X-Git-Url: http://git.indexdata.com/?a=blobdiff_plain;f=index%2Frecindex.c;h=73ef6339d94b3e9eadfc238d6a9d99472b23998a;hb=beda7a0a471cf2c19abb42dd9d1dad0d2aa28069;hp=f64816cda7585c55cdecbfb658d122cff0bca760;hpb=5214040d316f7b5228e46e1749505e3c0c8869d4;p=idzebra-moved-to-github.git diff --git a/index/recindex.c b/index/recindex.c index f64816c..73ef633 100644 --- a/index/recindex.c +++ b/index/recindex.c @@ -1,114 +1,623 @@ /* - * Copyright (C) 1994-1995, Index Data I/S + * Copyright (C) 1994-1998, Index Data I/S * All rights reserved. * Sebastian Hammer, Adam Dickmeiss * * $Log: recindex.c,v $ - * Revision 1.1 1995-11-15 14:46:20 adam + * Revision 1.20 1998-01-12 15:04:08 adam + * The test option (-s) only uses read-lock (and not write lock). + * + * Revision 1.19 1997/09/17 12:19:16 adam + * Zebra version corresponds to YAZ version 1.4. + * Changed Zebra server so that it doesn't depend on global common_resource. + * + * Revision 1.18 1997/07/15 16:28:42 adam + * Bug fix: storeData didn't work with files with multiple records. + * Bug fix: fixed memory management with records; not really well + * thought through. + * + * Revision 1.17 1997/02/12 20:39:46 adam + * Implemented options -f that limits the log to the first + * records. + * Changed some log messages also. + * + * Revision 1.16 1996/06/04 10:19:00 adam + * Minor changes - removed include of ctype.h. + * + * Revision 1.15 1996/05/13 14:23:06 adam + * Work on compaction of set/use bytes in dictionary. + * + * Revision 1.14 1996/02/01 20:48:15 adam + * The total size of records are always checked in rec_cache_insert to + * reduce memory usage. + * + * Revision 1.13 1995/12/11 09:12:49 adam + * The rec_get function returns NULL if record doesn't exist - will + * happen in the server if the result set records have been deleted since + * the creation of the set (i.e. the search). + * The server saves a result temporarily if it is 'volatile', i.e. the + * set is register dependent. + * + * Revision 1.12 1995/12/07 17:38:47 adam + * Work locking mechanisms for concurrent updates/commit. + * + * Revision 1.11 1995/12/06 13:58:26 adam + * Improved flushing of records - all flushes except the last one + * don't write the last accessed. Also flush takes place if record + * info occupy more than about 256k. + * + * Revision 1.10 1995/12/06 12:41:24 adam + * New command 'stat' for the index program. + * Filenames can be read from stdin by specifying '-'. + * Bug fix/enhancement of the transformation from terms to regular + * expressons in the search engine. + * + * Revision 1.9 1995/11/30 08:34:33 adam + * Started work on commit facility. + * Changed a few malloc/free to xmalloc/xfree. + * + * Revision 1.8 1995/11/28 14:26:21 adam + * Bug fix: recordId with constant wasn't right. + * Bug fix: recordId dictionary entry wasn't deleted when needed. + * + * Revision 1.7 1995/11/28 09:09:43 adam + * Zebra config renamed. + * Use setting 'recordId' to identify record now. + * Bug fix in recindex.c: rec_release_blocks was invokeded even + * though the blocks were already released. + * File traversal properly deletes records when needed. + * + * Revision 1.6 1995/11/25 10:24:06 adam + * More