1 /* $Id: kinput.c,v 1.59 2004-06-15 10:56:31 adam Exp $
2 Copyright (C) 1995,1996,1997,1998,1999,2000,2001,2002
5 This file is part of the Zebra server.
7 Zebra is free software; you can redistribute it and/or modify it under
8 the terms of the GNU General Public License as published by the Free
9 Software Foundation; either version 2, or (at your option) any later
12 Zebra is distributed in the hope that it will be useful, but WITHOUT ANY
13 WARRANTY; without even the implied warranty of MERCHANTABILITY or
14 FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
17 You should have received a copy of the GNU General Public License
18 along with Zebra; see the file LICENSE.zebra. If not, write to the
19 Free Software Foundation, 59 Temple Place - Suite 330, Boston, MA
38 #define KEY_SIZE (1+sizeof(struct it_key))
39 #define INP_NAME_MAX 768
40 #define INP_BUF_START 60000
41 #define INP_BUF_ADD 400000
46 off_t offset; /* file offset */
47 unsigned char *buf; /* buffer block */
48 size_t buf_size; /* number of read bytes in block */
49 size_t chunk; /* number of bytes allocated */
50 size_t buf_ptr; /* current position in buffer */
51 char *prev_name; /* last word read */
52 int sysno; /* last sysno */
53 int seqno; /* last seqno */
54 off_t length; /* length of file */
55 /* handler invoked in each read */
56 void (*readHandler)(struct key_file *keyp, void *rinfo);
61 void getFnameTmp (Res res, char *fname, int no)
65 pre = res_get_def (res, "keyTmpDir", ".");
66 sprintf (fname, "%s/key%d.tmp", pre, no);
69 void extract_get_fname_tmp (ZebraHandle zh, char *fname, int no)
73 pre = res_get_def (zh->res, "keyTmpDir", ".");
74 sprintf (fname, "%s/key%d.tmp", pre, no);
77 void key_file_chunk_read (struct key_file *f)
79 int nr = 0, r = 0, fd;
81 getFnameTmp (f->res, fname, f->no);
82 fd = open (fname, O_BINARY|O_RDONLY);
88 logf (LOG_WARN|LOG_ERRNO, "cannot open %s", fname);
93 if ((f->length = lseek (fd, 0L, SEEK_END)) == (off_t) -1)
95 logf (LOG_WARN|LOG_ERRNO, "cannot seek %s", fname);
100 if (lseek (fd, f->offset, SEEK_SET) == -1)
102 logf (LOG_WARN|LOG_ERRNO, "cannot seek %s", fname);
106 while (f->chunk - nr > 0)
108 r = read (fd, f->buf + nr, f->chunk - nr);
115 logf (LOG_WARN|LOG_ERRNO, "read of %s", fname);
121 (*f->readHandler)(f, f->readInfo);
125 void key_file_destroy (struct key_file *f)
128 xfree (f->prev_name);
132 struct key_file *key_file_init (int no, int chunk, Res res)
136 f = (struct key_file *) xmalloc (sizeof(*f));
144 f->readHandler = NULL;
145 f->buf = (unsigned char *) xmalloc (f->chunk);
146 f->prev_name = (char *) xmalloc (INP_NAME_MAX);
147 *f->prev_name = '\0';
148 key_file_chunk_read (f);
152 int key_file_getc (struct key_file *f)
154 if (f->buf_ptr < f->buf_size)
155 return f->buf[(f->buf_ptr)++];
156 if (f->buf_size < f->chunk)
158 f->offset += f->buf_size;
159 key_file_chunk_read (f);
160 if (f->buf_ptr < f->buf_size)
