Commit 877320cb authored by Bradley C. Kuszmaul's avatar Bradley C. Kuszmaul Committed by Yoni Fogel

Start working on examples. [t:1904]

git-svn-id: file:///svn/toku/tokudb@13636 c7de825b-a66e-492c-adef-691d508d4ae1
parent 3a739edb
/* -*- mode: C; c-basic-offset: 4 -*- */
#ident "Copyright (c) 2007, 2008 Tokutek Inc. All rights reserved."
/* Insert a bunch of stuff */
#include <db.h>
#include <assert.h>
#include <errno.h>
#include <string.h>
#include <stdio.h>
#include <stdlib.h>
#include <sys/stat.h>
#include <unistd.h>
#if !defined(DB_YESOVERWRITE)
#define DB_YESOVERWRITE 0
#endif
#if !defined(DB_PRELOCKED_WRITE)
#define NO_DB_PRELOCKED
#define DB_PRELOCKED_WRITE 0
#endif
int verbose=1;
enum { SERIAL_SPACING = 1<<6 };
enum { DEFAULT_ITEMS_TO_INSERT_PER_ITERATION = 1<<20 };
enum { DEFAULT_ITEMS_PER_TRANSACTION = 1<<14 };
static void insert (long long v);
#define CKERR(r) if (r!=0) fprintf(stderr, "%s:%d error %d %s\n", __FILE__, __LINE__, r, db_strerror(r)); assert(r==0);
#define CKERR2(r,rexpect) if (r!=rexpect) fprintf(stderr, "%s:%d error %d %s\n", __FILE__, __LINE__, r, db_strerror(r)); assert(r==rexpect);
/* default test parameters */
int keysize = sizeof (long long);
int valsize = sizeof (long long);
int pagesize = 0;
long long cachesize = 128*1024*1024;
int do_1514_point_query = 0;
int dupflags = 0;
int noserial = 0; // Don't do the serial stuff
int norandom = 0; // Don't do the random stuff
int prelock = 0;
int prelockflag = 0;
int items_per_transaction = DEFAULT_ITEMS_PER_TRANSACTION;
int items_per_iteration = DEFAULT_ITEMS_TO_INSERT_PER_ITERATION;
int finish_child_first = 0; // Commit or abort child first (before doing so to the parent). No effect if child does not exist.
int singlex_child = 0; // Do a single transaction, but do all work with a child
int singlex = 0; // Do a single transaction
int singlex_create = 0; // Create the db using the single transaction (only valid if singlex)
int insert1first = 0; // insert 1 before doing the rest
int check_small_rolltmp = 0; // verify that the rollback logs are small (only valid if singlex)
int do_transactions = 0;
int if_transactions_do_logging = DB_INIT_LOG; // set this to zero if we want no logging when transactions are used
int do_abort = 0;
int n_insertions_since_txn_began=0;
int env_open_flags = DB_CREATE|DB_PRIVATE|DB_INIT_MPOOL;
u_int32_t put_flags = DB_YESOVERWRITE;
double compressibility = -1; // -1 means make it very compressible. 1 means use random bits everywhere. 2 means half the bits are random.
int do_append = 0;
int do_checkpoint_period = 0;
u_int32_t checkpoint_period = 0;
static void do_prelock(DB* db, DB_TXN* txn) {
if (prelock) {
#if !defined(NO_DB_PRELOCKED)
int r = db->pre_acquire_table_lock(db, txn);
assert(r==0);
#else
db = db; txn = txn;
#endif
}
}
#define STRINGIFY2(s) #s
#define STRINGIFY(s) STRINGIFY2(s)
const char *dbdir = "./bench." STRINGIFY(DIRSUF); /* DIRSUF is passed in as a -D argument to the compiler. */
char *dbfilename = "bench.db";
char *dbname;
DB_ENV *dbenv;
DB *db;
DB_TXN *parenttid=0;
DB_TXN *tid=0;
static void benchmark_setup (void) {
int r;
if (!do_append) {
char unlink_cmd[strlen(dbdir) + strlen("rm -rf ") + 1];
snprintf(unlink_cmd, sizeof(unlink_cmd), "rm -rf %s", dbdir);
//printf("unlink_cmd=%s\n", unlink_cmd);
system(unlink_cmd);
if (strcmp(dbdir, ".") != 0) {
r = toku_os_mkdir(dbdir,S_IRWXU|S_IRGRP|S_IXGRP|S_IROTH|S_IXOTH);
assert(r == 0);
}
}
r = db_env_create(&dbenv, 0);
assert(r == 0);
#if DB_VERSION_MAJOR == 4 && DB_VERSION_MINOR <= 4
if (dbenv->set_lk_max) {
r = dbenv->set_lk_max(dbenv, items_per_transaction*2);
assert(r==0);
}
#endif
if (dbenv->set_lk_max_locks) {
r = dbenv->set_lk_max_locks(dbenv, items_per_transaction*2);
