← Back to branch summary

~ubuntu-branches/ubuntu/saucy/db/saucy-proposed

« back to all changes in this revision

Viewing changes to src/hash/hash_stat.c

  • Committer: Bazaar Package Importer
  • Author(s): Colin Watson
  • Date: 2010-11-05 15:02:09 UTC
  • mfrom: (13.1.12 sid)
  • Revision ID: james.westby@ubuntu.com-20101105150209-ppvyn0619pu014xo
Tags: 5.1.19-1ubuntu1
* Resynchronise with Debian.  Remaining changes:
  - Pass --build/--host to configure to support cross-building, and don't
    override CC.
  - Disable the Java build when cross-building, for now.

Show diffs side-by-side

added added

removed removed

Lines of Context:
src/hash/hash_stat.c
 
1
/*-
 
2
 * See the file LICENSE for redistribution information.
 
3
 *
 
4
 * Copyright (c) 1996, 2010 Oracle and/or its affiliates.  All rights reserved.
 
5
 *
 
6
 * $Id$
 
7
 */
 
8
 
 
9
#include "db_config.h"
 
10
 
 
11
#include "db_int.h"
 
12
#include "dbinc/db_page.h"
 
13
#include "dbinc/btree.h"
 
14
#include "dbinc/hash.h"
 
15
#include "dbinc/mp.h"
 
16
 
 
17
#ifdef HAVE_STATISTICS
 
18
static int __ham_stat_callback __P((DBC *, PAGE *, void *, int *));
 
19
 
 
20
/*
 
21
 * __ham_stat --
 
22
 *      Gather/print the hash statistics
 
23
 *
 
24
 * PUBLIC: int __ham_stat __P((DBC *, void *, u_int32_t));
 
25
 */
 
26
int
 
27
__ham_stat(dbc, spp, flags)
 
28
        DBC *dbc;
 
29
        void *spp;
 
30
        u_int32_t flags;
 
31
{
 
32
        DB *dbp;
 
33
        DB_HASH_STAT *sp;
 
34
        DB_MPOOLFILE *mpf;
 
35
        ENV *env;
 
36
        HASH_CURSOR *hcp;
 
37
        PAGE *h;
 
38
        db_pgno_t pgno;
 
39
        int ret;
 
40
 
 
41
        dbp = dbc->dbp;
 
42
        env = dbp->env;
 
43
 
 
44
        mpf = dbp->mpf;
 
45
        sp = NULL;
 
46
 
 
47
        hcp = (HASH_CURSOR *)dbc->internal;
 
48
 
 
49
        if ((ret = __ham_get_meta(dbc)) != 0)
 
50
                goto err;
 
51
 
 
52
        /* Allocate and clear the structure. */
 
53
        if ((ret = __os_umalloc(env, sizeof(*sp), &sp)) != 0)
 
54
                goto err;
 
55
        memset(sp, 0, sizeof(*sp));
 
56
        /* Copy the fields that we have. */
 
57
        sp->hash_nkeys = hcp->hdr->dbmeta.key_count;
 
58
        sp->hash_ndata = hcp->hdr->dbmeta.record_count;
 
59
        /*
 
60
         * Don't take the page number from the meta-data page -- that value is
 
61
         * only maintained in the primary database, we may have been called on
 
62
         * a subdatabase.
 
63
         */
 
64
        if ((ret = __memp_get_last_pgno(dbp->mpf, &pgno)) != 0)
 
65
                goto err;
 
66
        sp->hash_pagecnt = pgno + 1;
 
67
        sp->hash_pagesize = dbp->pgsize;
 
68
        sp->hash_buckets = hcp->hdr->max_bucket + 1;
 
69
        sp->hash_magic = hcp->hdr->dbmeta.magic;
 
70
        sp->hash_version = hcp->hdr->dbmeta.version;
 
71
        sp->hash_metaflags = hcp->hdr->dbmeta.flags;
 
72
        sp->hash_ffactor = hcp->hdr->ffactor;
 
73
 
 
74
        if (flags == DB_FAST_STAT)
 
75
                goto done;
 