161 return f->buf[(f->buf_ptr)++];
166 int key_file_decode (struct key_file *f)
170 c = key_file_getc (f);
177 d = ((c&63) << 8) + (key_file_getc (f) & 0xff);
180 d = ((c&63) << 8) + (key_file_getc (f) & 0xff);
181 d = (d << 8) + (key_file_getc (f) & 0xff);
184 d = ((c&63) << 8) + (key_file_getc (f) & 0xff);
185 d = (d << 8) + (key_file_getc (f) & 0xff);
186 d = (d << 8) + (key_file_getc (f) & 0xff);
192 int key_file_read (struct key_file *f, char *key)
197 c = key_file_getc (f);
200 strcpy (key, f->prev_name);
209 while ((key[i++] = key_file_getc (f)))
211 strcpy (f->prev_name, key);
214 d = key_file_decode (f);
217 itkey.sysno = d + f->sysno;
220 f->sysno = itkey.sysno;
223 d = key_file_decode (f);
224 itkey.seqno = d + f->seqno;
225 f->seqno = itkey.seqno;
226 memcpy (key + i, &itkey, sizeof(struct it_key));
227 return i + sizeof (struct it_key);
232 struct key_file **file;
237 int (*cmp)(const void *p1, const void *p2);
238 struct zebra_register *reg;
239 ZebraHandle zh; /* only used for raw reading that bypasses the heaps */
247 static struct heap_info *key_heap_malloc()
248 { /* malloc and clear it */
249 struct heap_info *hi;
250 hi = (struct heap_info *) xmalloc (sizeof(*hi));
259 hi->no_deletions = 0;
260 hi->no_insertions = 0;
261 hi->no_iterations = 0;
265 struct heap_info *key_heap_init (int nkeys,
266 int (*cmp)(const void *p1, const void *p2))
268 struct heap_info *hi;
271 hi = key_heap_malloc();
272 hi->info.file = (struct key_file **)
273 xmalloc (sizeof(*hi->info.file) * (1+nkeys));
274 hi->info.buf = (char **) xmalloc (sizeof(*hi->info.buf) * (1+nkeys));
275 hi->ptr = (int *) xmalloc (sizeof(*hi->ptr) * (1+nkeys));
277 for (i = 0; i<= nkeys; i++)
280 hi->info.buf[i] = (char *) xmalloc (INP_NAME_MAX);
285 struct heap_info *key_heap_init_buff ( ZebraHandle zh,
286 int (*cmp)(const void *p1, const void *p2))
288 struct heap_info *hi=key_heap_malloc();
294 void key_heap_destroy (struct heap_info *hi, int nkeys)
297 yaz_log (LOG_DEBUG, "key_heap_destroy");
298 yaz_log (LOG_DEBUG, "key_heap_destroy nk=%d",nkeys);
300 for (i = 0; i<=nkeys; i++)
301 xfree (hi->info.buf[i]);
303 xfree (hi->info.buf);
305 xfree (hi->info.file);
309 static void key_heap_swap (struct heap_info *hi, int i1, int i2)
314 hi->ptr[i1] = hi->ptr[i2];
319 static void key_heap_delete (struct heap_info *hi)
321 int cur = 1, child = 2;
323 assert (hi->heapnum > 0);
325 key_heap_swap (hi, 1, hi->heapnum);
327 while (child <= hi->heapnum) {
328 if (child < hi->heapnum &&
329 (*hi->cmp)(&hi->info.buf[hi->ptr[child]],
330 &hi->info.buf[hi->ptr[child+1]]) > 0)
332 if ((*hi->cmp)(&hi->info.buf[hi->ptr[cur]],
333 &hi->info.buf[hi->ptr[child]]) > 0)
335 key_heap_swap (hi, cur, child);
344 static void key_heap_insert (struct heap_info *hi, const char *buf, int nbytes,
349 cur = ++(hi->heapnum);
350 memcpy (hi->info.buf[hi->ptr[cur]], buf, nbytes);
351 hi->info.file[hi->ptr[cur]] = kf;