assert(r == 0);
}
if (dbenv->set_cachesize) {
r = dbenv->set_cachesize(dbenv, cachesize / (1024*1024*1024), cachesize % (1024*1024*1024), 1);
if (r != 0)
printf("WARNING: set_cachesize %d\n", r);
}
{
r = dbenv->open(dbenv, dbdir, env_open_flags, S_IRUSR|S_IWUSR|S_IRGRP|S_IROTH);
assert(r == 0);
}
#if defined(TOKUDB)
if (do_checkpoint_period) {
r = dbenv->checkpointing_set_period(dbenv, checkpoint_period);
assert(r == 0);
u_int32_t period;
r = dbenv->checkpointing_get_period(dbenv, &period);
assert(r == 0 && period == checkpoint_period);
}
#endif
r = db_create(&db, dbenv, 0);
assert(r == 0);
if (do_transactions) {
r=dbenv->txn_begin(dbenv, 0, &tid, 0); CKERR(r);
}
if (pagesize && db->set_pagesize) {
r = db->set_pagesize(db, pagesize);
assert(r == 0);
}
if (dupflags) {
r = db->set_flags(db, dupflags);
assert(r == 0);
}
r = db->open(db, tid, dbfilename, NULL, DB_BTREE, DB_CREATE, S_IRUSR|S_IWUSR|S_IRGRP|S_IROTH);
if (r!=0) fprintf(stderr, "errno=%d, %s\n", errno, strerror(errno));
assert(r == 0);
if (insert1first) {
if (do_transactions) {
r=tid->commit(tid, 0);
assert(r==0);
tid = NULL;
r=dbenv->txn_begin(dbenv, 0, &tid, 0); CKERR(r);
}
insert(-1);
if (singlex) {
r=tid->commit(tid, 0);
assert(r==0);
tid = NULL;
r=dbenv->txn_begin(dbenv, 0, &tid, 0); CKERR(r);
}
}
else if (singlex && !singlex_create) {
r=tid->commit(tid, 0);
assert(r==0);
tid = NULL;
r=dbenv->txn_begin(dbenv, 0, &tid, 0); CKERR(r);
}
if (do_transactions) {
if (singlex)
do_prelock(db, tid);
else {
r=tid->commit(tid, 0);
assert(r==0);
tid = NULL;
}
}
if (singlex_child) {
parenttid = tid;
tid = NULL;
r=dbenv->txn_begin(dbenv, parenttid, &tid, 0); CKERR(r);
}
}
#if defined(TOKUDB)
static void test1514(void);
#endif
static void benchmark_shutdown (void) {
int r;
#if defined(TOKUDB)
if (do_1514_point_query) test1514();
#endif
if (do_transactions && singlex && !insert1first && (singlex_create || prelock)) {
#if defined(TOKUDB)
//There should be a single 'truncate' in the rolltmp instead of many 'insert' entries.
struct txn_stat *s;
r = tid->txn_stat(tid, &s);
assert(r==0);
//TODO: #1125 Always do the test after performance testing is done.
if (singlex_child) fprintf(stderr, "SKIPPED 'small rolltmp' test for child txn\n");
else
assert(s->rolltmp_raw_count < 100); // gross test, not worth investigating details
os_free(s);
//system("ls -l bench.tokudb");
#endif
}
if (do_transactions && singlex) {
if (!singlex_child || finish_child_first) {
assert(tid);
r = (do_abort ? tid->abort(tid) : tid->commit(tid, 0)); assert(r==0);
tid = NULL;
}
if (singlex_child) {
assert(parenttid);
r = (do_abort ? parenttid->abort(parenttid) : parenttid->commit(parenttid, 0)); assert(r==0);
parenttid = NULL;
}
else
assert(!parenttid);
}
assert(!tid);
assert(!parenttid);
r = db->close(db, 0);
assert(r == 0);
r = dbenv->close(dbenv, 0);
assert(r == 0);
}
static void long_long_to_array (unsigned char *a, int array_size, unsigned long long l) {
int i;
for (i=0; i<8 && i<array_size; i++)
a[i] = (l>>(56-8*i))&0xff;
}
static DBT *fill_dbt(DBT *dbt, const void *data, int size) {
memset(dbt, 0, sizeof *dbt);
dbt->size = size;
dbt->data = (void *) data;
return dbt;
}
// Fill array with 0's if compressibilty==-1, otherwise fill array with data that is likely to compress by a factor of compressibility.
static void fill_array (unsigned char *data, int size) {
memset(data, 0, size);
if (compressibility>0) {
int i;
for (i=0; i<size/compressibility; i++) {
data[i] = (unsigned char) random();
}
}
}
static void insert (long long v) {
unsigned char kc[keysize], vc[valsize];
DBT kt, vt;
fill_array(kc, sizeof kc);
long_long_to_array(kc, keysize, v); // Fill in the array first, then write the long long in.
