-/*
- * Copyright (c) 1995-1996, Index Data.
- * See the file LICENSE for details.
- * Sebastian Hammer, Adam Dickmeiss
- *
- * $Log: isamc.c,v $
- * Revision 1.8 1997-09-17 12:19:20 adam
- * Zebra version corresponds to YAZ version 1.4.
- * Changed Zebra server so that it doesn't depend on global common_resource.
- *
- * Revision 1.7 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 1.6 1996/11/08 11:15:29 adam
- * Number of keys in chain are stored in first block and the function
- * to retrieve this information, isc_pp_num is implemented.
- *
- * Revision 1.5 1996/11/04 14:08:57 adam
- * Optimized free block usage.
- *
- * Revision 1.4 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 1.3 1996/11/01 08:59:14 adam
- * First version of isc_merge that supports update/delete.
- *
- * Revision 1.2 1996/10/29 16:44:56 adam
- * Work on isc_merge.
- *
- * Revision 1.1 1996/10/29 13:40:48 adam
- * First work.
- *
- */
+/* $Id: isamc.c,v 1.29 2005-01-15 19:38:31 adam Exp $
+ Copyright (C) 1995-2005
+ 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.
+*/
/*
* TODO:
#include <string.h>
#include <stdio.h>
-#include <log.h>
+#include <yaz/log.h>
+#include <yaz/xmalloc.h>
#include "isamc-p.h"
+static void flush_block (ISAMC is, int cat);
static void release_fc (ISAMC is, int cat);
static void init_fc (ISAMC is, int cat);
-ISAMC_M isc_getmethod (void)
+#define ISAMC_FREELIST_CHUNK 0
+
+#define SMALL_TEST 0
+
+void isc_getmethod (ISAMC_M *m)
{
+
static struct ISAMC_filecat_s def_cat[] = {
- { 32, 28, 0, 20 },
- { 512, 490, 100, 20 },
- { 4096, 3950, 1000, 20 },
- {32768, 32000, 10000, 0 },
+#if SMALL_TEST
+ { 32, 28, 0, 3 },
+ { 64, 54, 30, 0 },
+#else
+ { 64, 56, 40, 5 },
+ { 128, 120, 100, 10 },
+ { 512, 490, 350, 10 },
+ { 2048, 1900, 1700, 10 },
+ { 8192, 8000, 7900, 10 },
+ { 32768, 32000, 31000, 0 },
+#endif
};
- ISAMC_M m = xmalloc (sizeof(*m));
m->filecat = def_cat;
- m->code_start = NULL;
- m->code_item = NULL;
- m->code_stop = NULL;
+ m->codec.start = NULL;
+ m->codec.decode = NULL;
+ m->codec.encode = NULL;
+ m->codec.stop = NULL;
+ m->codec.reset = NULL;
m->compare_item = NULL;
+ m->log_item = NULL;
m->debug = 1;
m->max_blocks_mem = 10;
-
- return m;
}
-
-ISAMC isc_open (BFiles bfs, const char *name, int writeflag, ISAMC_M method)
+ISAMC isc_open (BFiles bfs, const char *name, int writeflag, ISAMC_M *method)
{
ISAMC is;
ISAMC_filecat filecat;
int i = 0;
int max_buf_size = 0;
- is = xmalloc (sizeof(*is));
+ is = (ISAMC) xmalloc (sizeof(*is));
- is->method = xmalloc (sizeof(*is->method));
+ is->method = (ISAMC_M *) xmalloc (sizeof(*is->method));
memcpy (is->method, method, sizeof(*method));
filecat = is->method->filecat;
assert (filecat);
/* determine number of block categories */
if (is->method->debug)
- logf (LOG_LOG, "isc: bsize ifill mfill mblocks");
+ yaz_log (YLOG_LOG, "isc: bsize ifill mfill mblocks");
do
{
if (is->method->debug)
- logf (LOG_LOG, "isc:%6d %6d %6d %6d",
+ yaz_log (YLOG_LOG, "isc:%6d %6d %6d %6d",
filecat[i].bsize, filecat[i].ifill,
filecat[i].mfill, filecat[i].mblocks);
if (max_buf_size < filecat[i].mblocks * filecat[i].bsize)
if (max_buf_size < (1+is->method->max_blocks_mem) * filecat[i].bsize)
max_buf_size = (1+is->method->max_blocks_mem) * filecat[i].bsize;
if (is->method->debug)
- logf (LOG_LOG, "isc: max_buf_size %d", max_buf_size);