354 while (parent && (*hi->cmp)(&hi->info.buf[hi->ptr[parent]],
355 &hi->info.buf[hi->ptr[cur]]) > 0)
357 key_heap_swap (hi, cur, parent);
363 static int heap_read_one_raw (struct heap_info *hi, char *name, char *key)
365 ZebraHandle zh=hi->zh;
366 size_t ptr_i = zh->reg->ptr_i;
371 cp=(zh->reg->key_buf)[zh->reg->ptr_top - ptr_i];
372 logf (LOG_DEBUG, " raw: i=%ld top=%ld cp=%p", (long) ptr_i,
373 (long) zh->reg->ptr_top, cp);
375 memcpy(key, cp+strlen(name)+1, KEY_SIZE);
380 static int heap_read_one (struct heap_info *hi, char *name, char *key)
383 char rbuf[INP_NAME_MAX];
386 if (hi->zh) /* bypass the heap stuff, we have a readymade buffer */
387 return heap_read_one_raw(hi, name, key);
392 strcpy (name, hi->info.buf[n]);
393 kf = hi->info.file[n];
395 memcpy (key, hi->info.buf[n] + r+1, KEY_SIZE);
396 key_heap_delete (hi);
397 if ((r = key_file_read (kf, rbuf)))
398 key_heap_insert (hi, rbuf, r, kf);
406 static void pkey(const char *b, int mode)
408 struct it_key *key = (struct it_key *) b;
409 printf ("%c %d:%d\n", mode + 48, key->sysno, key->seqno);
413 struct heap_cread_info {
414 char prev_name[INP_NAME_MAX];
415 char cur_name[INP_NAME_MAX];
420 struct heap_info *hi;
426 static int heap_cread_item (void *vp, char **dst, int *insertMode);
428 int heap_cread_item2 (void *vp, char **dst, int *insertMode)
430 struct heap_cread_info *p = (struct heap_cread_info *) vp;
433 if (p->ret == 0) /* lookahead was 0?. Return that in read next round */
438 else if (p->ret == -1) /* Must read new item ? */
440 char *dst_1 = p->key_1;
441 p->ret = heap_cread_item(vp, &dst_1, &p->mode_1);
442 p->sz_1 = dst_1 - p->key_1;
445 { /* lookahead in 2 . Now in 1. */
447 p->mode_1 = p->mode_2;
448 memcpy (p->key_1, p->key_2, p->sz_2);
451 level = 1; /* insert */
453 level = -1; /* delete */
456 char *dst_2 = p->key_2;
457 p->ret = heap_cread_item(vp, &dst_2, &p->mode_2);
465 p->sz_2 = dst_2 - p->key_2;
466 if (p->sz_1 == p->sz_2 && memcmp(p->key_1, p->key_2, p->sz_1) == 0)
468 if (p->mode_2) /* adjust level according to deletes/inserts */
477 /* all the same. new round .. */
479 p->mode_1 = p->mode_2;
480 memcpy (p->key_1, p->key_2, p->sz_1);
482 level = 1; /* insert */
484 level = -1; /* delete */
487 /* outcome is insert (1) or delete (0) depending on final level */
492 memcpy (*dst, p->key_1, p->sz_1);
495 pkey(*dst, *insertMode); fflush(stdout);
501 int heap_cread_item (void *vp, char **dst, int *insertMode)
503 struct heap_cread_info *p = (struct heap_cread_info *) vp;
504 struct heap_info *hi = p->hi;
506 if (p->first_in_list)
508 *insertMode = p->key[0];
509 memcpy (*dst, p->key+1, sizeof(struct it_key));
512 pkey(*dst, *insertMode);
514 (*dst) += sizeof(struct it_key);
515 p->first_in_list = 0;
518 strcpy (p->prev_name, p->cur_name);
519 if (!(p->more = heap_read_one (hi, p->cur_name, p->key)))
521 if (*p->cur_name && strcmp (p->cur_name, p->prev_name))
523 p->first_in_list = 1;
