X-Git-Url: http://git.indexdata.com/?a=blobdiff_plain;f=isamc%2Fisamc.c;h=bc81c16db76c147e05df80151f22b7e9fbf1bd1b;hb=e5e4f1de02396b2cfab488f8e5918853e96d6b53;hp=44155d22c9cf06166a3b96fa4db1a9e2479aaf65;hpb=801ef4b322574c793abdc1ba81dea2da62ca60bc;p=idzebra-moved-to-github.git diff --git a/isamc/isamc.c b/isamc/isamc.c index 44155d2..bc81c16 100644 --- a/isamc/isamc.c +++ b/isamc/isamc.c @@ -1,52 +1,26 @@ -/* - * Copyright (c) 1995-1998, Index Data. - * See the file LICENSE for details. - * Sebastian Hammer, Adam Dickmeiss - * - * $Log: isamc.c,v $ - * Revision 1.11 1998-03-13 15:30:50 adam - * New functions isc_block_used and isc_block_size. Fixed 'leak' - * in isc_alloc_block. - * - * Revision 1.10 1998/03/11 11:18:18 adam - * Changed the isc_merge to take into account the mfill (minimum-fill). - * - * Revision 1.9 1998/03/06 13:54:02 adam - * Fixed two nasty bugs in isc_merge. - * - * Revision 1.8 1997/09/17 12:19:20 adam - * Zebra version corresponds to YAZ version 1.4. - * Changed Zebra server so that it doesn't depend on global common_resource. - * - * Revision 1.7 1997/02/12 20:42:43 adam - * Bug fix: during isc_merge operations, some pages weren't marked dirty - * even though they should be. At this point the merge operation marks - * a page dirty if the previous page changed at all. A better approach is - * to mark it dirty if the last key written changed in previous page. - * - * Revision 1.6 1996/11/08 11:15:29 adam - * Number of keys in chain are stored in first block and the function - * to retrieve this information, isc_pp_num is implemented. - * - * Revision 1.5 1996/11/04 14:08:57 adam - * Optimized free block usage. - * - * Revision 1.4 1996/11/01 13:36:46 adam - * New element, max_blocks_mem, that control how many blocks of max size - * to store in memory during isc_merge. - * Function isc_merge now ignores delete/update of identical keys and - * the proper blocks are then non-dirty and not written in flush_blocks. - * - * Revision 1.3 1996/11/01 08:59:14 adam - * First version of isc_merge that supports update/delete. - * - * Revision 1.2 1996/10/29 16:44:56 adam - * Work on isc_merge. - * - * Revision 1.1 1996/10/29 13:40:48 adam - * First work. - * - */ +/* $Id: isamc.c,v 1.22 2002-08-02 19:26:56 adam Exp $ + Copyright (C) 1995,1996,1997,1998,1999,2000,2001,2002 + Index Data Aps + +This file is part of the Zebra server. + +Zebra is free software; you can redistribute it and/or modify it under +the terms of the GNU General Public License as published by the Free +Software Foundation; either version 2, or (at your option) any later +version. + +Zebra is distributed in the hope that it will be useful, but WITHOUT ANY +WARRANTY; without even the implied warranty of MERCHANTABILITY or +FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License +for more details. + +You should have received a copy of the GNU General Public License +along with Zebra; see the file LICENSE.zebra. If not, write to the +Free Software Foundation, 59 Temple Place - Suite 330, Boston, MA +02111-1307, USA. +*/ + + /* * TODO: @@ -57,44 +31,47 @@ #include #include -#include +#include #include "isamc-p.h" +static void flush_block (ISAMC is, int cat); static void release_fc (ISAMC is, int cat); static void init_fc (ISAMC is, int cat); +#define ISAMC_FREELIST_CHUNK 1 + #define SMALL_TEST 0 -ISAMC_M isc_getmethod (void) +void isc_getmethod (ISAMC_M m) { + static struct ISAMC_filecat_s def_cat[] = { #if SMALL_TEST - { 32, 28, 0, 3 }, - { 64, 54, 30, 0 }, + { 32, 28, 0, 3 }, + { 64, 54, 30, 0 }, #else - { 32, 28, 24, 20 }, - { 512, 490, 350, 20 }, - { 4096, 3950, 4200, 20 }, - {32768, 32000, 30000, 0 }, + { 32, 26, 20, 10 }, + { 128, 120, 100, 10 }, + { 512, 490, 350, 10 }, + { 2048, 1900, 1700, 10 }, + { 8192, 8000, 7900, 10 }, + { 32768, 32000, 31000, 0 }, #endif }; - ISAMC_M m = xmalloc (sizeof(*m)); m->filecat = def_cat; m->code_start = NULL; m->code_item = NULL; m->code_stop = NULL; + m->code_reset = NULL; m->compare_item = NULL; m->debug = 1; m->max_blocks_mem = 10; - - return m; } - ISAMC isc_open (BFiles bfs, const char *name, int writeflag, ISAMC_M method) { ISAMC is; @@ -102,9 +79,9 @@ ISAMC isc_open (BFiles bfs, const char *name, int writeflag, ISAMC_M method) int i = 0; int max_buf_size = 0; - is = xmalloc (sizeof(*is)); + is = (ISAMC) xmalloc (sizeof(*is)); - is->method = xmalloc (sizeof(*is->method)); + is->method = (ISAMC_M) xmalloc (sizeof(*is->method)); memcpy (is->method, method, sizeof(*method)); filecat = is->method->filecat; assert (filecat); @@ -131,10 +108,10 @@ ISAMC isc_open (BFiles bfs, const char *name, int writeflag, ISAMC_M method) logf (LOG_LOG, "isc: max_buf_size %d", max_buf_size); assert (is->no_files > 0); - is->files = xmalloc (sizeof(*is->files)*is->no_files); + is->files = (ISAMC_file) xmalloc (sizeof(*is->files)*is->no_files); if (writeflag) { - is->merge_buf = xmalloc (max_buf_size+256); + is->merge_buf = (char *) xmalloc (max_buf_size+256); memset (is->merge_buf, 0, max_buf_size+256); } else @@ -153,12 +130,23 @@ ISAMC isc_open (BFiles bfs, const char *name, int writeflag, ISAMC_M method) is->files[i].head.lastblock = 1; is->files[i].head.freelist = 0; } + is->files[i].alloc_entries_num = 0; + is->files[i].alloc_entries_max = + is->method->filecat[i].bsize / sizeof(int) - 1; + is->files[i].alloc_buf = (char *) + xmalloc (is->method->filecat[i].bsize); is->files[i].no_writes = 0; is->files[i].no_reads = 0; is->files[i].no_skip_writes = 0; is->files[i].no_allocated = 0; is->files[i].no_released = 0; is->files[i].no_remap = 0; + is->files[i].no_forward = 0; + is->files[i].no_backward = 0; + is->files[i].sum_forward = 0; + is->files[i].sum_backward = 0; + is->files[i].no_next = 0; + is->files[i].no_prev = 0; init_fc (is, i); } @@ -185,6 +173,22 @@ int isc_close (ISAMC is) int i; if (is->method->debug) + { + logf (LOG_LOG, "isc: next forw mid-f prev backw mid-b"); + for (i = 0; ino_files; i++) + logf (LOG_LOG, "isc:%8d%8d%8.1f%8d%8d%8.1f", + is->files[i].no_next, + is->files[i].no_forward, + is->files[i].no_forward ? + (double) is->files[i].sum_forward/is->files[i].no_forward + : 0.0, + is->files[i].no_prev, + is->files[i].no_backward, + is->files[i].no_backward ? + (double) is->files[i].sum_backward/is->files[i].no_backward + : 0.0); + } + if (is->method->debug) logf (LOG_LOG, "isc: writes reads skipped alloc released remap"); for (i = 0; ino_files; i++) { @@ -202,10 +206,12 @@ int isc_close (ISAMC is) is->files[i].no_released, is->files[i].no_remap); xfree (is->files[i].fc_list); + flush_block (is, i); bf_close (is->files[i].bf); } xfree (is->files); xfree (is->merge_buf); + xfree (is->method); xfree (is); return 0; } @@ -227,7 +233,7 @@ int isc_write_block (ISAMC is, int cat, int pos, char *src) int isc_write_dblock (ISAMC is, int cat, int pos, char *src, int nextpos, int offset) { - unsigned short size = offset + ISAMC_BLOCK_OFFSET_N; + ISAMC_BLOCK_SIZE size = offset + ISAMC_BLOCK_OFFSET_N; if (is->method->debug > 2) logf (LOG_LOG, "isc: write_dblock. size=%d nextpos=%d", (int) size, nextpos); @@ -237,6 +243,109 @@ int isc_write_dblock (ISAMC is, int cat, int pos, char *src, return isc_write_block (is, cat, pos, src); } +#if ISAMC_FREELIST_CHUNK +static void flush_block (ISAMC is, int cat) +{ + char *abuf = is->files[cat].alloc_buf; + int block = is->files[cat].head.freelist; + if (block && is->files[cat].alloc_entries_num) + { + memcpy (abuf, &is->files[cat].alloc_entries_num, sizeof(int)); + bf_write (is->files[cat].bf, block, 0, 0, abuf); + is->files[cat].alloc_entries_num = 0; + } + xfree (abuf); +} + +static int alloc_block (ISAMC is, int cat) +{ + int block = is->files[cat].head.freelist; + char *abuf = is->files[cat].alloc_buf; + + (is->files[cat].no_allocated)++; + + if (!block) + { + block = (is->files[cat].head.lastblock)++; /* no free list */ + is->files[cat].head_is_dirty = 1; + } + else + { + if (!is->files[cat].alloc_entries_num) /* read first time */ + { + bf_read (is->files[cat].bf, block, 0, 0, abuf); + memcpy (&is->files[cat].alloc_entries_num, abuf, + sizeof(is->files[cat].alloc_entries_num)); + assert (is->files[cat].alloc_entries_num > 0); + } + /* have some free blocks now */ + assert (is->files[cat].alloc_entries_num > 0); + is->files[cat].alloc_entries_num--; + if (!is->files[cat].alloc_entries_num) /* last one in block? */ + { + memcpy (&is->files[cat].head.freelist, abuf + sizeof(int), + sizeof(int)); + is->files[cat].head_is_dirty = 1; + + if (is->files[cat].head.freelist) + { + bf_read (is->files[cat].bf, is->files[cat].head.freelist, + 0, 0, abuf); + memcpy (&is->files[cat].alloc_entries_num, abuf, + sizeof(is->files[cat].alloc_entries_num)); + assert (is->files[cat].alloc_entries_num); + } + } + else + memcpy (&block, abuf + sizeof(int) + sizeof(int) * + is->files[cat].alloc_entries_num, sizeof(int)); + } + return block; +} + +static void release_block (ISAMC is, int cat, int pos) +{ + char *abuf = is->files[cat].alloc_buf; + int block = is->files[cat].head.freelist; + + (is->files[cat].no_released)++; + + if (block && !is->files[cat].alloc_entries_num) /* must read block */ + { + bf_read (is->files[cat].bf, block, 0, 0, abuf); + memcpy (&is->files[cat].alloc_entries_num, abuf, + sizeof(is->files[cat].alloc_entries_num)); + assert (is->files[cat].alloc_entries_num > 0); + } + assert (is->files[cat].alloc_entries_num <= is->files[cat].alloc_entries_max); + if (is->files[cat].alloc_entries_num == is->files[cat].alloc_entries_max) + { + assert (block); + memcpy (abuf, &is->files[cat].alloc_entries_num, sizeof(int)); + bf_write (is->files[cat].bf, block, 0, 0, abuf); + is->files[cat].alloc_entries_num = 0; + } + if (!is->files[cat].alloc_entries_num) /* make new buffer? */ + { + memcpy (abuf + sizeof(int), &block, sizeof(int)); + is->files[cat].head.freelist = pos; + is->files[cat].head_is_dirty = 1; + } + else + { + memcpy (abuf + sizeof(int) + + is->files[cat].alloc_entries_num*sizeof(int), + &pos, sizeof(int)); + } + is->files[cat].alloc_entries_num++; +} +#else +static void flush_block (ISAMC is, int cat) +{ + char *abuf = is->files[cat].alloc_buf; + xfree (abuf); +} + static int alloc_block (ISAMC is, int cat) { int block; @@ -254,6 +363,18 @@ static int alloc_block (ISAMC is, int cat) return block; } +static void release_block (ISAMC is, int cat, int pos) +{ + char buf[sizeof(int)]; + + (is->files[cat].no_released)++; + is->files[cat].head_is_dirty = 1; + memcpy (buf, &is->files[cat].head.freelist, sizeof(int)); + is->files[cat].head.freelist = pos; + bf_write (is->files[cat].bf, pos, 0, sizeof(int), buf); +} +#endif + int isc_alloc_block (ISAMC is, int cat) { int block = 0; @@ -276,17 +397,6 @@ int isc_alloc_block (ISAMC is, int cat) return block; } -static void release_block (ISAMC is, int cat, int pos) -{ - char buf[sizeof(int)]; - - (is->files[cat].no_released)++; - is->files[cat].head_is_dirty = 1; - memcpy (buf, &is->files[cat].head.freelist, sizeof(int)); - is->files[cat].head.freelist = pos; - bf_write (is->files[cat].bf, pos, 0, sizeof(int), buf); -} - void isc_release_block (ISAMC is, int cat, int pos) { if (is->method->debug > 3) @@ -309,7 +419,8 @@ static void init_fc (ISAMC is, int cat) int j = 100; is->files[cat].fc_max = j; - is->files[cat].fc_list = xmalloc (sizeof(*is->files[0].fc_list) * j); + is->files[cat].fc_list = (int *) + xmalloc (sizeof(*is->files[0].fc_list) * j); while (--j >= 0) is->files[cat].fc_list[j] = 0; } @@ -337,13 +448,13 @@ void isc_pp_close (ISAMC_PP pp) ISAMC_PP isc_pp_open (ISAMC is, ISAMC_P ipos) { - ISAMC_PP pp = xmalloc (sizeof(*pp)); + ISAMC_PP pp = (ISAMC_PP) xmalloc (sizeof(*pp)); char *src; pp->cat = isc_type(ipos); pp->pos = isc_block(ipos); - src = pp->buf = xmalloc (is->method->filecat[pp->cat].bsize); + src = pp->buf = (char *) xmalloc (is->method->filecat[pp->cat].bsize); pp->next = 0; pp->size = 0; @@ -391,10 +502,33 @@ int isc_read_item (ISAMC_PP pp, char **dst) if (pp->offset >= pp->size) { + if (!pp->next) + { + pp->pos = 0; + return 0; /* end of file */ + } + if (pp->next > pp->pos) + { + if (pp->next == pp->pos + 1) + is->files[pp->cat].no_next++; + else + { + is->files[pp->cat].no_forward++; + is->files[pp->cat].sum_forward += pp->next - pp->pos; + } + } + else + { + if (pp->next + 1 == pp->pos) + is->files[pp->cat].no_prev++; + else + { + is->files[pp->cat].no_backward++; + is->files[pp->cat].sum_backward += pp->pos - pp->next; + } + } /* out new block position */ pp->pos = pp->next; - if (!pp->pos) - return 0; /* end of file */ src = pp->buf; /* read block and save 'next' and 'size' entry */ isc_read_block (is, pp->cat, pp->pos, src);