1 /* $Id: kinput.c,v 1.59.2.1 2005-09-09 14:22:12 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);
210 while ((key[i++] = key_file_getc (f)))
212 strcpy (f->prev_name, key);
219 for (j = 0; j<i && j < 20; j++)
220 sprintf(s+strlen(s), "%02X", key[j]&0xff);
222 sprintf(s+strlen(s), " ..");
223 yaz_log(LOG_LOG, "key hex=%s", s);
226 d = key_file_decode (f);
229 itkey.sysno = d + f->sysno;
232 f->sysno = itkey.sysno;
235 d = key_file_decode (f);
236 itkey.seqno = d + f->seqno;
238 yaz_log(LOG_LOG, "key sysno=%d seqno=%d cmd=%d", itkey.sysno, itkey.seqno,
241 f->seqno = itkey.seqno;
242 memcpy (key + i, &itkey, sizeof(struct it_key));
243 return i + sizeof (struct it_key);
248 struct key_file **file;
253 int (*cmp)(const void *p1, const void *p2);
254 struct zebra_register *reg;
255 ZebraHandle zh; /* only used for raw reading that bypasses the heaps */
263 static struct heap_info *key_heap_malloc()
264 { /* malloc and clear it */
265 struct heap_info *hi;
266 hi = (struct heap_info *) xmalloc (sizeof(*hi));
275 hi->no_deletions = 0;
276 hi->no_insertions = 0;
277 hi->no_iterations = 0;
281 struct heap_info *key_heap_init (int nkeys,
282 int (*cmp)(const void *p1, const void *p2))
284 struct heap_info *hi;
287 hi = key_heap_malloc();
288 hi->info.file = (struct key_file **)
289 xmalloc (sizeof(*hi->info.file) * (1+nkeys));
290 hi->info.buf = (char **) xmalloc (sizeof(*hi->info.buf) * (1+nkeys));
291 hi->ptr = (int *) xmalloc (sizeof(*hi->ptr) * (1+nkeys));
293 for (i = 0; i<= nkeys; i++)
296 hi->info.buf[i] = (char *) xmalloc (INP_NAME_MAX);
301 struct heap_info *key_heap_init_buff ( ZebraHandle zh,
302 int (*cmp)(const void *p1, const void *p2))
304 struct heap_info *hi=key_heap_malloc();
310 void key_heap_destroy (struct heap_info *hi, int nkeys)
313 yaz_log (LOG_DEBUG, "key_heap_destroy");
314 yaz_log (LOG_DEBUG, "key_heap_destroy nk=%d",nkeys);
316 for (i = 0; i<=nkeys; i++)
317 xfree (hi->info.buf[i]);
319 xfree (hi->info.buf);
321 xfree (hi->info.file);
325 static void key_heap_swap (struct heap_info *hi, int i1, int i2)
330 hi->ptr[i1] = hi->ptr[i2];
335 static void key_heap_delete (struct heap_info *hi)
337 int cur = 1, child = 2;
339 assert (hi->heapnum > 0);
341 key_heap_swap (hi, 1, hi->heapnum);
343 while (child <= hi->heapnum) {
344 if (child < hi->heapnum &&
345 (*hi->cmp)(&hi->info.buf[hi->ptr[child]],
346 &hi->info.buf[hi->ptr[child+1]]) > 0)
348 if ((*hi->cmp)(&hi->info.buf[hi->ptr[cur]],
349 &hi->info.buf[hi->ptr[child]]) > 0)
351 key_heap_swap (hi, cur, child);
360 static void key_heap_insert (struct heap_info *hi, const char *buf, int nbytes,
365 cur = ++(hi->heapnum);
366 memcpy (hi->info.buf[hi->ptr[cur]], buf, nbytes);
367 hi->info.file[hi->ptr[cur]] = kf;
370 while (parent && (*hi->cmp)(&hi->info.buf[hi->ptr[parent]],
371 &hi->info.buf[hi->ptr[cur]]) > 0)
373 key_heap_swap (hi, cur, parent);