526 *insertMode = p->key[0];
527 memcpy (*dst, p->key+1, sizeof(struct it_key));
530 pkey(*dst, *insertMode);
532 (*dst) += sizeof(struct it_key);
536 int heap_inpc (struct heap_info *hi)
538 struct heap_cread_info hci;
539 ISAMC_I *isamc_i = (ISAMC_I *) xmalloc (sizeof(*isamc_i));
541 hci.key = (char *) xmalloc (KEY_SIZE);
542 hci.key_1 = (char *) xmalloc (KEY_SIZE);
543 hci.key_2 = (char *) xmalloc (KEY_SIZE);
545 hci.first_in_list = 1;
547 hci.more = heap_read_one (hi, hci.cur_name, hci.key);
549 isamc_i->clientData = &hci;
550 isamc_i->read_item = heap_cread_item2;
554 char this_name[INP_NAME_MAX];
555 ISAMC_P isamc_p, isamc_p2;
558 strcpy (this_name, hci.cur_name);
559 assert (hci.cur_name[1]);
561 if ((dict_info = dict_lookup (hi->reg->dict, hci.cur_name)))
563 memcpy (&isamc_p, dict_info+1, sizeof(ISAMC_P));
564 isamc_p2 = isc_merge (hi->reg->isamc, isamc_p, isamc_i);
568 if (!dict_delete (hi->reg->dict, this_name))
574 if (isamc_p2 != isamc_p)
575 dict_insert (hi->reg->dict, this_name,
576 sizeof(ISAMC_P), &isamc_p2);
581 isamc_p = isc_merge (hi->reg->isamc, 0, isamc_i);
583 dict_insert (hi->reg->dict, this_name, sizeof(ISAMC_P), &isamc_p);
594 /* for debugging only */
595 static void print_dict_item (ZebraMaps zm, const char *s)
598 char keybuf[IT_MAX_WORD+1];
600 const char *from = s + 2;
604 const char *res = zebra_maps_output (zm, reg_type, &from);
612 yaz_log (LOG_LOG, "%s", keybuf);
616 int heap_inpb (struct heap_info *hi)
618 struct heap_cread_info hci;
619 ISAMC_I *isamc_i = (ISAMC_I *) xmalloc (sizeof(*isamc_i));
621 hci.key = (char *) xmalloc (KEY_SIZE);
622 hci.key_1 = (char *) xmalloc (KEY_SIZE);
623 hci.key_2 = (char *) xmalloc (KEY_SIZE);
625 hci.first_in_list = 1;
627 hci.more = heap_read_one (hi, hci.cur_name, hci.key);
629 isamc_i->clientData = &hci;
630 isamc_i->read_item = heap_cread_item2;
634 char this_name[INP_NAME_MAX];
635 ISAMC_P isamc_p, isamc_p2;
638 strcpy (this_name, hci.cur_name);
639 assert (hci.cur_name[1]);
643 print_dict_item (hi->reg->zebra_maps, hci.cur_name);
645 if ((dict_info = dict_lookup (hi->reg->dict, hci.cur_name)))
647 memcpy (&isamc_p, dict_info+1, sizeof(ISAMC_P));
648 isamc_p2 = isamb_merge (hi->reg->isamb, isamc_p, isamc_i);
652 if (!dict_delete (hi->reg->dict, this_name))
658 if (isamc_p2 != isamc_p)
659 dict_insert (hi->reg->dict, this_name,
660 sizeof(ISAMC_P), &isamc_p2);
665 isamc_p = isamb_merge (hi->reg->isamb, 0, isamc_i);
667 dict_insert (hi->reg->dict, this_name, sizeof(ISAMC_P), &isamc_p);
677 int heap_inpd (struct heap_info *hi)
679 struct heap_cread_info hci;
680 ISAMD_I isamd_i = (ISAMD_I) xmalloc (sizeof(*isamd_i));
682 hci.key = (char *) xmalloc (KEY_SIZE);
683 hci.key_1 = (char *) xmalloc (KEY_SIZE);
684 hci.key_2 = (char *) xmalloc (KEY_SIZE);
686 hci.first_in_list = 1;
688 hci.more = heap_read_one (hi, hci.cur_name, hci.key);
690 isamd_i->clientData = &hci;
