Commit bfe13877 authored by Bradley C. Kuszmaul's avatar Bradley C. Kuszmaul

Start making a simpler version of the test. Addreses #162.

git-svn-id: file:///svn/tokudb@1029 c7de825b-a66e-492c-adef-691d508d4ae1
parent 88721b6a
/* Primary with two associated things. */
#include <arpa/inet.h>
#include <assert.h>
#include <db.h>
#include <errno.h>
#include <string.h>
#include <sys/stat.h>
#include <sys/time.h>
#include <unistd.h>
#include "test.h"
enum mode {
MODE_DEFAULT, MODE_DB_CREATE, MODE_MORE
} mode;
/* Primary is a map from a UID which consists of a random number followed by the current time. */
struct timestamp {
unsigned int tv_sec; /* in newtork order */
unsigned int tv_usec; /* in network order */
};
struct primary_key {
int rand; /* in network order */
struct timestamp ts;
};
struct name_key {
unsigned char* name;
};
struct primary_data {
struct timestamp creationtime;
struct timestamp expiretime; /* not valid if doesexpire==0 */
unsigned char doesexpire;
struct name_key name;
};
void free_pd (struct primary_data *pd) {
free(pd->name.name);
free(pd);
}
void write_uchar_to_dbt (DBT *dbt, const unsigned char c) {
assert(dbt->size+1 <= dbt->ulen);
((char*)dbt->data)[dbt->size++]=c;
}
void write_uint_to_dbt (DBT *dbt, const unsigned int v) {
write_uchar_to_dbt(dbt, (v>>24)&0xff);
write_uchar_to_dbt(dbt, (v>>16)&0xff);
write_uchar_to_dbt(dbt, (v>> 8)&0xff);
write_uchar_to_dbt(dbt, (v>> 0)&0xff);
}
void write_timestamp_to_dbt (DBT *dbt, const struct timestamp *ts) {
write_uint_to_dbt(dbt, ts->tv_sec);
write_uint_to_dbt(dbt, ts->tv_usec);
}
void write_pk_to_dbt (DBT *dbt, const struct primary_key *pk) {
write_uint_to_dbt(dbt, pk->rand);
write_timestamp_to_dbt(dbt, &pk->ts);
}
void write_name_to_dbt (DBT *dbt, const struct name_key *nk) {
int i;
for (i=0; 1; i++) {
write_uchar_to_dbt(dbt, nk->name[i]);
if (nk->name[i]==0) break;
}
}
void write_pd_to_dbt (DBT *dbt, const struct primary_data *pd) {
write_timestamp_to_dbt(dbt, &pd->creationtime);
write_timestamp_to_dbt(dbt, &pd->expiretime);
write_uchar_to_dbt(dbt, pd->doesexpire);
write_name_to_dbt(dbt, &pd->name);
}
void read_uchar_from_dbt (const DBT *dbt, int *off, unsigned char *uchar) {
assert(*off < dbt->size);
*uchar = ((unsigned char *)dbt->data)[(*off)++];
}
void read_uint_from_dbt (const DBT *dbt, int *off, unsigned int *uint) {
unsigned char a,b,c,d;
read_uchar_from_dbt(dbt, off, &a);
read_uchar_from_dbt(dbt, off, &b);
read_uchar_from_dbt(dbt, off, &c);
read_uchar_from_dbt(dbt, off, &d);
*uint = (a<<24)+(b<<16)+(c<<8)+d;
}
void read_timestamp_from_dbt (const DBT *dbt, int *off, struct timestamp *ts) {
read_uint_from_dbt(dbt, off, &ts->tv_sec);
read_uint_from_dbt(dbt, off, &ts->tv_usec);
}
void read_name_from_dbt (const DBT *dbt, int *off, struct name_key *nk) {
unsigned char buf[1000];
int i;
for (i=0; 1; i++) {
read_uchar_from_dbt(dbt, off, &buf[i]);
if (buf[i]==0) break;
}
nk->name=(unsigned char*)(strdup((char*)buf));
}
void read_pd_from_dbt (const DBT *dbt, int *off, struct primary_data *pd) {