fill_array(vc, sizeof vc);
long_long_to_array(vc, valsize, v);
int r = db->put(db, tid, fill_dbt(&kt, kc, keysize), fill_dbt(&vt, vc, valsize), put_flags);
CKERR(r);
if (do_transactions) {
if (n_insertions_since_txn_began>=items_per_transaction && !singlex) {
n_insertions_since_txn_began=0;
r = tid->commit(tid, 0); assert(r==0);
tid = NULL;
r=dbenv->txn_begin(dbenv, 0, &tid, 0); assert(r==0);
do_prelock(db, tid);
n_insertions_since_txn_began=0;
}
n_insertions_since_txn_began++;
}
}
static void serial_insert_from (long long from) {
long long i;
if (do_transactions && !singlex) {
int r = dbenv->txn_begin(dbenv, 0, &tid, 0); assert(r==0);
do_prelock(db, tid);
{
DBT k,v;
r=db->put(db, tid, fill_dbt(&k, "a", 1), fill_dbt(&v, "b", 1), put_flags);
CKERR(r);
}
}
for (i=0; i<items_per_iteration; i++) {
insert((from+i)*SERIAL_SPACING);
}
if (do_transactions && !singlex) {
int r= tid->commit(tid, 0); assert(r==0);
tid=NULL;
}
}
static long long llrandom (void) {
return (((long long)(random()))<<32) + random();
}
static void random_insert_below (long long below) {
long long i;
if (do_transactions && !singlex) {
int r = dbenv->txn_begin(dbenv, 0, &tid, 0); assert(r==0);
do_prelock(db, tid);
}
for (i=0; i<items_per_iteration; i++) {
insert(llrandom()%below);
}
if (do_transactions && !singlex) {
int r= tid->commit(tid, 0); assert(r==0);
tid=NULL;
}
}
static void biginsert (long long n_elements, struct timeval *starttime) {
long long i;
struct timeval t1,t2;
int iteration;
for (i=0, iteration=0; i<n_elements; i+=items_per_iteration, iteration++) {
if (!noserial) {
gettimeofday(&t1,0);
serial_insert_from(i);
gettimeofday(&t2,0);
if (verbose) printf("serial %9.6fs %8.0f/s ", toku_tdiff(&t2, &t1), items_per_iteration/toku_tdiff(&t2, &t1));
fflush(stdout);
}
if (!norandom) {
gettimeofday(&t1,0);
random_insert_below((i+items_per_iteration)*SERIAL_SPACING);
gettimeofday(&t2,0);
if (verbose) printf("random %9.6fs %8.0f/s ", toku_tdiff(&t2, &t1), items_per_iteration/toku_tdiff(&t2, &t1));
}
if (verbose) printf("cumulative %9.6fs %8.0f/s\n", toku_tdiff(&t2, starttime), (((float)items_per_iteration*(!noserial+!norandom))/toku_tdiff(&t2, starttime))*(iteration+1));
}
}
const long long default_n_items = 1LL<<22;
static int print_usage (const char *argv0) {
fprintf(stderr, "Usage:\n");
fprintf(stderr, " %s [-x] [--keysize KEYSIZE] [--valsize VALSIZE] [--noserial] [--norandom] [ n_iterations ]\n", argv0);
fprintf(stderr, " where\n");
fprintf(stderr, " -x do transactions (XCOUNT transactions per iteration) (default: no transactions at all)\n");
fprintf(stderr, " --keysize KEYSIZE sets the key size (default 8)\n");
fprintf(stderr, " --valsize VALSIZE sets the value size (default 8)\n");
fprintf(stderr, " --cachesize CACHESIZE set the database cache size\n");
fprintf(stderr, " --pagesize PAGESIZE sets the database page size\n");
fprintf(stderr, " --noserial causes the serial insertions to be skipped\n");
fprintf(stderr, " --norandom causes the random insertions to be skipped\n");
fprintf(stderr, " --compressibility C creates data that should compress by about a factor C. Default C is large. C is an float.\n");
fprintf(stderr, " --xcount N how many insertions per transaction (default=%d)\n", DEFAULT_ITEMS_PER_TRANSACTION);
fprintf(stderr, " --singlex (implies -x) Run the whole job as a single transaction. (Default don't run as a single transaction.)\n");
fprintf(stderr, " --singlex-child (implies -x) Run the whole job as a single transaction, do all work a child of that transaction.\n");
fprintf(stderr, " --finish-child-first Commit/abort child before doing so to parent (no effect if no child).\n");
fprintf(stderr, " --singlex-create (implies --singlex) Create the file using the single transaction (Default is to use a different transaction to create.)\n");
fprintf(stderr, " --check_small_rolltmp (Only valid in --singlex mode) Verify that very little data was saved in the rollback logs.\n");
fprintf(stderr, " --prelock Prelock the database.\n");
fprintf(stderr, " --prelockflag Prelock the database and send the DB_PRELOCKED_WRITE flag.\n");
fprintf(stderr, " --abort Abort the singlex after the transaction is over. (Requires --singlex.)\n");
fprintf(stderr, " --nolog If transactions are used, then don't write the recovery log\n");
fprintf(stderr, " --periter N how many insertions per iteration (default=%d)\n", DEFAULT_ITEMS_TO_INSERT_PER_ITERATION);
fprintf(stderr, " --DB_INIT_TXN (1|0) turn on or off the DB_INIT_TXN env_open_flag\n");
fprintf(stderr, " --DB_INIT_LOG (1|0) turn on or off the DB_INIT_LOG env_open_flag\n");
fprintf(stderr, " --DB_INIT_LOCK (1|0) turn on or off the DB_INIT_LOCK env_open_flag\n");
fprintf(stderr, " --1514 do a point query for something not there at end. See #1514. (Requires --norandom)\n");
fprintf(stderr, " --env DIR\n");
fprintf(stderr, " --append append to an existing file\n");
fprintf(stderr, " --checkpoint-period %"PRIu32" checkpoint period\n", checkpoint_period);
fprintf(stderr, " n_iterations how many iterations (default %lld)\n", default_n_items/DEFAULT_ITEMS_TO_INSERT_PER_ITERATION);
return 1;
}
#if defined(TOKUDB)
static int
nothing(DBT const* UU(key), DBT const* UU(val), void* UU(extra)) {
return 0;
}
static void
test1514(void) {
assert(norandom); //Otherwise we can't know the given element is missing.