record fields - they are enumerated now. + * New options: flagStoreData flagStoreKey. + * + * Revision 1.5 1995/11/22 17:19:18 adam + * Record management uses the bfile system. + * + * Revision 1.4 1995/11/20 16:59:46 adam + * New update method: the 'old' keys are saved for each records. + * + * Revision 1.3 1995/11/16 15:34:55 adam + * Uses new record management system in both indexer and server. + * + * Revision 1.2 1995/11/15 19:13:08 adam + * Work on record management. + * + * Revision 1.1 1995/11/15 14:46:20 adam * Started work on better record management system. * */ #include #include #include -#include -#include -#include -#include - -#include "recindex.h" -#define REC_HEAD_MAGIC "rechead" +#include "recindxp.h" static void rec_write_head (Records p) { int r; assert (p); - assert (p->fd != -1); - if (lseek (p->fd, (off_t) 0, SEEK_SET) == -1) + assert (p->index_BFile); + + r = bf_write (p->index_BFile, 0, 0, sizeof(p->head), &p->head); + if (r) { - logf (LOG_FATAL|LOG_ERRNO, "lseek to 0 in %s", p->fname); + logf (LOG_FATAL|LOG_ERRNO, "write head of %s", p->index_fname); exit (1); } - r = write (p->fd, &p->head, sizeof(p->head)); - switch (r) +} + +static void rec_tmp_expand (Records p, int size, int dst_type) +{ + if (p->tmp_size < size + 2048 || + p->tmp_size < p->head.block_size[dst_type]*2) { - case -1: - logf (LOG_FATAL|LOG_ERRNO, "write head of %s", p->fname); - exit (1); - case sizeof(p->head): - break; - default: - logf (LOG_FATAL, "write head of %s. wrote %d", p->fname, r); - exit (1); + xfree (p->tmp_buf); + p->tmp_size = size + p->head.block_size[dst_type]*2 + 2048; + p->tmp_buf = xmalloc (p->tmp_size); } } -Records rec_open (int rw) +static int read_indx (Records p, int sysno, void *buf, int itemsize, + int ignoreError) { - Records p; int r; + int pos = (sysno-1)*itemsize; - if (!(p = malloc (sizeof(*p)))) + r = bf_read (p->index_BFile, 1+pos/128, pos%128, itemsize, buf); + if (r != 1 && !ignoreError) { - logf (LOG_FATAL|LOG_ERRNO, "malloc"); + logf (LOG_FATAL|LOG_ERRNO, "read in %s at pos %ld", + p->index_fname, (long) pos); exit (1); } - p->fname = "recindex"; - p->fd = open (p->fname, rw ? (O_RDWR|O_CREAT) : O_RDONLY, 0666); - if (p->fd == -1) + return r; +} + +static void write_indx (Records p, int sysno, void *buf, int itemsize) +{ + int pos = (sysno-1)*itemsize; + + bf_write (p->index_BFile, 1+pos/128, pos%128, itemsize, buf); +} + +static void rec_release_blocks (Records p, int sysno) +{ + struct record_index_entry entry; + int freeblock, freenext; + int dst_type; + + if (read_indx (p, sysno, &entry, sizeof(entry), 1) != 1) + return ; + p->head.total_bytes -= entry.size; + freeblock = entry.next; + assert (freeblock > 0); + dst_type = freeblock & 7; + assert (dst_type < REC_BLOCK_TYPES); + freeblock = freeblock / 8; + while (freeblock) { - logf (LOG_FATAL|LOG_ERRNO, "open %s", p->fname); + if (bf_read (p->data_BFile[dst_type], freeblock, 0, sizeof(freenext), + &freenext) != 1) + { + logf (LOG_FATAL|LOG_ERRNO, "read in rec_del_single"); + exit (1); + } + if (bf_write (p->data_BFile[dst_type], freeblock, 0, sizeof(freenext), + &p->head.block_free[dst_type])) + { + logf (LOG_FATAL|LOG_ERRNO, "write in rec_del_single"); + exit (1); + } + p->head.block_free[dst_type] = freeblock; + freeblock = freenext; + p->head.block_used[dst_type]--; + } +} + +static void rec_delete_single (Records p, Record rec) +{ + struct record_index_entry entry; + + rec_release_blocks (p, rec->sysno); + + entry.next = p->head.index_free; + entry.size = 0; + p->head.index_free = rec->sysno; + write_indx (p, rec->sysno, &entry, sizeof(entry)); +} + + +static void rec_write_single (Records p, Record rec) +{ + int i, size = 0; + char *cptr; + int dst_type = 0; + int no_written = 0; + int block_prev = -1, block_free; + struct record_index_entry entry; + + for (i = 0; i < REC_NO_INFO; i++) + if (!rec->info[i]) + size += sizeof(*rec->size); + else + size += sizeof(*rec->size) + rec->size[i]; + + for (i = 1; i= p->head.block_move[i]) + dst_type = i; + + rec_tmp_expand (p, size, dst_type); + + cptr = p->tmp_buf + sizeof(int); /* a hack! */ + for (i = 0; i < REC_NO_INFO; i++) + { + memcpy (cptr, &rec->size[i], sizeof(*rec->size)); + cptr += sizeof(*rec->size); + if (rec->info[i]) + { + memcpy (cptr, rec->info[i], rec->size[i]); + cptr += rec->size[i]; + } + } + cptr = p->tmp_buf; + while (no_written < size) + { + block_free = p->head.block_free[dst_type]; + if (block_free) + { + if (bf_read (p->data_BFile[dst_type], + block_free, 0, sizeof(*p->head.block_free), + &p->head.block_free[dst_type]) != 1) + { + logf (LOG_FATAL|LOG_ERRNO, "read in %s at free block %d", + p->data_fname[dst_type], block_free); + exit (1); + } + } + else + block_free = p->head.block_last[dst_type]++; + if (block_prev == -1) + { + entry.next = block_free*8 + dst_type; + entry.size = size; + p->head.total_bytes += size; + write_indx (p, rec->sysno, &entry, sizeof(entry)); + } + else + { + memcpy (cptr, &block_free, sizeof(int)); + bf_write (p->data_BFile[dst_type], block_prev, 0, 0, cptr); + cptr = p->tmp_buf + no_written; + } + block_prev = block_free; + no_written += p->head.block_size[dst_type] - sizeof(int); + p->head.block_used[dst_type]++; + } + assert (block_prev != -1); + block_free = 0; + memcpy (cptr, &block_free, sizeof(int)); + bf_write (p->data_BFile[dst_type], block_prev, 0, + sizeof(int) + (p->tmp_buf+size) - cptr, cptr); +} + +static void rec_update_single (Records p, Record rec) +{ + rec_release_blocks (p, rec->sysno); + rec_write_single (p, rec); +} + +Records rec_open (BFiles bfs, int rw) +{ + Records p; + int i, r; + + p = xmalloc (sizeof(*p)); + p->rw = rw; + p->tmp_size = 1024; + p->tmp_buf = xmalloc (p->tmp_size); + p->index_fname = "recindex"; + p->index_BFile = bf_open (bfs, p->index_fname, 128, rw); + if (p->index_BFile == NULL) + { + logf (LOG_FATAL|LOG_ERRNO, "open %s", p->index_fname); exit (1); } - r = read (p->fd, &p->head, sizeof(p->head)); + r = bf_read (p->index_BFile, 0, 0, 0, p->tmp_buf); switch (r) { - case -1: - logf (LOG_FATAL|LOG_ERRNO, "read %s", p->fname); - exit (1); case 0: + memcpy (p->head.magic, REC_HEAD_MAGIC, sizeof(p->head.magic)); + p->head.index_free = 0; + p->head.index_last = 1; p->head.no_records = 0; - p->head.freelist = 0; + p->head.total_bytes = 0; + for (i = 0; ihead.block_free[i] = 0; + p->head.block_last[i] = 1; + p->head.block_used[i] = 