X-Git-Url: http://git.indexdata.com/?a=blobdiff_plain;f=isamc%2Fmerge.c;h=c8167663d248e032cc2f57b77b725f367687dd65;hb=840cf9f959b42b3f4b107de8e125cbc8c9cbbcfa;hp=6906176499b1dd6a84860b76e11b7bf833633bd1;hpb=bf25b1c32932f0f9dbe10119de1c361d9301e6a8;p=idzebra-moved-to-github.git diff --git a/isamc/merge.c b/isamc/merge.c index 6906176..c816766 100644 --- a/isamc/merge.c +++ b/isamc/merge.c @@ -1,24 +1,39 @@ -/* - * Copyright (c) 1996-1998, Index Data. - * See the file LICENSE for details. - * Sebastian Hammer, Adam Dickmeiss, Heikki Levanto - * - */ +/* $Id: merge.c,v 1.26 2004-08-06 12:55:02 adam Exp $ + Copyright (C) 1995,1996,1997,1998,1999,2000,2001,2002,2003,2004 + 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. +*/ #include #include #include #include -#include +#include #include "isamc-p.h" -#include "isamh-p.h" struct isc_merge_block { int offset; /* offset in r_buf */ - int block; /* block number of file (0 if none) */ + zint block; /* block number of file (0 if none) */ int dirty; /* block is different from that on file */ }; +#if 0 static void opt_blocks (ISAMC is, struct isc_merge_block *mb, int ptr, int last) { @@ -46,10 +61,11 @@ static void opt_blocks (ISAMC is, struct isc_merge_block *mb, int ptr, if (!last) mb[i].dirty = 1; } +#endif static void flush_blocks (ISAMC is, struct isc_merge_block *mb, int ptr, - char *r_buf, int *firstpos, int cat, int last, - int *numkeys) + char *r_buf, zint *firstpos, int cat, int last, + zint *numkeys) { int i; @@ -87,7 +103,7 @@ static void flush_blocks (ISAMC is, struct isc_merge_block *mb, int ptr, if (!*firstpos) *firstpos = mb[i].block; if (is->method->debug > 2) - logf (LOG_LOG, "isc: skip ptr=%d size=%d %d %d", + logf (LOG_LOG, "isc: skip ptr=%d size=%d %d " ZINT_FORMAT, i, ssize, cat, mb[i].block); ++(is->files[cat].no_skip_writes); continue; @@ -100,28 +116,27 @@ static void flush_blocks (ISAMC is, struct isc_merge_block *mb, int ptr, src = r_buf + mb[i].offset - ISAMC_BLOCK_OFFSET_1; ssize += ISAMC_BLOCK_OFFSET_1; - memcpy (src+sizeof(int)+sizeof(ssize), numkeys, - sizeof(*numkeys)); + memcpy (src+sizeof(zint)+sizeof(ssize), numkeys, sizeof(*numkeys)); if (is->method->debug > 2) - logf (LOG_LOG, "isc: flush ptr=%d numk=%d size=%d nextpos=%d", - i, *numkeys, (int) ssize, mb[i+1].block); + logf (LOG_LOG, "isc: flush ptr=%d numk=" ZINT_FORMAT " size=%d nextpos=" + ZINT_FORMAT, i, *numkeys, (int) ssize, mb[i+1].block); } else { src = r_buf + mb[i].offset - ISAMC_BLOCK_OFFSET_N; ssize += ISAMC_BLOCK_OFFSET_N; if (is->method->debug > 2) - logf (LOG_LOG, "isc: flush ptr=%d size=%d nextpos=%d", + logf (LOG_LOG, "isc: flush ptr=%d size=%d nextpos=" ZINT_FORMAT, i, (int) ssize, mb[i+1].block); } - memcpy (src, &mb[i+1].block, sizeof(int)); - memcpy (src+sizeof(int), &ssize, sizeof(ssize)); + memcpy (src, &mb[i+1].block, sizeof(zint)); + memcpy (src+sizeof(zint), &ssize, sizeof(ssize)); isc_write_block (is, cat, mb[i].block, src); } } -static int get_border (ISAMC is, struct isc_merge_block *mb, int ptr, - int cat, int firstpos) +static int get_border (ISAMC is, struct