+ yaz_log (YLOG_LOG, "isc: max_buf_size %d", max_buf_size);
assert (is->no_files > 0);
- is->files = xmalloc (sizeof(*is->files)*is->no_files);
+ is->files = (ISAMC_file) xmalloc (sizeof(*is->files)*is->no_files);
if (writeflag)
{
- is->merge_buf = xmalloc (max_buf_size+256);
+ is->merge_buf = (char *) xmalloc (max_buf_size+256);
memset (is->merge_buf, 0, max_buf_size+256);
}
else
is->files[i].head.lastblock = 1;
is->files[i].head.freelist = 0;
}
+ is->files[i].alloc_entries_num = 0;
+ is->files[i].alloc_entries_max =
+ is->method->filecat[i].bsize / sizeof(zint) - 1;
+ is->files[i].alloc_buf = (char *)
+ xmalloc (is->method->filecat[i].bsize);
is->files[i].no_writes = 0;
is->files[i].no_reads = 0;
is->files[i].no_skip_writes = 0;
is->files[i].no_allocated = 0;
is->files[i].no_released = 0;
is->files[i].no_remap = 0;
+ is->files[i].no_forward = 0;
+ is->files[i].no_backward = 0;
+ is->files[i].sum_forward = 0;
+ is->files[i].sum_backward = 0;
+ is->files[i].no_next = 0;
+ is->files[i].no_prev = 0;
init_fc (is, i);
}
return is;
}
+zint isc_block_used (ISAMC is, int type)
+{
+ if (type < 0 || type >= is->no_files)
+ return -1;
+ return is->files[type].head.lastblock-1;
+}
+
+int isc_block_size (ISAMC is, int type)
+{
+ ISAMC_filecat filecat = is->method->filecat;
+ if (type < 0 || type >= is->no_files)
+ return -1;
+ return filecat[type].bsize;
+}
+
int isc_close (ISAMC is)
{
int i;
if (is->method->debug)
- logf (LOG_LOG, "isc: writes reads skipped alloc released remap");
+ {
+ yaz_log (YLOG_LOG, "isc: next forw mid-f prev backw mid-b");
+ for (i = 0; i<is->no_files; i++)
+ yaz_log (YLOG_LOG, "isc:%8d%8d%8.1f%8d%8d%8.1f",
+ is->files[i].no_next,
+ is->files[i].no_forward,
+ is->files[i].no_forward ?
+ (double) is->files[i].sum_forward/is->files[i].no_forward
+ : 0.0,
+ is->files[i].no_prev,
+ is->files[i].no_backward,
+ is->files[i].no_backward ?
+ (double) is->files[i].sum_backward/is->files[i].no_backward
+ : 0.0);
+ }
+ if (is->method->debug)
+ yaz_log (YLOG_LOG, "isc: writes reads skipped alloc released remap");
for (i = 0; i<is->no_files; i++)
{
release_fc (is, i);
bf_write (is->files[i].bf, 0, 0, sizeof(ISAMC_head),
&is->files[i].head);
if (is->method->debug)
- logf (LOG_LOG, "isc:%8d%8d%8d%8d%8d%8d",
+ yaz_log (YLOG_LOG, "isc:%8d%8d%8d%8d%8d%8d",
is->files[i].no_writes,
is->files[i].no_reads,
is->files[i].no_skip_writes,
is->files[i].no_released,
is->files[i].no_remap);
xfree (is->files[i].fc_list);
+ flush_block (is, i);
bf_close (is->files[i].bf);
}
xfree (is->files);
xfree (is->merge_buf);
+ xfree (is->method);
xfree (is);
return 0;
}
-int isc_read_block (ISAMC is, int cat, int pos, char *dst)
+int isc_read_block (ISAMC is, int cat, zint pos, char *dst)
{
++(is->files[cat].no_reads);
return bf_read (is->files[cat].bf, pos, 0, 0, dst);
}
-int isc_write_block (ISAMC is, int cat, int pos, char *src)
+int isc_write_block (ISAMC is, int cat, zint pos, char *src)
{
++(is->files[cat].no_writes);
if (is->method->debug > 2)
- logf (LOG_LOG, "isc: write_block %d %d", cat, pos);
+ yaz_log (YLOG_LOG, "isc: write_block %d " ZINT_FORMAT, cat, pos);
return bf_write (is->files[cat].bf, pos, 0, 0, src);
}
-int isc_write_dblock (ISAMC is, int cat, int pos, char *src,
- int nextpos, int offset)
+int isc_write_dblock (ISAMC is, int cat, zint pos, char *src,
+ zint nextpos, int offset)
