X-Git-Url: http://git.indexdata.com/?p=idzebra-moved-to-github.git;a=blobdiff_plain;f=index%2Fkinput.c;h=b9d7f066da1b2e9c7b3cce036f3bc84e1723cb63;hp=dd7583e09c27cbfb3e48455054393bf53a0d206f;hb=7e75317bed8eecabcb57e59b16093a32238738e2;hpb=226fb73f42a86cc30ff4f27eb452ab3f6c19ae01 diff --git a/index/kinput.c b/index/kinput.c index dd7583e..b9d7f06 100644 --- a/index/kinput.c +++ b/index/kinput.c @@ -1,63 +1,38 @@ /* - * Copyright (C) 1994-1995, Index Data I/S + * Copyright (C) 1994-2002, Index Data * All rights reserved. - * Sebastian Hammer, Adam Dickmeiss + * Sebastian Hammer, Adam Dickmeiss, Heikki Levanto * - * $Log: kinput.c,v $ - * Revision 1.10 1995-12-06 12:41:22 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 1995/10/10 12:24:39 adam - * Temporary sort files are compressed. - * - * Revision 1.8 1995/10/04 16:57:19 adam - * Key input and merge sort in one pass. - * - * Revision 1.7 1995/10/02 15:18:52 adam - * New member in recRetrieveCtrl: diagnostic. - * - * Revision 1.6 1995/09/29 15:51:56 adam - * First work on multi-way read. - * - * Revision 1.5 1995/09/29 14:01:43 adam - * Bug fixes. - * - * Revision 1.4 1995/09/28 14:22:57 adam - * Sort uses smaller temporary files. - * - * Revision 1.3 1995/09/06 16:11:17 adam - * Option: only one word key per file. - * - * Revision 1.2 1995/09/04 12:33:42 adam - * Various cleanup. YAZ util used instead. - * - * Revision 1.1 1995/09/04 09:10:37 adam - * More work on index add/del/update. - * Merge sort implemented. - * Initial work on z39 server. + * $Id: kinput.c,v 1.45 2002-02-20 17:30:01 adam Exp $ * + * Bugs + * - Allocates a lot of memory for the merge process, but never releases it. + * Doesn't matter, as the program terminates soon after. + */ - + #include +#ifdef WIN32 +#include +#else #include +#endif #include #include #include -#include #include #include "index.h" +#include "zserver.h" #define KEY_SIZE (1+sizeof(struct it_key)) -#define INP_NAME_MAX 8192 +#define INP_NAME_MAX 768 #define INP_BUF_START 60000 #define INP_BUF_ADD 400000 static int no_diffs = 0; static int no_updates = 0; +static int no_deletions = 0; static int no_insertions = 0; static int no_iterations = 0; @@ -71,24 +46,57 @@ struct key_file { char *prev_name; /* last word read */ int sysno; /* last sysno */ int seqno; /* last seqno */ + off_t length; /* length of file */ + /* handler invoked in each read */ + void (*readHandler)(struct key_file *keyp, void *rinfo); + void *readInfo; + Res res; }; +void getFnameTmp (Res res, char *fname, int no) +{ + const char *pre; + + pre = res_get_def (res, "keyTmpDir", "."); + sprintf (fname, "%s/key%d.tmp", pre, no); +} + +void extract_get_fname_tmp (ZebraHandle zh, char *fname, int no) +{ + const char *pre; + + pre = res_get_def (zh->service->res, "keyTmpDir", "."); + sprintf (fname, "%s/key%d.tmp", pre, no); +} + void key_file_chunk_read (struct key_file *f) { - int nr = 0, r, fd; - char fname[256]; - sprintf (fname, TEMP_FNAME, f->no); - fd = open (fname, O_RDONLY); + int nr = 0, r = 0, fd; + char fname[1024]; + getFnameTmp (f->res, fname, f->no); + fd = open (fname, O_BINARY|O_RDONLY); + + f->buf_ptr = 0; + f->buf_size = 0; if (fd == -1) { - logf (LOG_FATAL|LOG_ERRNO, "cannot open %s", fname); - exit (1); + logf (LOG_WARN|LOG_ERRNO, "cannot open %s", fname); + return ; + } + if (!f->length) + { + if ((f->length = lseek (fd, 0L, SEEK_END)) == (off_t) -1) + { + logf (LOG_WARN|LOG_ERRNO, "cannot seek %s", fname); + close (fd); + return ; + } } - logf (LOG_LOG, "reading chunk from %s", fname); if (lseek (fd, f->offset, SEEK_SET) == -1) { - logf (LOG_FATAL|LOG_ERRNO, "cannot seek %s", fname); - exit (1); + logf (LOG_WARN|LOG_ERRNO, "cannot seek %s", fname); + close(fd); + return ; } while (f->chunk - nr > 0) { @@ -99,26 +107,31 @@ void key_file_chunk_read (struct key_file *f) } if (r == -1) { - logf (LOG_FATAL|LOG_ERRNO, "read of %s", fname); - exit (1); + logf (LOG_WARN|LOG_ERRNO, "read of %s", fname); + close (fd); + return; } f->buf_size = nr; - f->buf_ptr = 0; + if (f->readHandler) + (*f->readHandler)(f, f->readInfo); close (fd); } -struct key_file *key_file_init (int no, int chunk) +struct key_file *key_file_init (int no, int chunk, Res res) { struct key_file *f; - f = xmalloc (sizeof(*f)); + f = (struct key_file *) xmalloc (sizeof(*f)); + f->res = res; f->sysno = 0; f->seqno = 0; f->no = no; f->chunk = chunk; f->offset = 0; - f->buf = xmalloc (f->chunk); - f->prev_name = xmalloc (256); + f->length = 0; + f->readHandler = NULL; + f->buf = (unsigned char *) xmalloc (f->chunk); + f->prev_name = (char *) xmalloc (INP_NAME_MAX); *f->prev_name = '\0'; key_file_chunk_read (f); return f; @@ -210,6 +223,13 @@ struct heap_info { int heapnum; int *ptr; int (*cmp)(const void *p1, const void *p2); + Dict dict; + ISAMS isams; +#if ZMBOL + ISAM isam; + ISAMC isamc; + ISAMD isamd; +#endif }; struct heap_info *key_heap_init (int nkeys, @@ -218,16 +238,17 @@ struct heap_info *key_heap_init (int nkeys, struct heap_info *hi; int i; - hi = xmalloc (sizeof(*hi)); - hi->info.file = xmalloc (sizeof(*hi->info.file) * (1+nkeys)); - hi->info.buf = xmalloc (sizeof(*hi->info.buf) * (1+nkeys)); + hi = (struct heap_info *) xmalloc (sizeof(*hi)); + hi->info.file = (struct key_file **) + xmalloc (sizeof(*hi->info.file) * (1+nkeys)); + hi->info.buf = (char **) xmalloc (sizeof(*hi->info.buf) * (1+nkeys)); hi->heapnum = 0; - hi->ptr = xmalloc (sizeof(*hi->ptr) * (1+nkeys)); + hi->ptr = (int *) xmalloc (sizeof(*hi->ptr) * (1+nkeys)); hi->cmp = cmp; for (i = 0; i<= nkeys; i++) { hi->ptr[i] = i; - hi->info.buf[i] = xmalloc (512); + hi->info.buf[i] = (char *) xmalloc (INP_NAME_MAX); } return hi; } @@ -289,7 +310,7 @@ static void key_heap_insert (struct heap_info *hi, const char *buf, int nbytes, static int heap_read_one (struct heap_info *hi, char *name, char *key) { int n, r; - char rbuf[512]; + char rbuf[INP_NAME_MAX]; struct key_file *kf; if (!hi->heapnum) @@ -306,19 +327,159 @@ static int heap_read_one (struct heap_info *hi, char *name, char *key) return 1; } -int heap_inp (Dict dict, ISAM isam, struct heap_info *hi) +struct heap_cread_info { + char prev_name[INP_NAME_MAX]; + char cur_name[INP_NAME_MAX]; + char *key; + struct heap_info *hi; + int mode; + int more; +}; + +int heap_cread_item (void *vp, char **dst, int *insertMode) +{ + struct heap_cread_info *p = (struct heap_cread_info *) vp; + struct heap_info *hi = p->hi; + + if (p->mode == 1) + { + *insertMode = p->key[0]; + memcpy (*dst, p->key+1, sizeof(struct it_key)); + (*dst) += sizeof(struct it_key); + p->mode = 2; + return 1; + } + strcpy (p->prev_name, p->cur_name); + if (!