_length); |
| 79 | page_no < page_no_max; page_no++) |
| 80 | { |
| 81 | /** |
| 82 | @todo instead of reading pages one by one we could have a call |
| 83 | pagecache_read_several_pages() which does a single my_pread() for many |
| 84 | consecutive pages (like the my_pread() in mi_preload()). |
| 85 | */ |
| 86 | if (pagecache_read(share->pagecache, &share->kfile, page_no, |
| 87 | DFLT_INIT_HITS, buff, share->page_type, |
| 88 | PAGECACHE_LOCK_WRITE, &page_link) == NULL) |
| 89 | goto err; |
| 90 | keynr= _ma_get_keynr(share, buff); |
| 91 | if (((ignore_leaves && !_ma_test_if_nod(share, buff)) || |
| 92 | keynr == MARIA_DELETE_KEY_NR || |
| 93 | !(key_map & ((ulonglong) 1 << keynr))) && |
| 94 | (pagecache_pagelevel(page_link) == DFLT_INIT_HITS)) |
| 95 | { |
| 96 | /* |
| 97 | This page is not interesting, and (last condition above) we are the |
| 98 | ones who put it in the cache, so nobody else is interested in it. |
| 99 | */ |
| 100 | if (pagecache_delete_by_link(share->pagecache, page_link, |
| 101 | PAGECACHE_LOCK_LEFT_WRITELOCKED, FALSE)) |
| 102 | goto err; |
| 103 | } |
| 104 | else /* otherwise it stays in cache: */ |
| 105 | pagecache_unlock_by_link(share->pagecache, page_link, |
| 106 | PAGECACHE_LOCK_WRITE_UNLOCK, PAGECACHE_UNPIN, |
| 107 | LSN_IMPOSSIBLE, LSN_IMPOSSIBLE, FALSE, FALSE); |
| 108 | } |
| 109 | |
| 110 | my_free(buff); |
| 111 | DBUG_RETURN(0); |
| 112 | |
| 113 | err: |
| 114 | my_free(buff); |
| 115 | DBUG_RETURN(my_errno= errno); |
| 116 | } |
| 117 | |