unsigned char kc[keysize], vc[valsize];
DBT kt;
long long v = SERIAL_SPACING - 1;
fill_array(kc, sizeof kc);
long_long_to_array(kc, keysize, v); // Fill in the array first, then write the long long in.
fill_array(vc, sizeof vc);
long_long_to_array(vc, valsize, v);
int r;
DBC *c;
struct timeval t1,t2;
r = db->cursor(db, tid, &c, 0); CKERR(r);
gettimeofday(&t1,0);
r = c->c_getf_set(c, 0, fill_dbt(&kt, kc, keysize), nothing, NULL);
gettimeofday(&t2,0);
CKERR2(r, DB_NOTFOUND);
r = c->c_close(c); CKERR(r);
if (verbose) printf("(#1514) Single Point Query %9.6fs\n", toku_tdiff(&t2, &t1));
}
#endif
int main (int argc, const char *argv[]) {
struct timeval t1,t2,t3;
long long total_n_items = default_n_items;
char *endptr;
int i;
for (i=1; i<argc; i++) {
const char *arg = argv[i];
if (arg[0] != '-')
break;
if (strcmp(arg, "-q") == 0) {
verbose--; if (verbose<0) verbose=0;
} else if (strcmp(arg, "-x") == 0) {
do_transactions = 1;
} else if (strcmp(arg, "--noserial") == 0) {
noserial=1;
} else if (strcmp(arg, "--norandom") == 0) {
norandom=1;
} else if (strcmp(arg, "--compressibility") == 0) {
compressibility = atof(argv[++i]);
} else if (strcmp(arg, "--nolog") == 0) {
if_transactions_do_logging = 0;
} else if (strcmp(arg, "--singlex-create") == 0) {
do_transactions = 1;
singlex = 1;
singlex_create = 1;
} else if (strcmp(arg, "--finish-child-first") == 0) {
finish_child_first = 1;
} else if (strcmp(arg, "--singlex-child") == 0) {
do_transactions = 1;
singlex = 1;
singlex_child = 1;
} else if (strcmp(arg, "--singlex") == 0) {
do_transactions = 1;
singlex = 1;
} else if (strcmp(arg, "--insert1first") == 0) {
insert1first = 1;
} else if (strcmp(arg, "--check_small_rolltmp") == 0) {
check_small_rolltmp = 1;
} else if (strcmp(arg, "--xcount") == 0) {
if (i+1 >= argc) return print_usage(argv[0]);
items_per_transaction = strtoll(argv[++i], &endptr, 10); assert(*endptr == 0);
} else if (strcmp(arg, "--abort") == 0) {
do_abort = 1;
} else if (strcmp(arg, "--periter") == 0) {
if (i+1 >= argc) return print_usage(argv[0]);
items_per_iteration = strtoll(argv[++i], &endptr, 10); assert(*endptr == 0);
} else if (strcmp(arg, "--cachesize") == 0) {
if (i+1 >= argc) return print_usage(argv[0]);
cachesize = strtoll(argv[++i], &endptr, 10); assert(*endptr == 0);
} else if (strcmp(arg, "--keysize") == 0) {
if (i+1 >= argc) return print_usage(argv[0]);
keysize = atoi(argv[++i]);
} else if (strcmp(arg, "--valsize") == 0) {
if (i+1 >= argc) return print_usage(argv[0]);
valsize = atoi(argv[++i]);
} else if (strcmp(arg, "--pagesize") == 0) {
if (i+1 >= argc) return print_usage(argv[0]);
pagesize = atoi(argv[++i]);
} else if (strcmp(arg, "--dupsort") == 0) {
dupflags = DB_DUP + DB_DUPSORT;
continue;
} else if (strcmp(arg, "--env") == 0) {
if (i+1 >= argc) return print_usage(argv[0]);
dbdir = argv[++i];
#if defined(TOKUDB)
} else if (strcmp(arg, "--1514") == 0) {
do_1514_point_query=1;
#endif
} else if (strcmp(arg, "--prelock") == 0) {
prelock=1;
} else if (strcmp(arg, "--prelockflag") == 0) {
prelock=1;
prelockflag=1;
} else if (strcmp(arg, "--srandom") == 0) {
if (i+1 >= argc) return print_usage(argv[0]);
srandom(atoi(argv[++i]));
} else if (strcmp(arg, "--append") == 0) {
do_append = 1;
} else if (strcmp(arg, "--checkpoint-period") == 0) {
if (i+1 >= argc) return print_usage(argv[9]);
do_checkpoint_period = 1;
checkpoint_period = (u_int32_t) atoi(argv[++i]);
} else if (strcmp(arg, "--DB_INIT_TXN") == 0) {
if (i+1 >= argc) return print_usage(argv[0]);
if (atoi(argv[++i]))
env_open_flags |= DB_INIT_TXN;
else
env_open_flags &= ~DB_INIT_TXN;
} else if (strcmp(arg, "--DB_INIT_LOG") == 0) {
if (atoi(argv[++i]))
env_open_flags |= DB_INIT_LOG;
else
env_open_flags &= ~DB_INIT_LOG;