{
- unsigned short size = offset + ISAMC_BLOCK_OFFSET_N;
+ ISAMC_BLOCK_SIZE size = offset + ISAMC_BLOCK_OFFSET_N;
if (is->method->debug > 2)
- logf (LOG_LOG, "isc: write_dblock. size=%d nextpos=%d",
+ yaz_log (YLOG_LOG, "isc: write_dblock. size=%d nextpos=" ZINT_FORMAT,
(int) size, nextpos);
src -= ISAMC_BLOCK_OFFSET_N;
- memcpy (src, &nextpos, sizeof(int));
- memcpy (src + sizeof(int), &size, sizeof(size));
+ memcpy (src, &nextpos, sizeof(nextpos));
+ memcpy (src + sizeof(nextpos), &size, sizeof(size));
return isc_write_block (is, cat, pos, src);
}
-static int alloc_block (ISAMC is, int cat)
+#if ISAMC_FREELIST_CHUNK
+static void flush_block (ISAMC is, int cat)
+{
+ char *abuf = is->files[cat].alloc_buf;
+ zint block = is->files[cat].head.freelist;
+ if (block && is->files[cat].alloc_entries_num)
+ {
+ memcpy (abuf, &is->files[cat].alloc_entries_num, sizeof(block));
+ bf_write (is->files[cat].bf, block, 0, 0, abuf);
+ is->files[cat].alloc_entries_num = 0;
+ }
+ xfree (abuf);
+}
+
+static zint alloc_block (ISAMC is, int cat)
+{
+ zint block = is->files[cat].head.freelist;
+ char *abuf = is->files[cat].alloc_buf;
+
+ (is->files[cat].no_allocated)++;
+
+ if (!block)
+ {
+ block = (is->files[cat].head.lastblock)++; /* no free list */
+ is->files[cat].head_is_dirty = 1;
+ }
+ else
+ {
+ if (!is->files[cat].alloc_entries_num) /* read first time */
+ {
+ bf_read (is->files[cat].bf, block, 0, 0, abuf);
+ memcpy (&is->files[cat].alloc_entries_num, abuf,
+ sizeof(is->files[cat].alloc_entries_num));
+ assert (is->files[cat].alloc_entries_num > 0);
+ }
+ /* have some free blocks now */
+ assert (is->files[cat].alloc_entries_num > 0);
+ is->files[cat].alloc_entries_num--;
+ if (!is->files[cat].alloc_entries_num) /* last one in block? */
+ {
+ memcpy (&is->files[cat].head.freelist, abuf + sizeof(int),
+ sizeof(zint));
+ is->files[cat].head_is_dirty = 1;
+
+ if (is->files[cat].head.freelist)
+ {
+ bf_read (is->files[cat].bf, is->files[cat].head.freelist,
+ 0, 0, abuf);
+ memcpy (&is->files[cat].alloc_entries_num, abuf,
+ sizeof(is->files[cat].alloc_entries_num));
+ assert (is->files[cat].alloc_entries_num);
+ }
+ }
+ else
+ memcpy (&block, abuf + sizeof(zint) + sizeof(int) *
+ is->files[cat].alloc_entries_num, sizeof(zint));
+ }
+ return block;
+}
+
+static void release_block (ISAMC is, int cat, zint pos)
{
- int block;
- char buf[sizeof(int)];
+ char *abuf = is->files[cat].alloc_buf;
+ zint block = is->files[cat].head.freelist;
+
+ (is->files[cat].no_released)++;
+
+ if (block && !is->files[cat].alloc_entries_num) /* must read block */
+ {
+ bf_read (is->files[cat].bf, block, 0, 0, abuf);
+ memcpy (&is->files[cat].alloc_entries_num, abuf,
+ sizeof(is->files[cat].alloc_entries_num));
+ assert (is->files[cat].alloc_entries_num > 0);
+ }
+ assert (is->files[cat].alloc_entries_num <= is->files[cat].alloc_entries_max);
+ if (is->files[cat].alloc_entries_num == is->files[cat].alloc_entries_max)
+ {
+ assert (block);
+ memcpy (abuf, &is->files[cat].alloc_entries_num, sizeof(int));
+ bf_write (is->files[cat].bf, block, 0, 0, abuf);
+ is->files[cat].alloc_entries_num = 0;
+ }
+ if (!is->files[cat].alloc_entries_num) /* make new buffer? */
+ {
+ memcpy (abuf + sizeof(int), &block, sizeof(zint));
+ is->files[cat].head.freelist = pos;
+ is->files[cat].head_is_dirty = 1;
+ }
+ else