(p->more = heap_read_one (hi, p->cur_name, p->key))) + return 0; + if (*p->cur_name && strcmp (p->cur_name, p->prev_name)) + { + p->mode = 1; + return 0; + } + *insertMode = p->key[0]; + memcpy (*dst, p->key+1, sizeof(struct it_key)); + (*dst) += sizeof(struct it_key); + return 1; +} + +#if ZMBOL +int heap_inpc (struct heap_info *hi) +{ + struct heap_cread_info hci; + ISAMC_I isamc_i = (ISAMC_I) xmalloc (sizeof(*isamc_i)); + + hci.key = (char *) xmalloc (KEY_SIZE); + hci.mode = 1; + hci.hi = hi; + hci.more = heap_read_one (hi, hci.cur_name, hci.key); + + isamc_i->clientData = &hci; + isamc_i->read_item = heap_cread_item; + + while (hci.more) + { + char this_name[INP_NAME_MAX]; + ISAMC_P isamc_p, isamc_p2; + char *dict_info; + + strcpy (this_name, hci.cur_name); + assert (hci.cur_name[1]); + no_diffs++; + if ((dict_info = dict_lookup (hi->dict, hci.cur_name))) + { + memcpy (&isamc_p, dict_info+1, sizeof(ISAMC_P)); + isamc_p2 = isc_merge (hi->isamc, isamc_p, isamc_i); + if (!isamc_p2) + { + no_deletions++; + if (!dict_delete (hi->dict, this_name)) + abort(); + } + else + { + no_updates++; + if (isamc_p2 != isamc_p) + dict_insert (hi->dict, this_name, + sizeof(ISAMC_P), &isamc_p2); + } + } + else + { + isamc_p = isc_merge (hi->isamc, 0, isamc_i); + no_insertions++; + dict_insert (hi->dict, this_name, sizeof(ISAMC_P), &isamc_p); + } + } + xfree (isamc_i); + xfree (hci.key); + return 0; +} + +int heap_inpd (struct heap_info *hi) +{ + struct heap_cread_info hci; + ISAMD_I isamd_i = (ISAMD_I) xmalloc (sizeof(*isamd_i)); + + hci.key = (char *) xmalloc (KEY_SIZE); + hci.mode = 1; + hci.hi = hi; + hci.more = heap_read_one (hi, hci.cur_name, hci.key); + + isamd_i->clientData = &hci; + isamd_i->read_item = heap_cread_item; + + while (hci.more) + { + char this_name[INP_NAME_MAX]; + ISAMD_P isamd_p, isamd_p2; + char *dict_info; + + strcpy (this_name, hci.cur_name); + assert (hci.cur_name[1]); + no_diffs++; + if ((dict_info = dict_lookup (hi->dict, hci.cur_name))) + { + memcpy (&isamd_p, dict_info+1, sizeof(ISAMD_P)); + isamd_p2 = isamd_append (hi->isamd, isamd_p, isamd_i); + if (!isamd_p2) + { + no_deletions++; + if (!dict_delete (hi->dict, this_name)) + abort(); + } + else + { + no_updates++; + if (isamd_p2 != isamd_p) + dict_insert (hi->dict, this_name, + sizeof(ISAMD_P), &isamd_p2); + } + } + else + { + isamd_p = isamd_append (hi->isamd, 0, isamd_i); + no_insertions++; + dict_insert (hi->dict, this_name, sizeof(ISAMD_P), &isamd_p); + } + } + xfree (isamd_i); + return 0; +} + +int heap_inp (struct heap_info *hi) { char *info; - char next_name[INP_NAME_MAX+1]; - char cur_name[INP_NAME_MAX+1]; + char next_name[INP_NAME_MAX]; + char cur_name[INP_NAME_MAX]; int key_buf_size = INP_BUF_START; int key_buf_ptr; char *next_key; char *key_buf; int more; - next_key = xmalloc (KEY_SIZE); - key_buf = xmalloc (key_buf_size * (KEY_SIZE)); + next_key = (char *) xmalloc (KEY_SIZE); + key_buf = (char *) xmalloc (key_buf_size); more = heap_read_one (hi, cur_name, key_buf); while (more) /* EOF ? */ { @@ -332,10 +493,10 @@ int heap_inp (Dict dict, ISAM isam, struct heap_info *hi) break; memcpy (key_buf + key_buf_ptr, next_key, KEY_SIZE); key_buf_ptr += KEY_SIZE; - if (key_buf_ptr+KEY_SIZE >= key_buf_size) + if (key_buf_ptr+(int) KEY_SIZE >= key_buf_size) { char *new_key_buf; - new_key_buf = xmalloc (key_buf_size + INP_BUF_ADD); + new_key_buf = (char *) xmalloc (key_buf_size + INP_BUF_ADD); memcpy (new_key_buf, key_buf, key_buf_size); key_buf_size += INP_BUF_ADD; xfree (key_buf); @@ -344,24 +505,31 @@ int heap_inp (Dict dict, ISAM isam, struct heap_info *hi) } no_diffs++; nmemb = key_buf_ptr / KEY_SIZE; - assert (nmemb*KEY_SIZE == key_buf_ptr); - if ((info = dict_lookup (dict, cur_name))) + assert (nmemb * (int) KEY_SIZE == key_buf_ptr); + if ((info = dict_lookup (hi->dict, cur_name))) { ISAM_P isam_p, isam_p2; - logf (LOG_DEBUG, "updating %s", cur_name); - no_updates++; memcpy (&isam_p, info+1, sizeof(ISAM_P)); - isam_p2 = is_merge (isam, isam_p, nmemb, key_buf); - if (isam_p2 != isam_p) - dict_insert (dict, cur_name, sizeof(ISAM_P), &isam_p2); + isam_p2 = is_merge (hi->isam, isam_p, nmemb, key_buf); + if (!isam_p2) + { + no_deletions++; + if (!dict_delete (hi->dict, cur_name)) + abort (); + } + else + { + no_updates++; + if (isam_p2 != isam_p) + dict_insert (hi->dict, cur_name, sizeof(ISAM_P), &isam_p2); + } } else { ISAM_P isam_p; - logf (LOG_DEBUG, "inserting %s", cur_name); no_insertions++; - isam_p = is_merge (isam, 0, nmemb, key_buf); - dict_insert (dict, cur_name, sizeof(ISAM_P), &isam_p); + isam_p = is_merge (hi->isam, 0, nmemb, key_buf); + dict_insert (hi->dict, cur_name, sizeof(ISAM_P), &isam_p); } memcpy (key_buf, next_key, KEY_SIZE); strcpy (cur_name, next_name); @@ -369,47 +537,223 @@ int heap_inp (Dict dict, ISAM isam, struct heap_info *hi) return 0; } -void key_input (const char *dict_fname, const char *isam_fname, - int nkeys, int cache) - +#endif + +int heap_inps (struct heap_info *hi) +{ + struct heap_cread_info hci; + ISAMS_I isams_i = (ISAMS_I) xmalloc (sizeof(*isams_i)); + + hci.key = (char *) xmalloc (KEY_SIZE); + hci.mode = 1; + hci.hi = hi; + hci.more = heap_read_one (hi, hci.cur_name, hci.key); + + isams_i->clientData = &hci; + isams_i->read_item = heap_cread_item; + + while (hci.more) + { + char this_name[INP_NAME_MAX]; + ISAMS_P isams_p; + char *dict_info; + + strcpy (this_name, hci.cur_name); + assert (hci.cur_name[1]); + no_diffs++; + if (!(dict_info = dict_lookup (hi->dict, hci.cur_name))) + { + isams_p = isams_merge (hi->isams, isams_i); + no_insertions++; + dict_insert (hi->dict, this_name, sizeof(ISAMS_P), &isams_p); + } + else + { + logf (LOG_FATAL, "isams doesn't support this kind of update"); + break; + } + } + xfree (isams_i); + return 0; +} + +struct progressInfo { + time_t startTime; + time_t lastTime; + off_t totalBytes; + off_t totalOffset; +}; + +void progressFunc (struct key_file *keyp, void *info) +{ + struct progressInfo *p = (struct progressInfo *) info; + time_t now, remaining; + + if (keyp->buf_size <= 0 || p->totalBytes <= 0) + return ; + time (&now); + + if (now >= p->lastTime+10) + { + p->lastTime = now; + remaining = (time_t) ((now - p->startTime)* + ((double) p->totalBytes/p->totalOffset - 1.0)); + if (remaining <= 130) + logf (LOG_LOG, "Merge %2.1f%% completed; %ld seconds remaining", + (100.0*p->totalOffset) / p->totalBytes, (long) remaining); + else + logf (LOG_LOG, "Merge %2.1f%% completed; %ld minutes remaining", + (100.0*p->totalOffset) / p->totalBytes, (long) remaining/60); + } + p->totalOffset += keyp->buf_size; +} + +#ifndef R_OK +#define R_OK 4 +#endif + +void zebra_index_merge (ZebraHandle zh) { - Dict dict; - ISAM isam; struct key_file **kf; char rbuf[1024]; int i, r; struct heap_info *hi; - - dict = dict_open (dict_fname, cache, 1); - if (!dict) + struct progressInfo progressInfo; + int nkeys = zh->key_file_no; + + if (nkeys < 0) { - logf (LOG_FATAL, "dict_open fail of `%s'", dict_fname); - exit (1); + char fname[1024]; + nkeys = 0; + while (1) + { + extract_get_fname_tmp (zh, fname, nkeys+1); + if (access (fname, R_OK) == -1) + break; + nkeys++; + } + if (!nkeys) + return ; } - isam = is_open (isam_fname, key_compare, 1, sizeof(struct it_key)); - if (!isam) + kf = (struct key_file **) xmalloc ((1+nkeys) * sizeof(*kf)); + progressInfo.totalBytes = 0; + progressInfo.totalOffset = 0; + time (&progressInfo.startTime); + time (&progressInfo.lastTime); + for (i = 1; i<=nkeys; i++) { - logf (LOG_FATAL, "is_open fail of `%s'", isam_fname); - exit (1); + kf[i] = key_file_init (i, 8192, zh->service->res); + kf[i]->readHandler = progressFunc; + kf[i]->readInfo = &progressInfo; + progressInfo.totalBytes += kf[i]->length; + progressInfo.totalOffset += kf[i]->buf_size; } - - kf = xmalloc ((1+nkeys) * sizeof(*kf)); - for (i = 1; i<=nkeys; i++) - kf[i] = key_file_init (i, 32768); - hi = key_heap_init (nkeys, key_qsort_compare); + hi->dict = zh->service->dict; + hi->isams = zh->service->isams; +#if ZMBOL + hi->isam = zh->service->isam; + hi->isamc = zh->service->isamc; + hi->isamd = zh->service->isamd; +#endif + for (i = 1; i<=nkeys; i++) if ((r = key_file_read (kf[i], rbuf))) key_heap_insert (hi, rbuf, r, kf[i]); - heap_inp (dict, isam, hi); + if (zh->service->isams) + heap_inps (hi); +#if ZMBOL + else if (zh->service->isamc) + heap_inpc (hi); + else if (zh->service->isam) + heap_inp (hi); + else if (zh->service->isamd) + heap_inpd (hi); +#endif + + for (i = 1; i<=nkeys; i++) + { + extract_get_fname_tmp (zh, rbuf, i); + unlink (rbuf); + } + logf (LOG_LOG, "Iterations . . .%7d", no_iterations); + logf (LOG_LOG, "Distinct words .%7d", no_diffs); + logf (LOG_LOG, "Updates. . . . .%7d", no_updates); + logf (LOG_LOG, "Deletions. . . .%7d", no_deletions); + logf (LOG_LOG, "Insertions . . .%7d", no_insertions); + zh->key_file_no = 0; +} - dict_close (dict); - is_close (isam); +void key_input (ZebraHandle zh, int nkeys, int cache, Res res) + +{ + struct key_file **kf; + char rbuf[1024]; + int i, r; + struct heap_info *hi; + struct progressInfo progressInfo; + if (nkeys < 0) + { + char fname[1024]; + nkeys = 0; + while (1) + { + getFnameTmp (res, fname, nkeys+1); + if (access (fname, R_OK) == -1) + break; + nkeys++; + } + if (!nkeys) + return ; + } + kf = (struct key_file **) xmalloc ((1+nkeys) * sizeof(*kf)); + progressInfo.totalBytes = 0; + progressInfo.totalOffset = 0; + time (&progressInfo.startTime); + time (&progressInfo.lastTime); + for (i = 1; i<=nkeys; i++) + { + kf[i] = key_file_init (i, 8192, res); + kf[i]->readHandler = progressFunc; + kf[i]->readInfo = &progressInfo; + progressInfo.totalBytes += kf[i]->length; + progressInfo.totalOffset += kf[i]->buf_size; + } + hi = key_heap_init (nkeys, key_qsort_compare); + hi->dict = zh->service->dict; + hi->isams = zh->service->isams; +#if ZMBOL + hi->isam = zh->service->isam; + hi->isamc = zh->service->isamc; + hi->isamd = zh->service->isamd; +#endif + + for (i = 1; i<=nkeys; i++) + if ((r = key_file_read (kf[i], rbuf))) + key_heap_insert (hi, rbuf, r, kf[i]); + if (hi->isams) + heap_inps (hi); +#if ZMBOL + else if (hi->isamc) + heap_inpc (hi); + else if (hi->isam) + heap_inp (hi); + else if (hi->isamd) + heap_inpd (hi); +#endif + + for (i = 1; i<=nkeys; i++) + { + getFnameTmp (res, rbuf, i); + unlink (rbuf); + } logf (LOG_LOG, "Iterations . . .%7d", no_iterations); logf (LOG_LOG, "Distinct words .%7d", no_diffs); logf (LOG_LOG, "Updates. . . . .%7d", no_updates); + logf (LOG_LOG, "Deletions. . . .%7d", no_deletions); logf (LOG_LOG, "Insertions . . .%7d", no_insertions); -} + /* xmalloc_trav("unfreed"); while hunting leaks */ +}