379 static int heap_read_one_raw (struct heap_info *hi, char *name, char *key)
381 ZebraHandle zh=hi->zh;
382 size_t ptr_i = zh->reg->ptr_i;
387 cp=(zh->reg->key_buf)[zh->reg->ptr_top - ptr_i];
388 logf (LOG_DEBUG, " raw: i=%ld top=%ld cp=%p", (long) ptr_i,
389 (long) zh->reg->ptr_top, cp);
391 memcpy(key, cp+strlen(name)+1, KEY_SIZE);
396 static int heap_read_one (struct heap_info *hi, char *name, char *key)
399 char rbuf[INP_NAME_MAX];
402 if (hi->zh) /* bypass the heap stuff, we have a readymade buffer */
403 return heap_read_one_raw(hi, name, key);
408 strcpy (name, hi->info.buf[n]);
409 kf = hi->info.file[n];
411 memcpy (key, hi->info.buf[n] + r+1, KEY_SIZE);
412 key_heap_delete (hi);
413 if ((r = key_file_read (kf, rbuf)))
414 key_heap_insert (hi, rbuf, r, kf);
422 static void pkey(const char *b, int mode)
424 struct it_key *key = (struct it_key *) b;
425 printf ("%c %d:%d\n", mode + 48, key->sysno, key->seqno);
429 struct heap_cread_info {
430 char prev_name[INP_NAME_MAX];
431 char cur_name[INP_NAME_MAX];
436 struct heap_info *hi;
442 static int heap_cread_item (void *vp, char **dst, int *insertMode);
444 int heap_cread_item2 (void *vp, char **dst, int *insertMode)
446 struct heap_cread_info *p = (struct heap_cread_info *) vp;
449 if (p->ret == 0) /* lookahead was 0?. Return that in read next round */
454 else if (p->ret == -1) /* Must read new item ? */
456 char *dst_1 = p->key_1;
457 p->ret = heap_cread_item(vp, &dst_1, &p->mode_1);
458 p->sz_1 = dst_1 - p->key_1;
461 { /* lookahead in 2 . Now in 1. */
463 p->mode_1 = p->mode_2;
464 memcpy (p->key_1, p->key_2, p->sz_2);
467 level = 1; /* insert */
469 level = -1; /* delete */
472 char *dst_2 = p->key_2;
473 p->ret = heap_cread_item(vp, &dst_2, &p->mode_2);
481 p->sz_2 = dst_2 - p->key_2;
482 if (p->sz_1 == p->sz_2 && memcmp(p->key_1, p->key_2, p->sz_1) == 0)
484 if (p->mode_2) /* adjust level according to deletes/inserts */
493 /* all the same. new round .. */
495 p->mode_1 = p->mode_2;
496 memcpy (p->key_1, p->key_2, p->sz_1);
498 level = 1; /* insert */
500 level = -1; /* delete */
503 /* outcome is insert (1) or delete (0) depending on final level */
508 memcpy (*dst, p->key_1, p->sz_1);
511 pkey(*dst, *insertMode); fflush(stdout);
517 int heap_cread_item (void *vp, char **dst, int *insertMode)
519 struct heap_cread_info *p = (struct heap_cread_info *) vp;
520 struct heap_info *hi = p->hi;
522 if (p->first_in_list)
524 *insertMode = p->key[0];
525 memcpy (*dst, p->key+1, sizeof(struct it_key));
528 pkey(*dst, *insertMode);
530 (*dst) += sizeof(struct it_key);
531 p->first_in_list = 0;
534 strcpy (p->prev_name, p->cur_name);
535 if (!(p->more = heap_read_one (hi, p->cur_name, p->key)))
537 if (*p->cur_name && strcmp (p->cur_name, p->prev_name))
539 p->first_in_list = 1;
542 *insertMode = p->key[0];
543 memcpy (*dst, p->key+1, sizeof(struct it_key));
546 pkey(*dst, *insertMode);