691 isamd_i->read_item = heap_cread_item;
695 char this_name[INP_NAME_MAX];
697 char dictentry[ISAMD_MAX_DICT_LEN+1];
700 strcpy (this_name, hci.cur_name);
702 /* print_dict_item (hi->reg->zebra_maps, hci.cur_name); */
703 /*!*/ /* FIXME: depend on isamd-debug */
705 assert (hci.cur_name[1]);
707 if ((dict_info = dict_lookup (hi->reg->dict, hci.cur_name)))
709 dictlen=dict_info[0];
710 memcpy (dictentry, dict_info+1, dictlen );
712 logf(LOG_LOG,"dictentry before. len=%d: %d %d %d %d %d %d %d %d %d",
713 dictlen,dictentry[0], dictentry[1], dictentry[2],
714 dictentry[3], dictentry[4], dictentry[5],
715 dictentry[6], dictentry[7], dictentry[8]); /*!*/
717 dictlen= isamd_append(hi->reg->isamd, dictentry, dictlen, isamd_i);
718 /* logf dictentry after */
722 if ( (dictlen!=dict_info[0]) ||
723 (0!=memcmp(dictentry, dict_info+1, dictlen)) )
725 dict_insert(hi->reg->dict, this_name,
732 if (!dict_delete (hi->reg->dict, this_name))
734 logf (LOG_FATAL, "dict_delete failed");
742 memset (dictentry, '\0', ISAMD_MAX_DICT_LEN);
743 dictlen= isamd_append(hi->reg->isamd, dictentry, dictlen, isamd_i);
744 /* logf dictentry first */
747 dict_insert(hi->reg->dict, this_name,
758 int heap_inp (struct heap_info *hi)
761 char next_name[INP_NAME_MAX];
762 char cur_name[INP_NAME_MAX];
763 int key_buf_size = INP_BUF_START;
769 next_key = (char *) xmalloc (KEY_SIZE);
770 key_buf = (char *) xmalloc (key_buf_size);
771 more = heap_read_one (hi, cur_name, key_buf);
772 while (more) /* EOF ? */
775 key_buf_ptr = KEY_SIZE;
778 if (!(more = heap_read_one (hi, next_name, next_key)))
780 if (*next_name && strcmp (next_name, cur_name))
782 memcpy (key_buf + key_buf_ptr, next_key, KEY_SIZE);
783 key_buf_ptr += KEY_SIZE;
784 if (key_buf_ptr+(int) KEY_SIZE >= key_buf_size)
787 new_key_buf = (char *) xmalloc (key_buf_size + INP_BUF_ADD);
788 memcpy (new_key_buf, key_buf, key_buf_size);
789 key_buf_size += INP_BUF_ADD;
791 key_buf = new_key_buf;
795 nmemb = key_buf_ptr / KEY_SIZE;
796 assert (nmemb * (int) KEY_SIZE == key_buf_ptr);
797 if ((info = dict_lookup (hi->reg->dict, cur_name)))
799 ISAM_P isam_p, isam_p2;
800 memcpy (&isam_p, info+1, sizeof(ISAM_P));
801 isam_p2 = is_merge (hi->reg->isam, isam_p, nmemb, key_buf);
805 if (!dict_delete (hi->reg->dict, cur_name))
811 if (isam_p2 != isam_p)
812 dict_insert (hi->reg->dict, cur_name,
813 sizeof(ISAM_P), &isam_p2);
820 isam_p = is_merge (hi->reg->isam, 0, nmemb, key_buf);
821 dict_insert (hi->reg->dict, cur_name, sizeof(ISAM_P), &isam_p);
823 memcpy (key_buf, next_key, KEY_SIZE);
824 strcpy (cur_name, next_name);
829 int heap_inps (struct heap_info *hi)
831 struct heap_cread_info hci;
832 ISAMS_I isams_i = (ISAMS_I) xmalloc (sizeof(*isams_i));
834 hci.key = (char *) xmalloc (KEY_SIZE);
835 hci.key_1 = (char *) xmalloc (KEY_SIZE);
836 hci.key_2 = (char *) xmalloc (KEY_SIZE);
837 hci.first_in_list = 1;
840 hci.more = heap_read_one (hi, hci.cur_name, hci.key);