read_timestamp_from_dbt(dbt, off, &pd->creationtime);
read_timestamp_from_dbt(dbt, off, &pd->expiretime);
read_uchar_from_dbt(dbt, off, &pd->doesexpire);
read_name_from_dbt(dbt, off, &pd->name);
}
int name_offset_in_pd_dbt (void) {
return 17;
}
int name_callback (DB *secondary __attribute__((__unused__)), const DBT *key, const DBT *data, DBT *result) {
struct primary_data *pd = malloc(sizeof(*pd));
int off=0;
read_pd_from_dbt(data, &off, pd);
static int buf[1000];
result->ulen=1000;
result->data=buf;
result->size=0;
write_name_to_dbt(result, &pd->name);
free_pd(pd);
return 0;
}
int expire_callback (DB *secondary __attribute__((__unused__)), const DBT *key, const DBT *data, DBT *result) {
struct primary_data *d = data->data;
if (d->doesexpire) {
result->flags=0;
result->size=sizeof(struct timestamp);
result->data=&d->expiretime;
return 0;
} else {
return DB_DONOTINDEX;
}
}
// The expire_key is simply a timestamp.
DB_ENV *dbenv;
DB *dbp,*namedb,*expiredb;
DB_TXN * const null_txn=0;
DBC *delete_cursor=0, *name_cursor=0;
// We use a cursor to count the names.
int cursor_count_n_items=0; // The number of items the cursor saw as it scanned over.
int calc_n_items=0; // The number of items we expect the cursor to acount
int count_all_items=0; // The total number of items
DBT nc_key,nc_data;
void create_databases (void) {
int r;
r = db_env_create(&dbenv, 0); CKERR(r);
r = dbenv->open(dbenv, DIR, DB_PRIVATE|DB_INIT_MPOOL|DB_CREATE, 0); CKERR(r);
r = db_create(&dbp, dbenv, 0); CKERR(r);
r = dbp->open(dbp, null_txn, "primary.db", NULL, DB_BTREE, DB_CREATE, 0600); CKERR(r);
r = db_create(&namedb, dbenv, 0); CKERR(r);
r = namedb->open(namedb, null_txn, "name.db", NULL, DB_BTREE, DB_CREATE, 0600); CKERR(r);
r = db_create(&expiredb, dbenv, 0); CKERR(r);
r = expiredb->open(expiredb, null_txn, "expire.db", NULL, DB_BTREE, DB_CREATE, 0600); CKERR(r);
r = dbp->associate(dbp, NULL, namedb, name_callback, 0); CKERR(r);
r = dbp->associate(dbp, NULL, expiredb, expire_callback, 0); CKERR(r);
}
void close_databases (void) {
int r;
if (delete_cursor) {
r = delete_cursor->c_close(delete_cursor); CKERR(r);
}
if (name_cursor) {
r = name_cursor->c_close(name_cursor); CKERR(r);
}
if (nc_key.data) free(nc_key.data);
if (nc_data.data) free(nc_data.data);
r = namedb->close(namedb, 0); CKERR(r);
r = dbp->close(dbp, 0); CKERR(r);
r = expiredb->close(expiredb, 0); CKERR(r);
r = dbenv->close(dbenv, 0); CKERR(r);
}
void gettod (struct timestamp *ts) {
struct timeval tv;
int r = gettimeofday(&tv, 0);
assert(r==0);
ts->tv_sec = htonl(tv.tv_sec);
ts->tv_usec = htonl(tv.tv_usec);
}
void setup_for_db_create (void) {
// Remove name.db and then rebuild it with associate(... DB_CREATE)
int r=unlink(DIR "/name.db");
assert(r==0);
r = db_env_create(&dbenv, 0); CKERR(r);
r = dbenv->open(dbenv, DIR, DB_PRIVATE|DB_INIT_MPOOL, 0); CKERR(r);
r = db_create(&dbp, dbenv, 0); CKERR(r);
r = dbp->open(dbp, null_txn, "primary.db", NULL, DB_BTREE, 0, 0600); CKERR(r);