+ {
+ memcpy (abuf + sizeof(int) +
+ is->files[cat].alloc_entries_num*sizeof(zint),
+ &pos, sizeof(zint));
+ }
+ is->files[cat].alloc_entries_num++;
+}
+#else
+static void flush_block (ISAMC is, int cat)
+{
+ char *abuf = is->files[cat].alloc_buf;
+ xfree (abuf);
+}
+
+static zint alloc_block (ISAMC is, int cat)
+{
+ zint block;
+ char buf[sizeof(zint)];
is->files[cat].head_is_dirty = 1;
(is->files[cat].no_allocated)++;
if ((block = is->files[cat].head.freelist))
{
- bf_read (is->files[cat].bf, block, 0, sizeof(int), buf);
- memcpy (&is->files[cat].head.freelist, buf, sizeof(int));
+ bf_read (is->files[cat].bf, block, 0, sizeof(zint), buf);
+ memcpy (&is->files[cat].head.freelist, buf, sizeof(zint));
}
else
block = (is->files[cat].head.lastblock)++;
return block;
}
-int isc_alloc_block (ISAMC is, int cat)
+static void release_block (ISAMC is, int cat, zint pos)
+{
+ char buf[sizeof(zint)];
+
+ (is->files[cat].no_released)++;
+ is->files[cat].head_is_dirty = 1;
+ memcpy (buf, &is->files[cat].head.freelist, sizeof(zint));
+ is->files[cat].head.freelist = pos;
+ bf_write (is->files[cat].bf, pos, 0, sizeof(zint), buf);
+}
+#endif
+
+zint isc_alloc_block (ISAMC is, int cat)
{
- int block = 0;
+ zint block = 0;
if (is->files[cat].fc_list)
{
- int j, nb;
+ int j;
+ zint nb;
for (j = 0; j < is->files[cat].fc_max; j++)
if ((nb = is->files[cat].fc_list[j]) && (!block || nb < block))
{
is->files[cat].fc_list[j] = 0;
+ block = nb;
break;
}
}
if (!block)
block = alloc_block (is, cat);
if (is->method->debug > 3)
- logf (LOG_LOG, "isc: alloc_block in cat %d: %d", cat, block);
+ yaz_log (YLOG_LOG, "isc: alloc_block in cat %d: " ZINT_FORMAT, cat, block);
return block;
}
-static void release_block (ISAMC is, int cat, int pos)
-{
- char buf[sizeof(int)];
-
- (is->files[cat].no_released)++;
- is->files[cat].head_is_dirty = 1;
- memcpy (buf, &is->files[cat].head.freelist, sizeof(int));
- is->files[cat].head.freelist = pos;
- bf_write (is->files[cat].bf, pos, 0, sizeof(int), buf);
-}
-
-void isc_release_block (ISAMC is, int cat, int pos)
+void isc_release_block (ISAMC is, int cat, zint pos)
{
if (is->method->debug > 3)
- logf (LOG_LOG, "isc: release_block in cat %d: %d", cat, pos);
+ yaz_log (YLOG_LOG, "isc: release_block in cat %d:" ZINT_FORMAT, cat, pos);
if (is->files[cat].fc_list)
{
int j;
int j = 100;
is->files[cat].fc_max = j;
- is->files[cat].fc_list = xmalloc (sizeof(*is->files[0].fc_list) * j);
+ is->files[cat].fc_list = (zint *)
+ xmalloc (sizeof(*is->files[0].fc_list) * j);
while (--j >= 0)
is->files[cat].fc_list[j] = 0;
}
static void release_fc (ISAMC is, int cat)
{
- int b, j = is->files[cat].fc_max;
+ int j = is->files[cat].fc_max;
+ zint b;
while (--j >= 0)
if ((b = is->files[cat].fc_list[j]))
{
ISAMC is = pp->is;
- (*is->method->code_stop)(ISAMC_DECODE, pp->decodeClientData);
+ (*is->method->codec.stop)(pp->decodeClientData);
xfree (pp->buf);
xfree (pp);
}
ISAMC_PP isc_pp_open (ISAMC is, ISAMC_P ipos)
{
- ISAMC_PP pp = xmalloc (sizeof(*pp));
+ ISAMC_PP pp = (ISAMC_PP) xmalloc (sizeof(*pp));
char *src;
- pp->cat = isc_type(ipos);
+ pp->cat = (int) isc_type(ipos);
pp->pos = isc_block(ipos);
- src = pp->buf = xmalloc (is->method->filecat[pp->cat].bsize);
+ src = pp->buf = (char *) xmalloc (is->method->filecat[pp->cat].bsize);
pp->next = 0;
pp->size = 0;
pp->offset = 0;
pp->is = is;