76
 
 
77
        /* Walk the free list, counting pages. */
 
78
        for (sp->hash_free = 0, pgno = hcp->hdr->dbmeta.free;
 
79
            pgno != PGNO_INVALID;) {
 
80
                ++sp->hash_free;
 
81
 
 
82
                if ((ret = __memp_fget(mpf,
 
83
                     &pgno, dbc->thread_info, dbc->txn, 0, &h)) != 0)
 
84
                        goto err;
 
85
 
 
86
                pgno = h->next_pgno;
 
87
                (void)__memp_fput(mpf, dbc->thread_info, h, dbc->priority);
 
88
        }
 
89
 
 
90
        /* Now traverse the rest of the table. */
 
91
        sp->hash_nkeys = 0;
 
92
        sp->hash_ndata = 0;
 
93
        if ((ret = __ham_traverse(dbc,
 
94
            DB_LOCK_READ, __ham_stat_callback, sp, 0)) != 0)
 
95
                goto err;
 
96
 
 
97
        if (!F_ISSET(dbp, DB_AM_RDONLY)) {
 
98
                /*
 
99
                 * A transaction is not required for DB->stat, so this update
 
100
                 * can't safely make a copy of the meta page.  We have to
 
101
                 * update in place.
 
102
                 */
 
103
                if ((ret = __ham_dirty_meta(dbc,
 
104
                    (dbc->txn == NULL) ? DB_MPOOL_EDIT : 0)) != 0)
 
105
                        goto err;
 
106
                hcp->hdr->dbmeta.key_count = sp->hash_nkeys;
 
107
                hcp->hdr->dbmeta.record_count = sp->hash_ndata;
 
108
        }
 
109
 
 
110
done:   if ((ret = __ham_release_meta(dbc)) != 0)
 
111
                goto err;
 
112
 
 
113
        *(DB_HASH_STAT **)spp = sp;
 
114
        return (0);
 
115
 
 
116
err:    if (sp != NULL)
 
117
                __os_ufree(env, sp);
 
118
 
 
119
        if (hcp->hdr != NULL)
 
120
                (void)__ham_release_meta(dbc);
 
121
 
 
122
        return (ret);
 
123
}
 
124
 
 
125
/*
 
126
 * __ham_stat_print --
 
127
 *      Display hash statistics.
 
128
 *
 
129
 * PUBLIC: int __ham_stat_print __P((DBC *, u_int32_t));
 
130
 */
 
131
int
 
132
__ham_stat_print(dbc, flags)
 
133
        DBC *dbc;
 
134
        u_int32_t flags;
 
135
{
 
136
        static const FN fn[] = {
 
137
                { DB_HASH_DUP,          "duplicates" },
 
138
                { DB_HASH_SUBDB,        "multiple-databases" },
 
139
                { DB_HASH_DUPSORT,      "sorted duplicates" },
 
140
                { 0,                    NULL }
 
141
        };
 
142
        DB *dbp;
 
143
        ENV *env;
 
144
        DB_HASH_STAT *sp;
 
145
        int lorder, ret;
 
146
        const char *s;
 
147
 
 
148
        dbp = dbc->dbp;
 
149
        env = dbp->env;
 
150
 
 
151
        if ((ret = __ham_stat(dbc, &sp, LF_ISSET(DB_FAST_STAT))) != 0)
 
152
                return (ret);
 
153
 
 
154
        if (LF_ISSET(DB_STAT_ALL)) {
 
155
                __db_msg(env, "%s", DB_GLOBAL(db_line));
 
156
                __db_msg(env, "Default Hash database information:");
 
157
        }
 
158
        __db_msg(env, "%lx\tHash magic number", (u_long)sp->hash_magic);
 
159
        __db_msg(env,
 
160
            "%lu\tHash version number", (u_long)sp->hash_version);
 
161
        (void)__db_get_lorder(dbp, &lorder);
 
162
        switch (lorder) {
 
163
        case 1234:
 
164
                s = "Little-endian";
 
165
                break;
 
166
        case 4321:
 
167
                s = "Big-endian";
 