r = db_create(&namedb, dbenv, 0); CKERR(r);
r = namedb->open(namedb, null_txn, "name.db", NULL, DB_BTREE, DB_CREATE, 0600); CKERR(r);
r = db_create(&expiredb, dbenv, 0); CKERR(r);
r = expiredb->open(expiredb, null_txn, "expire.db", NULL, DB_BTREE, 0, 0600); CKERR(r);
r = dbp->associate(dbp, NULL, expiredb, expire_callback, 0); CKERR(r);
r = dbp->associate(dbp, NULL, namedb, name_callback, DB_CREATE); CKERR(r);
}
int count_entries (DB *db) {
DBC *dbc;
int r = db->cursor(db, null_txn, &dbc, 0); CKERR(r);
DBT key,data;
memset(&key, 0, sizeof(key));
memset(&data, 0, sizeof(data));
int n_found=0;
for (r = dbc->c_get(dbc, &key, &data, DB_FIRST);
r==0;
r = dbc->c_get(dbc, &key, &data, DB_NEXT)) {
n_found++;
}
assert(r==DB_NOTFOUND);
r=dbc->c_close(dbc); CKERR(r);
return n_found;
}
void do_create (void) {
setup_for_db_create();
// Now check to see if the number of names matches the number of associated things.
int n_named = count_entries(namedb);
int n_prim = count_entries(dbp);
assert(n_named==n_prim);
}
void insert_person (void) {
int namelen = 5+random()%245;
struct primary_key pk;
struct primary_data pd;
char keyarray[1000], dataarray[1000];
unsigned char namearray[1000];
pk.rand = random();
gettod(&pk.ts);
pd.creationtime = pk.ts;
pd.expiretime = pk.ts;
pd.expiretime.tv_sec += 24*60*60*366;
pd.doesexpire = (random()%10==0);
int i;
pd.name.name = namearray;
pd.name.name[0] = 'A'+random()%26;
for (i=1; i<namelen; i++) {
pd.name.name[i] = 'a'+random()%26;
}
pd.name.name[i]=0;
DBT key,data;
memset(&key,0,sizeof(DBT));
memset(&data,0,sizeof(DBT));
key.data = keyarray;
key.ulen = 1000;
key.size = 0;
data.data = dataarray;
data.ulen = 1000;
data.size = 0;
write_pk_to_dbt(&key, &pk);
write_pd_to_dbt(&data, &pd);
int r=dbp->put(dbp, null_txn, &key, &data,0); CKERR(r);
// If the cursor is to the left of the current item, then increment count_items
{
int compare=strcmp((char*)namearray, nc_key.data);
//printf("%s:%d compare=%d insert %s, cursor at %s\n", __FILE__, __LINE__, compare, namearray, (char*)nc_key.data);
if (compare>0) calc_n_items++;
count_all_items++;
}
}
void delete_oldest_expired (void) {
int r;
if (delete_cursor==0) {
r = expiredb->cursor(expiredb, null_txn, &delete_cursor, 0); CKERR(r);
}
DBT key,pkey,data, savepkey;
memset(&key, 0, sizeof(key));
memset(&pkey, 0, sizeof(pkey));
memset(&data, 0, sizeof(data));
r = delete_cursor->c_pget(delete_cursor, &key, &pkey, &data, DB_FIRST);
if (r==DB_NOTFOUND) return;
CKERR(r);
{
char *deleted_key = ((char*)data.data)+name_offset_in_pd_dbt();
int compare=strcmp(deleted_key, nc_key.data);
if (compare>0) {
//printf("%s:%d r3=%d compare=%d count=%d cacount=%d cucount=%d deleting %s cursor=%s\n", __FILE__, __LINE__, r3, compare, count_all_items, calc_n_items, cursor_count_n_items, deleted_key, (char*)nc_key.data);
calc_n_items--;
}
count_all_items--;
}
savepkey = pkey;
savepkey.data = malloc(pkey.size);
memcpy(savepkey.data, pkey.data, pkey.size);
r = dbp->del(dbp, null_txn, &pkey, 0); CKERR(r);
// Make sure it's really gone.