0; + } + p->head.block_size[0] = 128; + p->head.block_move[0] = 0; + for (i = 1; ihead.block_size[i] = p->head.block_size[i-1] * 4; + p->head.block_move[i] = p->head.block_size[i] * 3; + } if (rw) rec_write_head (p); break; - case sizeof(p->head): + case 1: + memcpy (&p->head, p->tmp_buf, sizeof(p->head)); if (memcmp (p->head.magic, REC_HEAD_MAGIC, sizeof(p->head.magic))) { - logf (LOG_FATAL, "read %s. bad header", p->fname); + logf (LOG_FATAL, "read %s. bad header", p->index_fname); exit (1); } break; - default: - logf (LOG_FATAL, "read head of %s. expected %d. got %d", - p->fname, sizeof(p->head), r); - exit (1); } + for (i = 0; idata_fname[i] = xmalloc (strlen(str)+1); + strcpy (p->data_fname[i], str); + p->data_BFile[i] = NULL; + } + for (i = 0; idata_BFile[i] = bf_open (bfs, p->data_fname[i], + p->head.block_size[i], + rw))) + { + logf (LOG_FATAL|LOG_ERRNO, "bf_open %s", p->data_fname[i]); + exit (1); + } + } + p->cache_max = 10; + p->cache_cur = 0; + p->record_cache = xmalloc (sizeof(*p->record_cache)*p->cache_max); return p; } -void rec_close (Records p) +static void rec_cache_flush (Records p, int saveCount) { - if (p->fd != -1) - close (p->fd); - free (p); + int i, j; + + if (saveCount >= p->cache_cur) + saveCount = 0; + for (i = 0; icache_cur - saveCount; i++) + { + struct record_cache_entry *e = p->record_cache + i; + switch (e->flag) + { + case recordFlagNop: + break; + case recordFlagNew: + rec_write_single (p, e->rec); + break; + case recordFlagWrite: + rec_update_single (p, e->rec); + break; + case recordFlagDelete: + rec_delete_single (p, e->rec); + break; + } + rec_rm (&e->rec); + } + for (j = 0; jrecord_cache+j, p->record_cache+i, + sizeof(*p->record_cache)); + p->cache_cur = saveCount; +} + +static Record *rec_cache_lookup (Records p, int sysno, + enum recordCacheFlag flag) +{ + int i; + for (i = 0; icache_cur; i++) + { + struct record_cache_entry *e = p->record_cache + i; + if (e->rec->sysno == sysno) + { + if (flag != recordFlagNop && e->flag == recordFlagNop) + e->flag = flag; + return &e->rec; + } + } + return NULL; +} + +static void rec_cache_insert (Records p, Record rec, enum recordCacheFlag flag) +{ + struct record_cache_entry *e; + + if (p->cache_cur == p->cache_max) + rec_cache_flush (p, 1); + else if (p->cache_cur > 0) + { + int i, j; + int used = 0; + for (i = 0; icache_cur; i++) + { + Record r = (p->record_cache + i)->rec; + for (j = 0; jsize[j]; + } + if (used > 256000) + rec_cache_flush (p, 1); + } + assert (p->cache_cur < p->cache_max); + + e = p->record_cache + (p->cache_cur)++; + e->flag = flag; + e->rec = rec_cp (rec); +} + +void rec_close (Records *pp) +{ + Records p = *pp; + int i; + + assert (p); + + rec_cache_flush (p, 0); + xfree (p->record_cache); + + if (p->rw) + rec_write_head (p); + + if (p->index_BFile) + bf_close (p->index_BFile); + + for (i = 0; idata_BFile[i]) + bf_close (p->data_BFile[i]); + xfree (p->data_fname[i]); + } + xfree (p->tmp_buf); + xfree (p); + *pp = NULL; } + Record rec_get (Records p, int sysno) { + int i; + Record rec, *recp; + struct record_index_entry entry; + int freeblock, dst_type; + char *nptr, *cptr; + + assert (sysno > 0); assert (p); - return NULL; + + if ((recp = rec_cache_lookup (p, sysno, recordFlagNop))) + return rec_cp (*recp); + + if (!read_indx (p, sysno, &entry, sizeof(entry), 1)) + return NULL; /* record is not there! */ + + if (!entry.size) + return NULL; /* record is deleted */ + + dst_type = entry.next & 7; + assert (dst_type < REC_BLOCK_TYPES); + freeblock = entry.next / 8; + + assert (freeblock > 0); + + rec = xmalloc (sizeof(*rec)); + rec_tmp_expand (p, entry.size, dst_type); + + cptr = p->tmp_buf; + bf_read (p->data_BFile[dst_type], freeblock, 0, 0, cptr); + memcpy (&freeblock, cptr, sizeof(freeblock)); + + while (freeblock) + { + int tmp; + + cptr += p->head.block_size[dst_type] - sizeof(freeblock); + + memcpy (&tmp, cptr, sizeof(tmp)); + bf_read (p->data_BFile[dst_type], freeblock, 0, 0, cptr); + memcpy (&freeblock, cptr, sizeof(freeblock)); + memcpy (cptr, &tmp, sizeof(tmp)); + } + + rec->sysno = sysno; + nptr = p->tmp_buf + sizeof(freeblock); + for (i = 0; i < REC_NO_INFO; i++) + { + memcpy (&rec->size[i], nptr, sizeof(*rec->size)); + nptr += sizeof(*rec->size); + if (rec->size[i]) + { + rec->info[i] = xmalloc (rec->size[i]); + memcpy (rec->info[i], nptr, rec->size[i]); + nptr += rec->size[i]; + } + else + rec->info[i] = NULL; + } + rec_cache_insert (p, rec, recordFlagNop); + return rec; } Record rec_new (Records p) { + int sysno, i; + Record rec; + assert (p); - return NULL; + rec = xmalloc (sizeof(*rec)); + if (1 || p->head.index_free == 0) + sysno = (p->head.index_last)++; + else + { + struct record_index_entry entry; + + read_indx (p, p->head.index_free, &entry, sizeof(entry), 0); + sysno = p->head.index_free; + p->head.index_free = entry.next; + } + (p->head.no_records)++; + rec->sysno = sysno; + for (i = 0; i < REC_NO_INFO; i++) + { + rec->info[i] = NULL; + rec->size[i] = 0; + } + rec_cache_insert (p, rec, recordFlagNew); + return rec; } -void rec_put (Records p, Record rec) +void rec_del (Records p, Record *recpp) { - assert (p); + Record *recp; + + (p->head.no_records)--; + if ((recp = rec_cache_lookup (p, (*recpp)->sysno, recordFlagDelete))) + { + rec_rm (recp); + *recp = *recpp; + } + else + { + rec_cache_insert (p, *recpp, recordFlagDelete); + rec_rm (recpp); + } + *recpp = NULL; +} + +void rec_put (Records p, Record *recpp) +{ + Record *recp; + + if ((recp = rec_cache_lookup (p, (*recpp)->sysno, recordFlagWrite))) + { + rec_rm (recp); + *recp = *recpp; + } + else + { + rec_cache_insert (p, *recpp, recordFlagWrite); + rec_rm (recpp); + } + *recpp = NULL; +} + +void rec_rm (Record *recpp) +{ + int i; + + if (!*recpp) + return ; + for (i = 0; i < REC_NO_INFO; i++) + xfree ((*recpp)->info[i]); + xfree (*recpp); + *recpp = NULL; } + +Record rec_cp (Record rec) +{ + Record n; + int i; + + n = xmalloc (sizeof(*n)); + n->sysno = rec->sysno; + for (i = 0; i < REC_NO_INFO; i++) + if (!rec->info[i]) + { + n->info[i] = NULL; + n->size[i] = 0; + } + else + { + n->size[i] = rec->size[i]; + n->info[i] = xmalloc (rec->size[i]); + memcpy (n->info[i], rec->info[i], rec->size[i]); + } + return n; +} + + +char *rec_strdup (const char *s, size_t *len) +{ + char *p; + + if (!s) + { + *len = 0; + return NULL; + } + *len = strlen(s)+1; + p = xmalloc (*len); + strcpy (p, s); + return p; +} +