842 isams_i->clientData = &hci;
843 isams_i->read_item = heap_cread_item;
847 char this_name[INP_NAME_MAX];
851 strcpy (this_name, hci.cur_name);
852 assert (hci.cur_name[1]);
854 if (!(dict_info = dict_lookup (hi->reg->dict, hci.cur_name)))
856 isams_p = isams_merge (hi->reg->isams, isams_i);
858 dict_insert (hi->reg->dict, this_name, sizeof(ISAMS_P), &isams_p);
862 logf (LOG_FATAL, "isams doesn't support this kind of update");
870 struct progressInfo {
877 void progressFunc (struct key_file *keyp, void *info)
879 struct progressInfo *p = (struct progressInfo *) info;
880 time_t now, remaining;
882 if (keyp->buf_size <= 0 || p->totalBytes <= 0)
886 if (now >= p->lastTime+10)
889 remaining = (time_t) ((now - p->startTime)*
890 ((double) p->totalBytes/p->totalOffset - 1.0));
891 if (remaining <= 130)
892 logf (LOG_LOG, "Merge %2.1f%% completed; %ld seconds remaining",
893 (100.0*p->totalOffset) / p->totalBytes, (long) remaining);
895 logf (LOG_LOG, "Merge %2.1f%% completed; %ld minutes remaining",
896 (100.0*p->totalOffset) / p->totalBytes, (long) remaining/60);
898 p->totalOffset += keyp->buf_size;
905 void zebra_index_merge (ZebraHandle zh)
907 struct key_file **kf = 0;
910 struct heap_info *hi;
911 struct progressInfo progressInfo;
912 int nkeys = zh->reg->key_file_no;
915 logf (LOG_DEBUG, " index_merge called with nk=%d b=%p",
916 nkeys, zh->reg->key_buf);
917 if ( (nkeys==0) && (zh->reg->key_buf==0) )
918 return; /* nothing to merge - probably flush after end-trans */
920 usefile = (nkeys!=0);
930 extract_get_fname_tmp (zh, fname, nkeys+1);
931 if (access (fname, R_OK) == -1)
938 kf = (struct key_file **) xmalloc ((1+nkeys) * sizeof(*kf));
939 progressInfo.totalBytes = 0;
940 progressInfo.totalOffset = 0;
941 time (&progressInfo.startTime);
942 time (&progressInfo.lastTime);
943 for (i = 1; i<=nkeys; i++)
945 kf[i] = key_file_init (i, 8192, zh->res);
946 kf[i]->readHandler = progressFunc;
947 kf[i]->readInfo = &progressInfo;
948 progressInfo.totalBytes += kf[i]->length;
949 progressInfo.totalOffset += kf[i]->buf_size;
951 hi = key_heap_init (nkeys, key_qsort_compare);
954 for (i = 1; i<=nkeys; i++)
955 if ((r = key_file_read (kf[i], rbuf)))
956 key_heap_insert (hi, rbuf, r, kf[i]);
959 { /* do not use file, read straight from buffer */
960 hi = key_heap_init_buff (zh,key_qsort_compare);
976 for (i = 1; i<=nkeys; i++)
978 extract_get_fname_tmp (zh, rbuf, i);
981 for (i = 1; i<=nkeys; i++)
982 key_file_destroy (kf[i]);
985 if (hi->no_iterations)
986 { /* do not log if nothing happened */
987 logf (LOG_LOG, "Iterations . . .%7d", hi->no_iterations);
988 logf (LOG_LOG, "Distinct words .%7d", hi->no_diffs);
989 logf (LOG_LOG, "Updates. . . . .%7d", hi->no_updates);
990 logf (LOG_LOG, "Deletions. . . .%7d", hi->no_deletions);
991 logf (LOG_LOG, "Insertions . . .%7d", hi->no_insertions);
993 zh->reg->key_file_no = 0;
995 key_heap_destroy (hi, nkeys);