r = delete_cursor->c_get(delete_cursor, &key, &data, DB_CURRENT);
assert(r==DB_KEYEMPTY);
r = dbp->get(dbp, null_txn, &savepkey, &data, 0);
assert(r==DB_NOTFOUND);
free(savepkey.data);
}
// Use a cursor to step through the names.
void step_name (void) {
int r;
if (name_cursor==0) {
r = namedb->cursor(namedb, null_txn, &name_cursor, 0); CKERR(r);
}
r = name_cursor->c_get(name_cursor, &nc_key, &nc_data, DB_NEXT); // an uninitialized cursor should do a DB_FIRST.
if (r==0) {
cursor_count_n_items++;
} else if (r==DB_NOTFOUND) {
// Got to the end.
//printf("%s:%d Got to end count=%d curscount=%d\n", __FILE__, __LINE__, calc_n_items, cursor_count_n_items);
assert(cursor_count_n_items==calc_n_items);
r = name_cursor->c_get(name_cursor, &nc_key, &nc_data, DB_FIRST);
if (r==DB_NOTFOUND) {
nc_key.data = realloc(nc_key.data, 1);
((char*)nc_key.data)[0]=0;
cursor_count_n_items=0;
} else {
cursor_count_n_items=1;
}
calc_n_items = count_all_items;
}
}
int cursor_load=2; /* Set this to a higher number to do more cursor work for every insertion. Needed to get to the end. */
void activity (void) {
if (random()%20==0) {
// Delete the oldest expired one. Keep the cursor open
delete_oldest_expired();
} else if (random()%cursor_load==0) {
insert_person();
} else {
step_name();
}
//assert(count_all_items==count_entries(dbp));
}
void usage (const char *argv1) {
fprintf(stderr, "Usage:\n %s [ --DB-CREATE | --more ] seed ", argv1);
exit(1);
}
int main (int argc, const char *argv[]) {
const char *progname=argv[0];
int useseed;
{
struct timeval tv;
gettimeofday(&tv, 0);
useseed = tv.tv_sec+tv.tv_usec*997; // magic: 997 is a prime, and a million (microseconds/second) times 997 is still 32 bits.
}
memset(&nc_key, 0, sizeof(nc_key));
memset(&nc_data, 0, sizeof(nc_data));
nc_key.flags = DB_DBT_REALLOC;
nc_key.data = malloc(1); // Iniitalize it.
((char*)nc_key.data)[0]=0;
nc_data.flags = DB_DBT_REALLOC;
nc_data.data = malloc(1); // Iniitalize it.
mode = MODE_DEFAULT;
argv++; argc--;
while (argc>0) {
if (strcmp(argv[0], "--DB_CREATE")==0) {
mode = MODE_DB_CREATE;
} else if (strcmp(argv[0], "--more")==0) {
mode = MODE_MORE;
} else {
errno=0;
char *endptr;
useseed = strtoul(argv[0], &endptr, 10);
if (errno!=0 || *endptr!=0 || endptr==argv[0]) {
usage(progname);
}
}
argc--; argv++;
}
printf("seed=%d\n", useseed);
srandom(useseed);
switch (mode) {
case MODE_DEFAULT:
system("rm -rf " DIR);
mkdir(DIR, 0777);
create_databases();
{
int i;
for (i=0; i<100; i++)
activity();
}
break;
case MODE_MORE:
create_databases();
calc_n_items = count_all_items = count_entries(dbp);
//printf("%s:%d n_items initially=%d\n", __FILE__, __LINE__, count_all_items);
{
const int n_activities = 100000;
int i;
cursor_load = 8*(1+2*count_all_items/n_activities);
printf("%s:%d count=%d cursor_load=%d\n", __FILE__, __LINE__, count_all_items, cursor_load);
for (i=0; i<n_activities; i++)
activity();
}
break;
case MODE_DB_CREATE:
do_create();
break;
}
close_databases();
return 0;
}
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