} else if (strcmp(arg, "--DB_INIT_LOCK") == 0) {
if (atoi(argv[++i]))
env_open_flags |= DB_INIT_LOCK;
else
env_open_flags &= ~DB_INIT_LOCK;
} else if (strcmp(arg, "--unique_checks") == 0) {
if (i+1 >= argc) return print_usage(argv[0]);
int unique_checks = atoi(argv[++i]);
if (unique_checks)
put_flags = DB_NOOVERWRITE;
else
put_flags = DB_YESOVERWRITE;
} else {
return print_usage(argv[0]);
}
}
if (do_transactions) {
env_open_flags |= DB_INIT_TXN | if_transactions_do_logging | DB_INIT_LOCK;
}
if (do_transactions && prelockflag) {
put_flags |= DB_PRELOCKED_WRITE;
}
if (i<argc) {
/* if it looks like a number */
char *end;
errno=0;
long n_iterations = strtol(argv[i], &end, 10);
if (errno!=0 || *end!=0 || end==argv[i]) {
print_usage(argv[0]);
return 1;
}
total_n_items = items_per_iteration * (long long)n_iterations;
}
if (verbose) {
if (!noserial) printf("serial ");
if (!noserial && !norandom) printf("and ");
if (!norandom) printf("random ");
printf("insertions of %d per batch%s\n", items_per_iteration, do_transactions ? " (with transactions)" : "");
}
#if !defined TOKUDB
if (check_small_rolltmp) {
fprintf(stderr, "--check_small_rolltmp only works on the TokuDB (not BDB)\n");
return print_usage(argv[0]);
}
#endif
if (check_small_rolltmp && !singlex) {
fprintf(stderr, "--check_small_rolltmp requires --singlex\n");
return print_usage(argv[0]);
}
benchmark_setup();
gettimeofday(&t1,0);
biginsert(total_n_items, &t1);
gettimeofday(&t2,0);
benchmark_shutdown();
gettimeofday(&t3,0);
if (verbose) {
printf("Shutdown %9.6fs\n", toku_tdiff(&t3, &t2));
printf("Total time %9.6fs for %lld insertions = %8.0f/s\n", toku_tdiff(&t3, &t1),
(!noserial+!norandom)*total_n_items, (!noserial+!norandom)*total_n_items/toku_tdiff(&t3, &t1));
}
#if 0 && defined TOKUDB
if (verbose) {
extern int toku_os_get_max_rss(int64_t*);
int64_t mrss;
int r = toku_os_get_max_rss(&mrss);
assert(r==0);
printf("maxrss=%.2fMB\n", mrss/256.0);
}
if (0) {
extern void print_hash_histogram (void) __attribute__((__visibility__("default")));
print_hash_histogram();
}
#endif
return 0;
}
#ifndef _DB_H
#define _DB_H
/* This code generated by make_db_h. Copyright (c) 2007, 2008 Tokutek */
#ident "Copyright (c) 2007, 2008 Tokutek Inc. All rights reserved."
#include <sys/types.h>
/*stdio is needed for the FILE* in db->verify*/
#include <stdio.h>
#if defined(__cplusplus)
extern "C" {
#endif
#define TOKUDB 1
#define TOKUDB_NATIVE_H 1
#define DB_VERSION_MAJOR 4
#define DB_VERSION_MINOR 6
#define DB_VERSION_PATCH 19
#ifndef _TOKUDB_WRAP_H
#define DB_VERSION_STRING "Tokutek: TokuDB 4.6.19"
#else
#define DB_VERSION_STRING_ydb "Tokutek: TokuDB (wrapped bdb)"
#endif
#ifndef TOKU_OFF_T_DEFINED
#define TOKU_OFF_T_DEFINED
typedef int64_t toku_off_t;
#endif
typedef struct __toku_db_env DB_ENV;
typedef struct __toku_db_key_range DB_KEY_RANGE;
typedef struct __toku_db_lsn DB_LSN;
typedef struct __toku_db DB;
typedef struct __toku_db_txn DB_TXN;
typedef struct __toku_db_txn_active DB_TXN_ACTIVE;
typedef struct __toku_db_txn_stat DB_TXN_STAT;
typedef struct __toku_dbc DBC;
typedef struct __toku_dbt DBT;
typedef u_int32_t db_recno_t;
typedef int(*YDB_CALLBACK_FUNCTION)(DBT const*, DBT const*, void*);
typedef int(*YDB_HEAVISIDE_CALLBACK_FUNCTION)(DBT const *key, DBT const *value, void *extra_f, int r_h);
typedef int(*YDB_HEAVISIDE_FUNCTION)(const DBT *key, const DBT *value, void *extra_h);
#include <tdb-internal.h>
typedef struct __toku_db_btree_stat64 {
u_int64_t bt_nkeys; /* how many unique keys (guaranteed only to be an estimate, even when flattened) */
u_int64_t bt_ndata; /* how many key-value pairs (an estimate, but exact when flattened) */
u_int64_t bt_dsize; /* how big are the keys+values (not counting the lengths) (an estimate, unless flattened) */