isc_merge_block *mb, zint ptr, + int cat, zint firstpos) { /* Border set to initial fill or block size depending on whether we are creating a new one or updating and old one. @@ -136,7 +151,7 @@ static int get_border (ISAMC is, struct isc_merge_block *mb, int ptr, return mb[ptr].offset + fill - off; } -ISAMC_P isc_merge (ISAMC is, ISAMC_P ipos, ISAMC_I data) +ISAMC_P isc_merge (ISAMC is, ISAMC_P ipos, ISAMC_I *data) { char i_item[128], *i_item_ptr; @@ -150,17 +165,17 @@ ISAMC_P isc_merge (ISAMC is, ISAMC_P ipos, ISAMC_I data) struct isc_merge_block mb[200]; - int firstpos = 0; + zint firstpos = 0; int cat = 0; char r_item_buf[128]; /* temporary result output */ char *r_buf; /* block with resulting data */ int r_offset = 0; /* current offset in r_buf */ int ptr = 0; /* pointer */ void *r_clientData; /* encode client data */ - int border; - int numKeys = 0; + zint border; + zint numKeys = 0; - r_clientData = (*is->method->code_start)(ISAMC_ENCODE); + r_clientData = (*is->method->codec.start)(); r_buf = is->merge_buf + 128; pp = isc_pp_open (is, ipos); @@ -172,7 +187,7 @@ ISAMC_P isc_merge (ISAMC is, ISAMC_P ipos, ISAMC_I data) cat = pp->cat; if (debug > 1) - logf (LOG_LOG, "isc: isc_merge begin %d %d", cat, pp->pos); + logf (LOG_LOG, "isc: isc_merge begin %d " ZINT_FORMAT, cat, pp->pos); /* read first item from i */ i_item_ptr = i_item; @@ -319,10 +334,10 @@ ISAMC_P isc_merge (ISAMC is, ISAMC_P ipos, ISAMC_I data) if (r_item) /* insert resulting item? */ { char *r_out_ptr = r_buf + r_offset; + const char *src = r_item; int new_offset; - (*is->method->code_item)(ISAMC_ENCODE, r_clientData, - &r_out_ptr, &r_item); + (*is->method->codec.encode)(r_clientData, &r_out_ptr, &src); new_offset = r_out_ptr - r_buf; numKeys++; @@ -330,7 +345,8 @@ ISAMC_P isc_merge (ISAMC is, ISAMC_P ipos, ISAMC_I data) if (border < new_offset && border >= r_offset) { if (debug > 2) - logf (LOG_LOG, "isc: border %d %d", ptr, border); + logf (LOG_LOG, "isc: border %d " ZINT_FORMAT, + ptr, border); /* Max size of current block category reached ... make new virtual block entry */ mb[++ptr].block = 0; @@ -406,7 +422,8 @@ ISAMC_P isc_merge (ISAMC is, ISAMC_P ipos, ISAMC_I data) ptr = j; border = get_border (is, mb, ptr, cat, firstpos); if (debug > 3) - logf (LOG_LOG, "isc: border=%d r_offset=%d", border, r_offset); + logf (LOG_LOG, "isc: border=" ZINT_FORMAT " r_offset=%d", + border, r_offset); } } if (mb[ptr].offset < r_offset) @@ -438,7 +455,7 @@ ISAMC_P isc_merge (ISAMC is, ISAMC_P ipos, ISAMC_I data) if (numKeys != isc_pp_num (pp)) { if (debug > 2) - logf (LOG_LOG, "isc: patch num keys firstpos=%d num=%d", + logf (LOG_LOG, "isc: patch num keys firstpos=" ZINT_FORMAT " num=" ZINT_FORMAT, firstpos, numKeys); bf_write (is->files[cat].bf, firstpos, ISAMC_BLOCK_OFFSET_N, sizeof(numKeys), &numKeys); @@ -453,315 +470,12 @@ ISAMC_P isc_merge (ISAMC is, ISAMC_P ipos, ISAMC_I data) /* flush rest of block(s) in r_buf */ flush_blocks (is, mb, ptr, r_buf, &firstpos, cat, 1, &numKeys); - (*is->method->code_stop)(ISAMC_ENCODE, r_clientData); + (*is->method->codec.stop)(r_clientData); if (!firstpos) cat = 0; if (debug > 1) - logf (LOG_LOG, "isc: isc_merge return %d %d", cat, firstpos); + logf (LOG_LOG, "isc: isc_merge return %d " ZINT_FORMAT, cat, firstpos); isc_pp_close (pp); return cat + firstpos * 8; } -static char *hexdump(unsigned char *p, int len, char *buff) { - static char localbuff[128]; - char bytebuff[8]; - if (!buff) buff=localbuff; - *buff='\0'; - while (len--) { - sprintf(bytebuff,"%02x",*p); - p++; - strcat(buff,bytebuff); - if (len) strcat(buff,","); - } - return buff; -} - - -/* isamh - heikki's append-only isam - * Idea: When allocating a new block, allocate memory for a very large block - * (maximal blocksize). When done, see if you can shrink it to some - * smaller size. First-time indexing will go in optimal blocks, and - * following small additions will go to the end of the last of the - * maximal ones. Only later, when new blocks need to be allocated, it - * may make sense to reserve some extra space... -*/ - - -static void isamh_reduceblock(ISAMH is, ISAMH_PP pp, int numKeys) -{ - if (pp->is->method->debug > 2) - logf(LOG_LOG,"isamh_reduce: block p=%d c=%d o=%d nk=%d ", - pp->pos, pp->cat, pp->offset, numKeys); - if (pp->pos != 0) - return; /* already allocated in some size */ - while ( ( pp->cat > 0 ) && - ( pp->offset < is->method->filecat[pp->cat-1].bsize) && - ( numKeys < is->method->filecat[pp->cat-1].mblocks) ) - pp->cat--; - pp->pos = isamh_alloc_block(is,pp->cat) ; - if (pp->is->method->debug > 2) - logf(LOG_LOG,"isamh_reduced block p=%d to c=%d o=%d nk=%d bs=%d", - pp->pos, pp->cat, pp->offset, numKeys, - is->method->filecat[pp->cat].bsize); - - -} /* reduceblock */ - -ISAMC_P isamh_append (ISAMH is, ISAMH_P ipos, ISAMH_I data) -{ - - ISAMH_PP pp; - - char i_item[128]; - char *i_item_ptr; - int i_more=1, i_mode; - - char codebuffer[128]; - char *codeptr; - char *bufptr; - int codelen; - - ISAMH_PP firstpp; - void *r_clientData; /* encode client data */ - int newblock; - int newcat; - int maxkeys; - int maxsize; - int retval; - int maxcat; - - pp = firstpp = isamh_pp_open (is, ipos); - assert (*is->method->code_reset); - - maxcat=0; /* find the largest block size for default allocation */ - for ( newcat=0; is->method->filecat[newcat].mblocks > 0; newcat++) - if ( is->method->filecat[newcat].bsize > - is->method->filecat[maxcat].bsize) - maxcat = newcat; - - if ( 0==ipos) - { /* new block */ - pp->cat=maxcat; /* start large... */ - pp->pos = 0; /* not allocated yet */ - pp->size= pp->offset = ISAMH_BLOCK_OFFSET_1 ; - r_clientData = (*is->method->code_start)(ISAMH_ENCODE); - if (pp->is->method->debug > 2) - logf(LOG_LOG,"isamh_append: starting with new block %d",pp->pos); - } - else - { /* existing block */ - if (isamh_block(firstpp->lastblock) == firstpp->pos) - { /* only one block, we have it already */ - pp->offset=ISAMH_BLOCK_OFFSET_1; - if (pp->is->method->debug > 2) - logf(LOG_LOG,"isamh_append: starting with one block %d",pp->pos); - } - else - { - if (pp->is->method->debug > 2) - logf(LOG_LOG,"isamh_append: starting with multiple blocks %d>%d>%d", - firstpp->pos,isamh_block(firstpp->next),isamh_block(firstpp->lastblock)); - pp=isamh_pp_open(is,firstpp->lastblock); - /* dirty, but this can also read a N-block. Just clear extra values*/ - pp->lastblock=0; - pp->offset=ISAMH_BLOCK_OFFSET_N; - } /* get last */ - r_clientData = (*is->method->code_start)(ISAMH_ENCODE); - if (pp->is->method->debug > 3) - logf(LOG_LOG,"isamh_append: scanning to end of block %d %d->%d", - pp->pos, pp->offset, pp->size); - codeptr=codebuffer; - while (pp->offsetsize) { - codeptr=codebuffer; - bufptr=pp->buf + pp->offset; - (*is->method->code_item)(ISAMH_DECODE, r_clientData, &codeptr, &bufptr); - codelen = bufptr - (pp->buf+pp->offset) ; - if (pp->is->method->debug > 3) - logf(LOG_LOG,"isamh_append: dec at %d %d/%d:%s", - pp->offset, codelen, codeptr-codebuffer, - hexdump(codebuffer,codeptr-codebuffer,0) ); - pp->offset += codelen; - } - } /* existing block */ - - - i_item_ptr = i_item; - i_more = (*data->read_item)(data->clientData,&i_item_ptr,&i_mode); - if (pp->is->method->debug > 3) - logf(LOG_LOG,"isamh_append 1: m=%d l=%d %s", - i_mode, i_item_ptr-i_item, hexdump(i_item,i_item_ptr-i_item,0)); - - maxsize = is->method->filecat[pp->cat].bsize; - - while(i_more) { - if (i_mode) - { /* insert key, ignore all delete keys time being... */ - codeptr = codebuffer; - i_item_ptr=i_item; - (*is->method->code_item)(ISAMH_ENCODE, r_clientData, &codeptr, &i_item_ptr); - codelen = codeptr-codebuffer; - - assert( (codelen < 128) && (codelen>0)); - - if (pp->is->method->debug > 3) - logf(LOG_LOG,"isamh_append: coded into %d:%s (nk=%d)", - codelen,hexdump(codebuffer,codelen,0),firstpp->numKeys); - - if ( pp->offset + codelen > maxsize ) - { /* oops, block full, do something */ - newcat = maxcat; /* start with a large block again */ - maxkeys = is->method->filecat[pp->cat].mblocks; /* max keys */ - if (pp->is->method->debug > 2) - logf(LOG_LOG,"isamh_append: need new block: %d > %d (k:%d/%d)", - pp->offset + codelen, maxsize, firstpp->numKeys,maxkeys ); - -#ifdef SKIPTHIS - if ( (maxkeys>0) && (firstpp->numKeys > maxkeys) ) - { /* time to increase block size */ - newcat++; - maxsize = is->method->filecat[newcat].bsize; - pp->buf=xrealloc(pp->buf,maxsize); - if (pp->is->method->debug > 2) - logf(LOG_LOG,"isamh_append: increased to cat %d ",newcat); - } -#endif - - newblock = 0; /* isamh_alloc_block(is,newcat);*/ - pp->next = isamh_addr(newblock,newcat); - if (firstpp!=pp) - { /* not first block, write to disk already now */ - isamh_reduceblock(is,pp,firstpp->numKeys); - isamh_buildlaterblock(pp); - isamh_write_block(is,pp->cat,pp->pos,pp->buf); - } - else - { /* we had only one block, allocate a second buffer */ - pp = isamh_pp_open(is,0); - } - pp->cat = newcat; - pp->pos = newblock; - - pp->size=pp->offset=ISAMH_BLOCK_OFFSET_N ; - pp->next=0; - pp->lastblock=0; - if (pp->is->method->debug > 2) - logf(LOG_LOG,"isamh_append: got a new block %d:%d",pp->cat,pp->pos); - - /* reset the encoding, and code again */ - (*is->method->code_reset)(r_clientData); - codeptr = codebuffer; - i_item_ptr=i_item; - (*is->method->code_item)(ISAMH_ENCODE, r_clientData, &codeptr, &i_item_ptr); - codelen = codeptr-codebuffer; - if (pp->is->method->debug > 3) - logf(LOG_LOG,"isamh_append: coded again %d:%s (nk=%d)", - codelen,hexdump(codebuffer,codelen,0),firstpp->numKeys); - - } /* block full */ - - /* ok, now we can write it */ - memcpy(&(pp->buf[pp->offset]), codebuffer, codelen); - pp->offset += codelen; - pp->size += codelen; - firstpp->numKeys++; - } /* not a delete */ - - /* try to read the next element */ - i_item_ptr = i_item; - i_more = (*data->read_item)(data->clientData,&i_item_ptr,&i_mode); - if (pp->is->method->debug > 3) - logf(LOG_LOG,"isamh_append 2: m=%d l=%d %s", - i_mode, i_item_ptr-i_item, hexdump(i_item,i_item_ptr-i_item,0)); - - } /* while */ - - /* Write the last (partial) block, if needed. */ - if (pp!=firstpp) - { - pp->next=0; /* just to be sure */ - isamh_reduceblock(is,pp,firstpp->numKeys); - isamh_buildlaterblock(pp); - isamh_write_block(is,pp->cat,pp->pos,pp->buf); - } - - /* update first block and write it */ - firstpp->lastblock = isamh_addr(pp->pos,pp->cat); - isamh_reduceblock(is,firstpp,firstpp->numKeys); - isamh_buildfirstblock(firstpp); - isamh_write_block(is,firstpp->cat,firstpp->pos,firstpp->buf); - - /* release the second block, if we allocated one */ - if ( firstpp != pp ) - isamh_pp_close(pp); - - /* get return value (before it disappears at close! */ - retval = isamh_addr(firstpp->pos,firstpp->cat); - - isamh_pp_close(firstpp); - - return retval; - -} /* isamh_append */ - - -/* - * $Log: merge.c,v $ - * Revision 1.17 1999-07-13 14:22:17 heikki - * Better allocation strategy in isamh_merge - * - * Revision 1.16 1999/07/08 14:23:27 heikki - * Fixed a bug in isamh_pp_read and cleaned up a bit - * - * Revision 1.15 1999/07/07 09:36:04 heikki - * Fixed an assertion in isamh - * - * Revision 1.13 1999/07/06 09:37:05 heikki - * Working on isamh - not ready yet. - * - * Revision 1.12 1999/06/30 15:03:55 heikki - * first take on isamh, the append-only isam structure - * - * Revision 1.11 1999/05/26 07:49:14 adam - * C++ compilation. - * - * Revision 1.10 1998/03/19 12:22:09 adam - * Minor change. - * - * Revision 1.9 1998/03/19 10:04:38 adam - * Minor changes. - * - * Revision 1.8 1998/03/18 09:23:55 adam - * Blocks are stored in chunks on free list - up to factor 2 in speed. - * Fixed bug that could occur in block category rearrangemen. - * - * Revision 1.7 1998/03/11 11:18:18 adam - * Changed the isc_merge to take into account the mfill (minimum-fill). - * - * Revision 1.6 1998/03/06 13:54:03 adam - * Fixed two nasty bugs in isc_merge. - * - * Revision 1.5 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 1.4 1996/11/08 11:15:31 adam - * Number of keys in chain are stored in first block and the function - * to retrieve this information, isc_pp_num is implemented. - * - * Revision 1.3 1996/11/04 14:08:59 adam - * Optimized free block usage. - * - * Revision 1.2 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 1.1 1996/11/01 08:59:15 adam - * First version of isc_merge that supports update/delete. - * - */ - -