548 (*dst) += sizeof(struct it_key);
552 int heap_inpc (struct heap_info *hi)
554 struct heap_cread_info hci;
555 ISAMC_I *isamc_i = (ISAMC_I *) xmalloc (sizeof(*isamc_i));
557 hci.key = (char *) xmalloc (KEY_SIZE);
558 hci.key_1 = (char *) xmalloc (KEY_SIZE);
559 hci.key_2 = (char *) xmalloc (KEY_SIZE);
561 hci.first_in_list = 1;
563 hci.more = heap_read_one (hi, hci.cur_name, hci.key);
565 isamc_i->clientData = &hci;
566 isamc_i->read_item = heap_cread_item2;
570 char this_name[INP_NAME_MAX];
571 ISAMC_P isamc_p, isamc_p2;
574 strcpy (this_name, hci.cur_name);
575 assert (hci.cur_name[1]);
577 if ((dict_info = dict_lookup (hi->reg->dict, hci.cur_name)))
579 memcpy (&isamc_p, dict_info+1, sizeof(ISAMC_P));
580 isamc_p2 = isc_merge (hi->reg->isamc, isamc_p, isamc_i);
584 if (!dict_delete (hi->reg->dict, this_name))
590 if (isamc_p2 != isamc_p)
591 dict_insert (hi->reg->dict, this_name,
592 sizeof(ISAMC_P), &isamc_p2);
597 isamc_p = isc_merge (hi->reg->isamc, 0, isamc_i);
599 dict_insert (hi->reg->dict, this_name, sizeof(ISAMC_P), &isamc_p);
610 /* for debugging only */
611 static void print_dict_item (ZebraMaps zm, const char *s)
614 char keybuf[IT_MAX_WORD+1];
616 const char *from = s + 2;
620 const char *res = zebra_maps_output (zm, reg_type, &from);
628 yaz_log (LOG_LOG, "%s", keybuf);
632 int heap_inpb (struct heap_info *hi)
634 struct heap_cread_info hci;
635 ISAMC_I *isamc_i = (ISAMC_I *) xmalloc (sizeof(*isamc_i));
637 hci.key = (char *) xmalloc (KEY_SIZE);
638 hci.key_1 = (char *) xmalloc (KEY_SIZE);
639 hci.key_2 = (char *) xmalloc (KEY_SIZE);
641 hci.first_in_list = 1;
643 hci.more = heap_read_one (hi, hci.cur_name, hci.key);
645 isamc_i->clientData = &hci;
646 isamc_i->read_item = heap_cread_item2;
650 char this_name[INP_NAME_MAX];
651 ISAMC_P isamc_p, isamc_p2;
654 strcpy (this_name, hci.cur_name);
655 assert (hci.cur_name[1]);
659 print_dict_item (hi->reg->zebra_maps, hci.cur_name);
661 if ((dict_info = dict_lookup (hi->reg->dict, hci.cur_name)))
663 memcpy (&isamc_p, dict_info+1, sizeof(ISAMC_P));
664 isamc_p2 = isamb_merge (hi->reg->isamb, isamc_p, isamc_i);
668 if (!dict_delete (hi->reg->dict, this_name))
674 if (isamc_p2 != isamc_p)
675 dict_insert (hi->reg->dict, this_name,
676 sizeof(ISAMC_P), &isamc_p2);
681 isamc_p = isamb_merge (hi->reg->isamb, 0, isamc_i);
683 dict_insert (hi->reg->dict, this_name, sizeof(ISAMC_P), &isamc_p);
693 int heap_inpd (struct heap_info *hi)
695 struct heap_cread_info hci;
696 ISAMD_I isamd_i = (ISAMD_I) xmalloc (sizeof(*isamd_i));
698 hci.key = (char *) xmalloc (KEY_SIZE);
699 hci.key_1 = (char *) xmalloc (KEY_SIZE);
700 hci.key_2 = (char *) xmalloc (KEY_SIZE);
702 hci.first_in_list = 1;
704 hci.more = heap_read_one (hi, hci.cur_name, hci.key);
706 isamd_i->clientData = &hci;
707 isamd_i->read_item = heap_cread_item;
711 char this_name[INP_NAME_MAX];