u_int64_t bt_fsize; /* how big is the underlying file */
} DB_BTREE_STAT64;
typedef enum {
DB_BTREE=1,
DB_UNKNOWN=5
} DBTYPE;
#ifndef _TOKUDB_WRAP_H
#define DB_VERB_DEADLOCK 1
#define DB_VERB_RECOVERY 8
#define DB_VERB_REPLICATION 32
#define DB_VERB_WAITSFOR 64
#define DB_DBT_MALLOC 8
#define DB_DBT_REALLOC 64
#define DB_DBT_USERMEM 256
#define DB_DBT_DUPOK 2
#define DB_ARCH_ABS 1
#define DB_ARCH_LOG 4
#define DB_CREATE 1
#define DB_CXX_NO_EXCEPTIONS 1
#define DB_EXCL 16384
#define DB_PRIVATE 8388608
#define DB_RDONLY 32
#define DB_RECOVER 64
#define DB_THREAD 128
#define DB_TXN_NOSYNC 512
#define DB_LOCK_DEFAULT 1
#define DB_LOCK_OLDEST 7
#define DB_LOCK_RANDOM 8
#define DB_DUP 32768
#define DB_DUPSORT 65536
#define DB_KEYFIRST 13
#define DB_KEYLAST 14
#define DB_NODUPDATA 19
#define DB_NOOVERWRITE 20
#define DB_YESOVERWRITE 254
#define DB_OPFLAGS_MASK 255
#define DB_AUTO_COMMIT 33554432
#define DB_INIT_LOCK 131072
#define DB_INIT_LOG 262144
#define DB_INIT_MPOOL 524288
#define DB_INIT_TXN 2097152
#define DB_USE_ENVIRON 16384
#define DB_USE_ENVIRON_ROOT 32768
#define DB_READ_UNCOMMITTED 134217728
#define DB_KEYEXIST -30996
#define DB_LOCK_DEADLOCK -30995
#define DB_LOCK_NOTGRANTED -30994
#define DB_NOTFOUND -30989
#define DB_SECONDARY_BAD -30974
#define DB_DONOTINDEX -30998
#define DB_BUFFER_SMALL -30999
#define DB_BADFORMAT -30500
#define DB_DELETE_ANY 65536
#define DB_TRUNCATE_WITHCURSORS 131072
#define DB_FIRST 7
#define DB_GET_BOTH 8
#define DB_GET_BOTH_RANGE 10
#define DB_LAST 15
#define DB_CURRENT 6
#define DB_NEXT 16
#define DB_NEXT_DUP 17
#define DB_NEXT_NODUP 18
#define DB_PREV 23
#define DB_PREV_DUP 24
#define DB_PREV_NODUP 25
#define DB_SET 26
#define DB_SET_RANGE 27
#define DB_CURRENT_BINDING 253
#define DB_RMW 1073741824
#define DB_PRELOCKED 0x00800000
#define DB_PRELOCKED_WRITE 0x00400000
#define DB_DBT_APPMALLOC 1
#define DB_DBT_MULTIPLE 16
#define DB_LOG_AUTOREMOVE 524288
#define DB_TXN_WRITE_NOSYNC 4096
#define DB_TXN_NOWAIT 1024
#define DB_TXN_SYNC 16384
#endif
/* TOKUDB specific error codes */
#define TOKUDB_OUT_OF_LOCKS -100000
#define TOKUDB_SUCCEEDED_EARLY -100001
#define TOKUDB_DICTIONARY_TOO_OLD -100004
#define TOKUDB_DICTIONARY_TOO_NEW -100005
#define TOKUDB_DICTIONARY_NO_HEADER -100006
#define TOKUDB_FOUND_BUT_REJECTED -100002
#define TOKUDB_USER_CALLBACK_ERROR -100003
/* in wrap mode, top-level function txn_begin is renamed, but the field isn't renamed, so we have to hack it here.*/
#ifdef _TOKUDB_WRAP_H
#undef txn_begin
#endif
struct __toku_db_env {
struct __toku_db_env_internal *i;
#define db_env_struct_i(x) ((x)->i)
int (*checkpointing_set_period) (DB_ENV*, u_int32_t) /* Change the delay between automatic checkpoints. 0 means disabled. */;
int (*checkpointing_get_period) (DB_ENV*, u_int32_t*) /* Retrieve the delay between automatic checkpoints. 0 means disabled. */;
int (*checkpointing_postpone) (DB_ENV*) /* Use for 'rename table' or any other operation that must be disjoint from a checkpoint */;
int (*checkpointing_resume) (DB_ENV*) /* Alert tokudb 'postpone' is no longer necessary */;
int (*checkpointing_begin_atomic_operation) (DB_ENV*) /* Begin a set of operations (that must be atomic as far as checkpoints are concerned). i.e. inserting into every index in one table */;
int (*checkpointing_end_atomic_operation) (DB_ENV*) /* End a set of operations (that must be atomic as far as checkpoints are concerned). */;
int (*set_default_bt_compare) (DB_ENV*,int (*bt_compare) (DB *, const DBT *, const DBT *)) /* Set default (key) comparison function for all DBs in this environment. Required for RECOVERY since you cannot open the DBs manually. */;