168
                break;
 
169
        default:
 
170
                s = "Unrecognized byte order";
 
171
                break;
 
172
        }
 
173
        __db_msg(env, "%s\tByte order", s);
 
174
        __db_prflags(env, NULL, sp->hash_metaflags, fn, NULL, "\tFlags");
 
175
        __db_dl(env,
 
176
            "Number of pages in the database", (u_long)sp->hash_pagecnt);
 
177
        __db_dl(env,
 
178
            "Underlying database page size", (u_long)sp->hash_pagesize);
 
179
        __db_dl(env, "Specified fill factor", (u_long)sp->hash_ffactor);
 
180
        __db_dl(env,
 
181
            "Number of keys in the database", (u_long)sp->hash_nkeys);
 
182
        __db_dl(env,
 
183
            "Number of data items in the database", (u_long)sp->hash_ndata);
 
184
 
 
185
        __db_dl(env, "Number of hash buckets", (u_long)sp->hash_buckets);
 
186
        __db_dl_pct(env, "Number of bytes free on bucket pages",
 
187
            (u_long)sp->hash_bfree, DB_PCT_PG(
 
188
            sp->hash_bfree, sp->hash_buckets, sp->hash_pagesize), "ff");
 
189
 
 
190
        __db_dl(env,
 
191
            "Number of overflow pages", (u_long)sp->hash_bigpages);
 
192
        __db_dl_pct(env, "Number of bytes free in overflow pages",
 
193
            (u_long)sp->hash_big_bfree, DB_PCT_PG(
 
194
            sp->hash_big_bfree, sp->hash_bigpages, sp->hash_pagesize), "ff");
 
195
 
 
196
        __db_dl(env,
 
197
            "Number of bucket overflow pages", (u_long)sp->hash_overflows);
 
198
        __db_dl_pct(env,
 
199
            "Number of bytes free in bucket overflow pages",
 
200
            (u_long)sp->hash_ovfl_free, DB_PCT_PG(
 
201
            sp->hash_ovfl_free, sp->hash_overflows, sp->hash_pagesize), "ff");
 
202
 
 
203
        __db_dl(env, "Number of duplicate pages", (u_long)sp->hash_dup);
 
204
        __db_dl_pct(env, "Number of bytes free in duplicate pages",
 
205
            (u_long)sp->hash_dup_free, DB_PCT_PG(
 
206
            sp->hash_dup_free, sp->hash_dup, sp->hash_pagesize), "ff");
 
207
 
 
208
        __db_dl(env,
 
209
            "Number of pages on the free list", (u_long)sp->hash_free);
 
210
 
 
211
        __os_ufree(env, sp);
 
212
 
 
213
        return (0);
 
214
}
 
215
 
 
216
static int
 
217
__ham_stat_callback(dbc, pagep, cookie, putp)
 
218
        DBC *dbc;
 
219
        PAGE *pagep;
 
220
        void *cookie;
 
221
        int *putp;
 
222
{
 
223
        DB *dbp;
 
224
        DB_BTREE_STAT bstat;
 
225
        DB_HASH_STAT *sp;
 
226
        db_indx_t indx, len, off, tlen, top;
 
227
        u_int8_t *hk;
 
228
        int ret;
 
229
 
 
230
        *putp = 0;
 
231
        sp = cookie;
 
232
        dbp = dbc->dbp;
 
233
 
 
234
        switch (pagep->type) {
 
235
        case P_INVALID:
 
236
                /*
 
237
                 * Hash pages may be wholly zeroed;  this is not a bug.
 
238
                 * Obviously such pages have no data, so we can just proceed.
 
239
                 */
 
240
                break;
 
241
        case P_HASH_UNSORTED:
 
242
        case P_HASH:
 
243
                /*
 
244
                 * We count the buckets and the overflow pages
 
245
                 * separately and tally their bytes separately
 
246
                 * as well.  We need to figure out if this page
 
247
                 * is a bucket.
 
248
                 */
 
249
                if (PREV_PGNO(pagep) == PGNO_INVALID)
 
250
                        sp->hash_bfree += P_FREESPACE(dbp, pagep);
 
251
                else {
 
252
                        sp->hash_overflows++;
 
253
                        sp->hash_ovfl_free += P_FREESPACE(dbp, pagep);
 
254
                }
 
255
                top = NUM_ENT(pagep);
 
256
                /* Correct for on-page duplicates and deleted items. */
 
257
                for (indx = 0; indx < top; indx += P_INDX) {
 
258
                        switch (*H_PAIRDATA(dbp, pagep, indx)) {
 
259
                        case H_OFFDUP:
 
260
                                break;
 
261
                        case H_OFFPAGE:
 
262
                        case H_KEYDATA:
 
263
                                sp->hash_ndata++;
 
264
                                break;
 
265
                        case H_DUPLICATE:
 
266
                                tlen = LEN_HDATA(dbp, pagep, 0, indx);
 
267
                                hk = H_PAIRDATA(dbp, pagep, indx);
 
268
                                for (off = 0; off < tlen;
 
269
                                    off += len + 2 * sizeof(db_indx_t)) {
 
270
                                        sp->hash_ndata++;
 
271
                                        memcpy(&len,
 
272
                                            HKEYDATA_DATA(hk)
 
273
                                            + off, sizeof(db_indx_t));
 
274
                                }
 
275
                                break;
 
276
                        default:
 
277
                                return (__db_pgfmt(dbp->env, PGNO(pagep)));
 
278
                        }
 
279
                }
 
280
                sp->hash_nkeys += H_NUMPAIRS(pagep);
 
281
                break;
 
282
        case P_IBTREE:
 
283
        case P_IRECNO:
 
284
        case P_LBTREE:
 
285
        case P_LRECNO:
 
286
        case P_LDUP:
 
287
                /*
 
288
                 * These are all btree pages; get a correct
 
289
                 * cookie and call them.  Then add appropriate
 
290
                 * fields into our stat structure.
 
291
                 */
 
292
                memset(&bstat, 0, sizeof(bstat));
 
293
                if ((ret = __bam_stat_callback(dbc, pagep, &bstat, putp)) != 0)
 
294
                        return (ret);
 
295
                sp->hash_dup++;
 
296
                sp->hash_dup_free += bstat.bt_leaf_pgfree +
 
297
                    bstat.bt_dup_pgfree + bstat.bt_int_pgfree;
 
298
                sp->hash_ndata += bstat.bt_ndata;
 
299
                break;
 
300
        case P_OVERFLOW:
 
301
                sp->hash_bigpages++;
 
302
                sp->hash_big_bfree += P_OVFLSPACE(dbp, dbp->pgsize, pagep);
 
303
                break;
 
304
        default:
 
305
                return (__db_pgfmt(dbp->env, PGNO(pagep)));
 
306
        }
 
307
 
 
308
        return (0);
 
309
}
 
310
 
 
311
/*
 
312
 * __ham_print_cursor --
 
313
 *      Display the current cursor.
 
314
 *
 
315
 * PUBLIC: void __ham_print_cursor __P((DBC *));
 
316
 */
 
317
void
 
318
__ham_print_cursor(dbc)
 
319
        DBC *dbc;
 
320
{
 
321
        static const FN fn[] = {
 
322
                { H_CONTINUE,   "H_CONTINUE" },
 
323
                { H_DELETED,    "H_DELETED" },
 
324
                { H_DUPONLY,    "H_DUPONLY" },
 
325
                { H_EXPAND,     "H_EXPAND" },
 
326
                { H_ISDUP,      "H_ISDUP" },
 
327
                { H_NEXT_NODUP, "H_NEXT_NODUP" },
 
328
                { H_NOMORE,     "H_NOMORE" },
 
329
                { H_OK,         "H_OK" },
 
330
                { 0,            NULL }
 
331
        };
 
332
        ENV *env;
 
333
        HASH_CURSOR *cp;
 
334
 
 
335
        env = dbc->env;
 
336
        cp = (HASH_CURSOR *)dbc->internal;
 
337
 
 
338
        STAT_ULONG("Bucket traversing", cp->bucket);
 
339
        STAT_ULONG("Bucket locked", cp->lbucket);
 
340
        STAT_ULONG("Duplicate set offset", cp->dup_off);
 
341
        STAT_ULONG("Current duplicate length", cp->dup_len);
 
342
        STAT_ULONG("Total duplicate set length", cp->dup_tlen);
 
343
        STAT_ULONG("Bytes needed for add", cp->seek_size);
 
344
        STAT_ULONG("Page on which we can insert", cp->seek_found_page);
 
345
        STAT_ULONG("Order", cp->order);
 
346
        __db_prflags(env, NULL, cp->flags, fn, NULL, "\tInternal Flags");
 
347
}
 
348
 
 
349
#else /* !HAVE_STATISTICS */
 
350
 
 
351
int
 
352
__ham_stat(dbc, spp, flags)
 
353
        DBC *dbc;
 
354
        void *spp;
 
355
        u_int32_t flags;
 
356
{
 
357
        COMPQUIET(spp, NULL);
 
358
        COMPQUIET(flags, 0);
 
359
 
 
360
        return (__db_stat_not_built(dbc->env));
 
361
}
 
362
#endif
 
363
 
 
364
/*
 
365
 * __ham_traverse
 
366
 *       Traverse an entire hash table.  We use the callback so that we
 
367
 * can use this both for stat collection and for deallocation.
 
368
 *
 
369
 * PUBLIC: int __ham_traverse __P((DBC *, db_lockmode_t,
 
370
 * PUBLIC:     int (*)(DBC *, PAGE *, void *, int *), void *, int));
 
371
 */
 
372
int
 
373
__ham_traverse(dbc, mode, callback, cookie, look_past_max)
 
374
        DBC *dbc;
 
375
        db_lockmode_t mode;
 
376
        int (*callback) __P((DBC *, PAGE *, void *, int *));
 
377
        void *cookie;
 
378
        int look_past_max;
 
379
{
 
380
        DB *dbp;
 
381
        DBC *opd;
 
382
        DB_MPOOLFILE *mpf;
 
383
        HASH_CURSOR *hcp;
 
384
        HKEYDATA *hk;
 
385
        db_pgno_t pgno, opgno;
 
386
        int did_put, i, ret, t_ret;
 
387
        u_int32_t bucket, spares_entry;
 
388
 
 
389
        dbp = dbc->dbp;
 
390
        opd = NULL;
 
391
        mpf = dbp->mpf;
 
392
        hcp = (HASH_CURSOR *)dbc->internal;
 
393
        ret = 0;
 
394
 
 
395
        /*
 
396
         * In a perfect world, we could simply read each page in the file
 
397
         * and look at its page type to tally the information necessary.
 
398
         * Unfortunately, the bucket locking that hash tables do to make
 
399
         * locking easy, makes this a pain in the butt.  We have to traverse
 
400
         * duplicate, overflow and big pages from the bucket so that we
 
401
         * don't access anything that isn't properly locked.
 
402
         *
 
403
         */
 
404
        for (bucket = 0;; bucket++) {
 
405
                /*
 
406
                 * We put the loop exit condition check here, because
 
407
                 * it made for a really vile extended ?: that made SCO's
 
408
                 * compiler drop core.
 
409
                 *
 
410
                 * If look_past_max is not set, we can stop at max_bucket;
 
411
                 * if it is set, we need to include pages that are part of
 
412
                 * the current doubling but beyond the highest bucket we've
 
413
                 * split into, as well as pages from a "future" doubling
 
414
                 * that may have been created within an aborted
 
415
                 * transaction.  To do this, keep looping (and incrementing
 
416
                 * bucket) until the corresponding spares array entries
 
417
                 * cease to be defined.
 