713 char dictentry[ISAMD_MAX_DICT_LEN+1];
716 strcpy (this_name, hci.cur_name);
718 /* print_dict_item (hi->reg->zebra_maps, hci.cur_name); */
719 /*!*/ /* FIXME: depend on isamd-debug */
721 assert (hci.cur_name[1]);
723 if ((dict_info = dict_lookup (hi->reg->dict, hci.cur_name)))
725 dictlen=dict_info[0];
726 memcpy (dictentry, dict_info+1, dictlen );
728 logf(LOG_LOG,"dictentry before. len=%d: %d %d %d %d %d %d %d %d %d",
729 dictlen,dictentry[0], dictentry[1], dictentry[2],
730 dictentry[3], dictentry[4], dictentry[5],
731 dictentry[6], dictentry[7], dictentry[8]); /*!*/
733 dictlen= isamd_append(hi->reg->isamd, dictentry, dictlen, isamd_i);
734 /* logf dictentry after */
738 if ( (dictlen!=dict_info[0]) ||
739 (0!=memcmp(dictentry, dict_info+1, dictlen)) )
741 dict_insert(hi->reg->dict, this_name,
748 if (!dict_delete (hi->reg->dict, this_name))
750 logf (LOG_FATAL, "dict_delete failed");
758 memset (dictentry, '\0', ISAMD_MAX_DICT_LEN);
759 dictlen= isamd_append(hi->reg->isamd, dictentry, dictlen, isamd_i);
760 /* logf dictentry first */
763 dict_insert(hi->reg->dict, this_name,
774 int heap_inp (struct heap_info *hi)
777 char next_name[INP_NAME_MAX];
778 char cur_name[INP_NAME_MAX];
779 int key_buf_size = INP_BUF_START;
785 next_key = (char *) xmalloc (KEY_SIZE);
786 key_buf = (char *) xmalloc (key_buf_size);
787 more = heap_read_one (hi, cur_name, key_buf);
788 while (more) /* EOF ? */
791 key_buf_ptr = KEY_SIZE;
794 if (!(more = heap_read_one (hi, next_name, next_key)))
796 if (*next_name && strcmp (next_name, cur_name))
798 memcpy (key_buf + key_buf_ptr, next_key, KEY_SIZE);
799 key_buf_ptr += KEY_SIZE;
800 if (key_buf_ptr+(int) KEY_SIZE >= key_buf_size)
803 new_key_buf = (char *) xmalloc (key_buf_size + INP_BUF_ADD);
804 memcpy (new_key_buf, key_buf, key_buf_size);
805 key_buf_size += INP_BUF_ADD;
807 key_buf = new_key_buf;
811 nmemb = key_buf_ptr / KEY_SIZE;
812 assert (nmemb * (int) KEY_SIZE == key_buf_ptr);
813 if ((info = dict_lookup (hi->reg->dict, cur_name)))
815 ISAM_P isam_p, isam_p2;
816 memcpy (&isam_p, info+1, sizeof(ISAM_P));
817 isam_p2 = is_merge (hi->reg->isam, isam_p, nmemb, key_buf);
821 if (!dict_delete (hi->reg->dict, cur_name))
827 if (isam_p2 != isam_p)
828 dict_insert (hi->reg->dict, cur_name,
829 sizeof(ISAM_P), &isam_p2);
836 isam_p = is_merge (hi->reg->isam, 0, nmemb, key_buf);
837 dict_insert (hi->reg->dict, cur_name, sizeof(ISAM_P), &isam_p);
839 memcpy (key_buf, next_key, KEY_SIZE);
840 strcpy (cur_name, next_name);
845 int heap_inps (struct heap_info *hi)
847 struct heap_cread_info hci;
848 ISAMS_I isams_i = (ISAMS_I) xmalloc (sizeof(*isams_i));
850 hci.key = (char *) xmalloc (KEY_SIZE);
851 hci.key_1 = (char *) xmalloc (KEY_SIZE);
852 hci.key_2 = (char *) xmalloc (KEY_SIZE);
853 hci.first_in_list = 1;
856 hci.more = heap_read_one (hi, hci.cur_name, hci.key);
858 isams_i->clientData = &hci;
859 isams_i->read_item = heap_cread_item;
863 char this_name[INP_NAME_MAX];
867 strcpy (this_name, hci.cur_name);
868 assert (hci.cur_name[1]);
870 if (!(dict_info = dict_lookup (hi->reg->dict, hci.cur_name)))
872 isams_p = isams_merge (hi->reg->isams, isams_i);
874 dict_insert (hi->reg->dict, this_name, sizeof(ISAMS_P), &isams_p);
878 logf (LOG_FATAL, "isams doesn't support this kind of update");
886 struct progressInfo {
893 void progressFunc (struct key_file *keyp, void *info)
895 struct progressInfo *p = (struct progressInfo *) info;
896 time_t now, remaining;
898 if (keyp->buf_size <= 0 || p->totalBytes <= 0)
902 if (now >= p->lastTime+10)
905 remaining = (time_t) ((now - p->startTime)*
906 ((double) p->totalBytes/p->totalOffset - 1.0));
907 if (remaining <= 130)
908 logf (LOG_LOG, "Merge %2.1f%% completed; %ld seconds remaining",
909 (100.0*p->totalOffset) / p->totalBytes, (long) remaining);
911 logf (LOG_LOG, "Merge %2.1f%% completed; %ld minutes remaining",
912 (100.0*p->totalOffset) / p->totalBytes, (long) remaining/60);
914 p->totalOffset += keyp->buf_size;
921 void zebra_index_merge (ZebraHandle zh)
923 struct key_file **kf = 0;
926 struct heap_info *hi;
927 struct progressInfo progressInfo;
928 int nkeys = zh->reg->key_file_no;
931 logf (LOG_DEBUG, " index_merge called with nk=%d b=%p",
932 nkeys, zh->reg->key_buf);
933 if ( (nkeys==0) && (zh->reg->key_buf==0) )
934 return; /* nothing to merge - probably flush after end-trans */
936 usefile = (nkeys!=0);
946 extract_get_fname_tmp (zh, fname, nkeys+1);
947 if (access (fname, R_OK) == -1)
954 kf = (struct key_file **) xmalloc ((1+nkeys) * sizeof(*kf));
955 progressInfo.totalBytes = 0;
956 progressInfo.totalOffset = 0;
957 time (&progressInfo.startTime);
958 time (&progressInfo.lastTime);
959 for (i = 1; i<=nkeys; i++)
961 kf[i] = key_file_init (i, 8192, zh->res);
962 kf[i]->readHandler = progressFunc;
963 kf[i]->readInfo = &progressInfo;
964 progressInfo.totalBytes += kf[i]->length;
965 progressInfo.totalOffset += kf[i]->buf_size;
967 hi = key_heap_init (nkeys, key_qsort_compare);
970 for (i = 1; i<=nkeys; i++)
971 if ((r = key_file_read (kf[i], rbuf)))
972 key_heap_insert (hi, rbuf, r, kf[i]);
975 { /* do not use file, read straight from buffer */
976 hi = key_heap_init_buff (zh,key_qsort_compare);
992 for (i = 1; i<=nkeys; i++)
994 extract_get_fname_tmp (zh, rbuf, i);
997 for (i = 1; i<=nkeys; i++)
998 key_file_destroy (kf[i]);
1001 if (hi->no_iterations)
1002 { /* do not log if nothing happened */
1003 logf (LOG_LOG, "Iterations . . .%7d", hi->no_iterations);
1004 logf (LOG_LOG, "Distinct words .%7d", hi->no_diffs);
1005 logf (LOG_LOG, "Updates. . . . .%7d", hi->no_updates);
1006 logf (LOG_LOG, "Deletions. . . .%7d", hi->no_deletions);
1007 logf (LOG_LOG, "Insertions . . .%7d", hi->no_insertions);
1009 zh->reg->key_file_no = 0;
1011 key_heap_destroy (hi, nkeys);