int (*set_default_dup_compare) (DB_ENV*,int (*bt_compare) (DB *, const DBT *, const DBT *)) /* Set default (val) comparison function for all DBs in this environment. Required for RECOVERY since you cannot open the DBs manually. */;
void *app_private;
void *api1_internal;
int (*close) (DB_ENV *, u_int32_t);
void (*err) (const DB_ENV *, int, const char *, ...);
int (*get_cachesize) (DB_ENV *, u_int32_t *, u_int32_t *, int *);
int (*get_flags) (DB_ENV *, u_int32_t *);
int (*get_lg_max) (DB_ENV *, u_int32_t*);
int (*get_lk_max_locks) (DB_ENV *, u_int32_t *);
int (*log_archive) (DB_ENV *, char **[], u_int32_t);
int (*log_flush) (DB_ENV *, const DB_LSN *);
int (*open) (DB_ENV *, const char *, u_int32_t, int);
int (*set_cachesize) (DB_ENV *, u_int32_t, u_int32_t, int);
int (*set_data_dir) (DB_ENV *, const char *);
void (*set_errcall) (DB_ENV *, void (*)(const DB_ENV *, const char *, const char *));
void (*set_errfile) (DB_ENV *, FILE*);
void (*set_errpfx) (DB_ENV *, const char *);
int (*set_flags) (DB_ENV *, u_int32_t, int);
int (*set_lg_bsize) (DB_ENV *, u_int32_t);
int (*set_lg_dir) (DB_ENV *, const char *);
int (*set_lg_max) (DB_ENV *, u_int32_t);
int (*set_lk_detect) (DB_ENV *, u_int32_t);
int (*set_lk_max_locks) (DB_ENV *, u_int32_t);
int (*set_tmp_dir) (DB_ENV *, const char *);
int (*set_verbose) (DB_ENV *, u_int32_t, int);
int (*txn_begin) (DB_ENV *, DB_TXN *, DB_TXN **, u_int32_t);
int (*txn_checkpoint) (DB_ENV *, u_int32_t, u_int32_t, u_int32_t);
int (*txn_stat) (DB_ENV *, DB_TXN_STAT **, u_int32_t);
};
struct __toku_db_key_range {
double less;
double equal;
double greater;
};
struct __toku_db_lsn {
};
struct __toku_dbt {
void*data;
u_int32_t size;
u_int32_t ulen;
u_int32_t flags;
};
typedef int (*toku_dbt_upgradef)(DB*,
u_int32_t old_version, const DBT *old_descriptor, const DBT *old_key, const DBT *old_val,
u_int32_t new_version, const DBT *new_descriptor, const DBT *new_key, const DBT *new_val);
struct __toku_db {
struct __toku_db_internal *i;
#define db_struct_i(x) ((x)->i)
int (*key_range64)(DB*, DB_TXN *, DBT *, u_int64_t *less, u_int64_t *equal, u_int64_t *greater, int *is_exact);
int (*stat64)(DB *, DB_TXN *, DB_BTREE_STAT64 *);
void *app_private;
DB_ENV *dbenv;
int (*pre_acquire_read_lock)(DB*, DB_TXN*, const DBT*, const DBT*, const DBT*, const DBT*);
int (*pre_acquire_table_lock)(DB*, DB_TXN*);
const DBT* (*dbt_pos_infty)(void) /* Return the special DBT that refers to positive infinity in the lock table.*/;
const DBT* (*dbt_neg_infty)(void)/* Return the special DBT that refers to negative infinity in the lock table.*/;
int (*delboth) (DB*, DB_TXN*, DBT*, DBT*, u_int32_t) /* Delete the key/value pair. */;
int (*row_size_supported) (DB*, u_int32_t) /* Test whether a row size is supported. */;
const DBT *descriptor /* saved row/dictionary descriptor for aiding in comparisons */;
int (*set_descriptor) (DB*, u_int32_t version, const DBT* descriptor, toku_dbt_upgradef dbt_userformat_upgrade) /* set row/dictionary descriptor for a db. Available only while db is open */;
void *api_internal;
int (*close) (DB*, u_int32_t);
int (*cursor) (DB *, DB_TXN *, DBC **, u_int32_t);
int (*del) (DB *, DB_TXN *, DBT *, u_int32_t);
int (*fd) (DB *, int *);
int (*get) (DB *, DB_TXN *, DBT *, DBT *, u_int32_t);
int (*get_flags) (DB *, u_int32_t *);
int (*get_pagesize) (DB *, u_int32_t *);
int (*key_range) (DB *, DB_TXN *, DBT *, DB_KEY_RANGE *, u_int32_t);
int (*open) (DB *, DB_TXN *, const char *, const char *, DBTYPE, u_int32_t, int);
int (*put) (DB *, DB_TXN *, DBT *, DBT *, u_int32_t);
int (*remove) (DB *, const char *, const char *, u_int32_t);
int (*rename) (DB *, const char *, const char *, const char *, u_int32_t);
int (*set_bt_compare) (DB *, int (*)(DB *, const DBT *, const DBT *));
int (*set_dup_compare) (DB *, int (*)(DB *, const DBT *, const DBT *));