418
                 */
 
419
                if (look_past_max) {
 
420
                        spares_entry = __db_log2(bucket + 1);
 
421
                        if (spares_entry >= NCACHED ||
 
422
                            hcp->hdr->spares[spares_entry] == 0)
 
423
                                break;
 
424
                } else {
 
425
                        if (bucket > hcp->hdr->max_bucket)
 
426
                                break;
 
427
                }
 
428
 
 
429
                hcp->bucket = bucket;
 
430
                hcp->pgno = pgno = BUCKET_TO_PAGE(hcp, bucket);
 
431
                for (ret = __ham_get_cpage(dbc, mode); ret == 0;
 
432
                    ret = __ham_next_cpage(dbc, pgno)) {
 
433
 
 
434
                        /*
 
435
                         * If we are cleaning up pages past the max_bucket,
 
436
                         * then they may be on the free list and have their
 
437
                         * next pointers set, but they should be ignored.  In
 
438
                         * fact, we really ought to just skip anybody who is
 
439
                         * not a valid page.
 
440
                         */
 
441
                        if (TYPE(hcp->page) == P_INVALID)
 
442
                                break;
 
443
                        pgno = NEXT_PGNO(hcp->page);
 
444
 
 
445
                        /*
 
446
                         * Go through each item on the page checking for
 
447
                         * duplicates (in which case we have to count the
 
448
                         * duplicate pages) or big key/data items (in which
 
449
                         * case we have to count those pages).
 
450
                         */
 
451
                        for (i = 0; i < NUM_ENT(hcp->page); i++) {
 
452
                                hk = (HKEYDATA *)P_ENTRY(dbp, hcp->page, i);
 
453
                                switch (HPAGE_PTYPE(hk)) {
 
454
                                case H_OFFDUP:
 
455
                                        memcpy(&opgno, HOFFDUP_PGNO(hk),
 
456
                                            sizeof(db_pgno_t));
 
457
                                        if ((ret = __dbc_newopd(dbc,
 
458
                                            opgno, NULL, &opd)) != 0)
 
459
                                                return (ret);
 
460
                                        if ((ret = __bam_traverse(opd,
 
461
                                            DB_LOCK_READ, opgno,
 
462
                                            callback, cookie))
 
463
                                            != 0)
 
464
                                                goto err;
 
465
                                        if ((ret = __dbc_close(opd)) != 0)
 
466
                                                return (ret);
 
467
                                        opd = NULL;
 
468
                                        break;
 
469
                                case H_OFFPAGE:
 
470
                                        /*
 
471
                                         * We are about to get a big page
 
472
                                         * which will use the same spot that
 
473
                                         * the current page uses, so we need
 
474
                                         * to restore the current page before
 
475
                                         * looking at it again.
 
476
                                         */
 
477
                                        memcpy(&opgno, HOFFPAGE_PGNO(hk),
 
478
                                            sizeof(db_pgno_t));
 
479
                                        if ((ret = __db_traverse_big(dbc,
 
480
                                            opgno, callback, cookie)) != 0)
 
481
                                                goto err;
 
482
                                        break;
 
483
                                case H_KEYDATA:
 
484
                                case H_DUPLICATE:
 
485
                                        break;
 
486
                                default:
 
487
                                        ret = __db_unknown_path(
 
488
                                            dbp->env, "__ham_traverse");
 
489
                                        goto err;
 
490
                                }
 
491
                        }
 
492
 
 
493
                        /* Call the callback on main pages. */
 
494
                        if ((ret = callback(dbc,
 
495
                            hcp->page, cookie, &did_put)) != 0)
 
496
                                goto err;
 
497
 
 
498
                        if (did_put)
 
499
                                hcp->page = NULL;
 
500
                        if (pgno == PGNO_INVALID)
 
501
                                break;
 
502
                }
 
503
                if (ret != 0)
 
504
                        goto err;
 
505
 
 
506
                if (hcp->page != NULL) {
 
507
                        if ((ret = __memp_fput(mpf,
 
508
                            dbc->thread_info, hcp->page, dbc->priority)) != 0)
 
509
                                return (ret);
 
510
                        hcp->page = NULL;
 
511
                }
 
512
 
 
513
        }
 
514
err:    if (opd != NULL &&
 
515
            (t_ret = __dbc_close(opd)) != 0 && ret == 0)
 
516
                ret = t_ret;
 
517
        return (ret);
 
518
}