- pp->decodeClientData = (*is->method->code_start)(ISAMC_DECODE);
+ pp->decodeClientData = (*is->method->codec.start)();
pp->deleteFlag = 0;
pp->numKeys = 0;
src += sizeof(pp->size);
memcpy (&pp->numKeys, src, sizeof(pp->numKeys));
src += sizeof(pp->numKeys);
- assert (pp->next != pp->pos);
+ if (pp->next == pp->pos)
+ {
+ yaz_log(YLOG_FATAL|YLOG_LOG, "pp->next = " ZINT_FORMAT, pp->next);
+ yaz_log(YLOG_FATAL|YLOG_LOG, "pp->pos = " ZINT_FORMAT, pp->pos);
+ assert (pp->next != pp->pos);
+ }
pp->offset = src - pp->buf;
assert (pp->offset == ISAMC_BLOCK_OFFSET_1);
if (is->method->debug > 2)
- logf (LOG_LOG, "isc: read_block size=%d %d %d",
- pp->size, pp->cat, pp->pos);
+ yaz_log (YLOG_LOG, "isc: read_block size=%d %d " ZINT_FORMAT " next="
+ ZINT_FORMAT, pp->size, pp->cat, pp->pos, pp->next);
}
return pp;
}
/* returns non-zero if item could be read; 0 otherwise */
int isc_pp_read (ISAMC_PP pp, void *buf)
{
- return isc_read_item (pp, (char **) &buf);
+ char *cp = buf;
+ return isc_read_item (pp, &cp);
}
-/* returns non-zero if item could be read; 0 otherwise */
+/* read one item from file - decode and store it in *dst.
+ Returns
+ 0 if end-of-file
+ 1 if item could be read ok and NO boundary
+ 2 if item could be read ok and boundary */
int isc_read_item (ISAMC_PP pp, char **dst)
{
ISAMC is = pp->is;
- char *src = pp->buf + pp->offset;
+ const char *src = pp->buf + pp->offset;
if (pp->offset >= pp->size)
{
+ if (!pp->next)
+ {
+ pp->pos = 0;
+ return 0; /* end of file */
+ }
+ if (pp->next > pp->pos)
+ {
+ if (pp->next == pp->pos + 1)
+ is->files[pp->cat].no_next++;
+ else
+ {
+ is->files[pp->cat].no_forward++;
+ is->files[pp->cat].sum_forward += pp->next - pp->pos;
+ }
+ }
+ else
+ {
+ if (pp->next + 1 == pp->pos)
+ is->files[pp->cat].no_prev++;
+ else
+ {
+ is->files[pp->cat].no_backward++;
+ is->files[pp->cat].sum_backward += pp->pos - pp->next;
+ }
+ }
+ /* out new block position */
pp->pos = pp->next;
- if (!pp->pos)
- return 0;
src = pp->buf;
- isc_read_block (is, pp->cat, pp->pos, src);
+ /* read block and save 'next' and 'size' entry */
+ isc_read_block (is, pp->cat, pp->pos, pp->buf);
memcpy (&pp->next, src, sizeof(pp->next));
src += sizeof(pp->next);
memcpy (&pp->size, src, sizeof(pp->size));
src += sizeof(pp->size);
/* assume block is non-empty */
assert (src - pp->buf == ISAMC_BLOCK_OFFSET_N);
- assert (pp->next != pp->pos);
+
+ if (pp->next == pp->pos)
+ {
+ yaz_log(YLOG_FATAL|YLOG_LOG, "pp->next = " ZINT_FORMAT, pp->next);
+ yaz_log(YLOG_FATAL|YLOG_LOG, "pp->pos = " ZINT_FORMAT, pp->pos);
+ assert (pp->next != pp->pos);
+ }
+
if (pp->deleteFlag)
isc_release_block (is, pp->cat, pp->pos);
- (*is->method->code_item)(ISAMC_DECODE, pp->decodeClientData, dst, &src);
+ (*is->method->codec.decode)(pp->decodeClientData, dst, &src);
pp->offset = src - pp->buf;
if (is->method->debug > 2)
- logf (LOG_LOG, "isc: read_block size=%d %d %d",
- pp->size, pp->cat, pp->pos);
+ yaz_log (YLOG_LOG, "isc: read_block size=%d %d " ZINT_FORMAT " next="
+ ZINT_FORMAT, pp->size, pp->cat, pp->pos, pp->next);
return 2;
}
- (*is->method->code_item)(ISAMC_DECODE, pp->decodeClientData, dst, &src);
+ (*is->method->codec.decode)(pp->decodeClientData, dst, &src);
pp->offset = src - pp->buf;
return 1;
}
-int isc_pp_num (ISAMC_PP pp)
+zint isc_pp_num (ISAMC_PP pp)
{
return pp->numKeys;
}