void (*set_errfile) (DB *, FILE*);
int (*set_flags) (DB *, u_int32_t);
int (*set_pagesize) (DB *, u_int32_t);
int (*stat) (DB *, void *, u_int32_t);
int (*truncate) (DB *, DB_TXN *, u_int32_t *, u_int32_t);
int (*verify) (DB *, const char *, const char *, FILE *, u_int32_t);
};
struct __toku_db_txn_active {
u_int32_t txnid;
DB_LSN lsn;
};
struct txn_stat {
u_int64_t rolltmp_raw_count;
};
struct __toku_db_txn {
DB_ENV *mgrp /*In TokuDB, mgrp is a DB_ENV not a DB_TXNMGR*/;
DB_TXN *parent;
struct __toku_db_txn_internal ii;
#define db_txn_struct_i(x) (&(x)->ii)
int (*txn_stat)(DB_TXN *, struct txn_stat **);
void *api_internal;
int (*abort) (DB_TXN *);
int (*commit) (DB_TXN*, u_int32_t);
u_int32_t (*id) (DB_TXN *);
};
struct __toku_db_txn_stat {
u_int32_t st_nactive;
DB_TXN_ACTIVE *st_txnarray;
};
struct __toku_dbc {
DB *dbp;
struct __toku_dbc_internal ii;
#define dbc_struct_i(x) (&(x)->ii)
int (*c_getf_first)(DBC *, u_int32_t, YDB_CALLBACK_FUNCTION, void *);
int (*c_getf_last)(DBC *, u_int32_t, YDB_CALLBACK_FUNCTION, void *);
int (*c_getf_next)(DBC *, u_int32_t, YDB_CALLBACK_FUNCTION, void *);
int (*c_getf_next_dup)(DBC *, u_int32_t, YDB_CALLBACK_FUNCTION, void *);
int (*c_getf_next_nodup)(DBC *, u_int32_t, YDB_CALLBACK_FUNCTION, void *);
int (*c_getf_prev)(DBC *, u_int32_t, YDB_CALLBACK_FUNCTION, void *);
int (*c_getf_prev_dup)(DBC *, u_int32_t, YDB_CALLBACK_FUNCTION, void *);
int (*c_getf_prev_nodup)(DBC *, u_int32_t, YDB_CALLBACK_FUNCTION, void *);
int (*c_getf_current)(DBC *, u_int32_t, YDB_CALLBACK_FUNCTION, void *);
int (*c_getf_current_binding)(DBC *, u_int32_t, YDB_CALLBACK_FUNCTION, void *);
int (*c_getf_heaviside)(DBC *, u_int32_t, YDB_HEAVISIDE_CALLBACK_FUNCTION f, void *extra_f, YDB_HEAVISIDE_FUNCTION h, void *extra_h, int direction);
int (*c_getf_set)(DBC *, u_int32_t, DBT *, YDB_CALLBACK_FUNCTION, void *);
int (*c_getf_set_range)(DBC *, u_int32_t, DBT *, YDB_CALLBACK_FUNCTION, void *);
int (*c_getf_get_both)(DBC *, u_int32_t, DBT *, DBT *, YDB_CALLBACK_FUNCTION, void *);
int (*c_getf_get_both_range)(DBC *, u_int32_t, DBT *, DBT *, YDB_CALLBACK_FUNCTION, void *);
int (*c_close) (DBC *);
int (*c_count) (DBC *, db_recno_t *, u_int32_t);
int (*c_del) (DBC *, u_int32_t);
int (*c_get) (DBC *, DBT *, DBT *, u_int32_t);
};
#ifdef _TOKUDB_WRAP_H
#define txn_begin txn_begin_tokudb
#endif
int db_env_create(DB_ENV **, u_int32_t) __attribute__((__visibility__("default")));
int db_create(DB **, DB_ENV *, u_int32_t) __attribute__((__visibility__("default")));
char *db_strerror(int) __attribute__((__visibility__("default")));
const char *db_version(int*,int *,int *) __attribute__((__visibility__("default")));
int log_compare (const DB_LSN*, const DB_LSN *) __attribute__((__visibility__("default")));
int db_env_set_func_fsync (int (*)(int)) __attribute__((__visibility__("default")));
int toku_set_trace_file (char *fname) __attribute__((__visibility__("default")));
int toku_close_trace_file (void) __attribute__((__visibility__("default")));
int db_env_set_func_free (void (*)(void*)) __attribute__((__visibility__("default")));
int db_env_set_func_malloc (void *(*)(size_t)) __attribute__((__visibility__("default")));
int db_env_set_func_pwrite (ssize_t (*)(int, const void *, size_t, toku_off_t)) __attribute__((__visibility__("default")));
int db_env_set_func_write (ssize_t (*)(int, const void *, size_t)) __attribute__((__visibility__("default")));
int db_env_set_func_realloc (void *(*)(void*, size_t)) __attribute__((__visibility__("default")));
void db_env_set_checkpoint_callback (void (*)(void*), void*) __attribute__((__visibility__("default")));
#if defined(__cplusplus)
}
#endif
#endif
Markdown is supported
0%
or
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment