/**
* Delete an item from the cache
* @param key the key of the item to delete
* @param nkey the length of the key
* @return true if the item was deleted from the cache
*/
bool delete_item(const void* key, size_t nkey) {
ib_trx_t transaction= ib_trx_begin(IB_TRX_REPEATABLE_READ);
bool ret= do_delete_item(transaction, key, nkey);
if (ret)
{
/* object found. commit transaction */
ib_err_t error= ib_trx_commit(transaction);
if (error != DB_SUCCESS)
{
fprintf(stderr, "Failed to delete key:\n\t%s\n",
ib_strerror(error));
ret= false;
}
}
else
{
ib_err_t error= ib_trx_rollback(transaction);
if (error != DB_SUCCESS)
fprintf(stderr, "Failed to roll back the transaction:\n\t%s\n",
ib_strerror(error));
}
return ret;
}
/**
* Flush the entire cache
* @param when when the cache should be flushed (0 == immediately)
*/
void flush(uint32_t when __attribute__((unused)))
{
/* @TODO implement support for when != 0 */
ib_trx_t transaction= ib_trx_begin(IB_TRX_REPEATABLE_READ);
ib_crsr_t cursor= NULL;
ib_err_t err= DB_SUCCESS;
checked(ib_cursor_open_table(tablename, transaction, &cursor));
checked(ib_cursor_first(cursor));
checked(ib_cursor_lock(cursor, IB_LOCK_X));
do
{
checked(ib_cursor_delete_row(cursor));
} while ((err= ib_cursor_next(cursor)) == DB_SUCCESS);
if (err != DB_END_OF_INDEX)
{
fprintf(stderr, "Failed to flush the cache: %s\n", ib_strerror(err));
goto error_exit;
}
ib_cursor_close(cursor);
cursor= NULL;
checked(ib_trx_commit(transaction));
return;
error_exit:
if (cursor != NULL)
ib_cursor_close(cursor);
ib_err_t error= ib_trx_rollback(transaction);
if (error != DB_SUCCESS)
fprintf(stderr, "Failed to roll back the transaction:\n\t%s\n",
ib_strerror(error));
}
bool init_db(void)
{
ib_err_t err;
/* Initialize the memory sub-system. */
ib_init();
/* Call the ib_cfg_*() functions to setup the directory etc. */
err = ib_cfg_set_bool_off("adaptive_hash_index");
err = ib_cfg_set_int("additional_mem_pool_size", 16*1024*1024);
err = ib_cfg_set_int("buffer_pool_size", 1024*1024*1024);
err = ib_cfg_set_int("flush_log_at_trx_commit", 2);
err = ib_cfg_set_int("log_buffer_size", 8*1024*1024);
err = ib_cfg_set_int("log_file_size", 256*1024*1024);
err = ib_cfg_set_text("data_home_dir", "/data/");
err = ib_cfg_set_text("log_group_home_dir", "/data/");
err = ib_cfg_set_text("flush_method", "O_DIRECT");
/* Create system files if this is the first time
* or do recovery if starting an existing instance. */
err = ib_startup("barracuda");
/* File format "barracuda" supports all
* the currently available table formats. */
if (err == DB_SUCCESS) {
printf("InnoDB started!\n");
} else {
printf("Error starting up InnoDB: %s\n", ib_strerror(err));
return false;
}
return true;
}
/**
* Initialize the database storage
* @return true if the database was initialized successfully, false otherwise
*/
bool initialize_storage(void)
{
ib_err_t error;
ib_id_t tid;
checked(ib_init());
checked(ib_cfg_set_text("data_home_dir", "/tmp/memcached_light"));
checked(ib_cfg_set_text("log_group_home_dir", "/tmp/memcached_light"));
checked(ib_cfg_set_bool_on("file_per_table"));
checked(ib_startup("barracuda"));
/* check to see if the table exists or if we should create the schema */
error= ib_table_get_id(tablename, &tid);
if (error == DB_TABLE_NOT_FOUND)
{
if (!create_schema())
{
return false;
}
}
else if (error != DB_SUCCESS)
{
fprintf(stderr, "Failed to get table id: %s\n", ib_strerror(error));
return false;
}
return true;
error_exit:
return false;
}
static void innostore_drv_finish()
{
erl_drv_mutex_destroy(G_ENGINE_STATE_LOCK);
// Shutdown the engine, if it's running -- note that this blocks the
// the calling VM thread and may be a long running operation.
if (G_ENGINE_STATE == ENGINE_STARTED)
{
ib_err_t result = ib_shutdown(IB_SHUTDOWN_NORMAL);
if (result != DB_SUCCESS)
{
log("ib_shutdown failed: %s\n", ib_strerror(result));
}
}
// Clean up logging after inno has shutdown completely
erl_drv_mutex_destroy(G_LOGGER_LOCK);
if (G_LOGGER_BUF != NULL)
{
driver_free(G_LOGGER_BUF);
G_LOGGER_BUF = NULL;
G_LOGGER_SIZE = 0;
}
if (G_LOGGER_FH != NULL)
{
fclose(G_LOGGER_FH);
G_LOGGER_FH = NULL;
G_LOGGER_FN = raw_logger;
}
G_ENGINE_STATE = ENGINE_STOPPED;
}
/*********************************************************************
CREATE TABLE T(C1 INT, C2 VARCHAR(10), C3 BLOB); */
static
ib_err_t
create_table(
/*=========*/
const char* dbname, /*!< in: database name */
const char* name) /*!< in: table name */
{
ib_trx_t ib_trx;
ib_id_t table_id = 0;
ib_err_t err = DB_SUCCESS;
ib_tbl_sch_t ib_tbl_sch = NULL;
char table_name[IB_MAX_TABLE_NAME_LEN];
#ifdef __WIN__
sprintf(table_name, "%s/%s", dbname, name);
#else
snprintf(table_name, sizeof(table_name), "%s/%s", dbname, name);
#endif
/* Pass a table page size of 0, ie., use default page size. */
err = ib_table_schema_create(
table_name, &ib_tbl_sch, IB_TBL_COMPACT, 0);
assert(err == DB_SUCCESS);
err = ib_table_schema_add_col(
ib_tbl_sch, "c1", IB_INT, IB_COL_NONE, 0, sizeof(ib_i32_t));
assert(err == DB_SUCCESS);
err = ib_tbl_sch_add_varchar_col(ib_tbl_sch, "c2", 10);
assert(err == DB_SUCCESS);
err = ib_tbl_sch_add_blob_col(ib_tbl_sch, "c3");
assert(err == DB_SUCCESS);
/* create table */
ib_trx = ib_trx_begin(IB_TRX_REPEATABLE_READ);
err = ib_schema_lock_exclusive(ib_trx);
assert(err == DB_SUCCESS);
err = ib_table_create(ib_trx, ib_tbl_sch, &table_id);
if (err == DB_SUCCESS) {
err = ib_trx_commit(ib_trx);
} else {
fprintf(stderr, "Table: %s create failed: %s\n",
table_name, ib_strerror(err));
err = ib_trx_rollback(ib_trx);
}
assert(err == DB_SUCCESS);
if (ib_tbl_sch != NULL) {
ib_table_schema_delete(ib_tbl_sch);
}
return(err);
}
int main(int argc, char *argv[])
{
if (!init_db()) { exit(1); }
ib_crsr_t crsr, index_crsr;
ib_tpl_t tpl;
ib_err_t err;
char *l6_orderkey_rle = "l5_customer_key_rle";
char cstore_l6_orderkey_rle[64];
sprintf(cstore_l6_orderkey_rle, "%s/%s", db_name, l6_orderkey_rle);
ib_trx_t ib_trx = ib_trx_begin(IB_TRX_REPEATABLE_READ);
err = ib_cursor_open_table(cstore_l6_orderkey_rle, ib_trx, &crsr);
if (err != DB_SUCCESS) { std::cout << ib_strerror(err) << " " << cstore_l6_orderkey_rle << std::endl; }
err = ib_cursor_open_index_using_name(crsr, "SECONDARY_KEY", &index_crsr);
if (err != DB_SUCCESS) { std::cout << ib_strerror(err) << std::endl; }
ib_cursor_set_cluster_access(index_crsr);
ib_tpl_t key = ib_sec_search_tuple_create(index_crsr);
ib_tuple_write_u32(key, 0, 330575);
int res;
err = ib_cursor_moveto(index_crsr, key, IB_CUR_GE, &res);
std::cout << "res: " << res << std::endl;
assert(err == DB_SUCCESS || err == DB_END_OF_INDEX || err == DB_RECORD_NOT_FOUND);
tpl = ib_clust_read_tuple_create(crsr);
while (err == DB_SUCCESS) {
err = ib_cursor_read_row(index_crsr, tpl);
if (err == DB_RECORD_NOT_FOUND || err == DB_END_OF_INDEX) {
std::cerr << "not found" << std::endl;
}
ib_u32_t orderkey, pos, freq;
ib_tuple_read_u32(tpl, 0, &orderkey);
ib_tuple_read_u32(tpl, 1, &pos);
ib_tuple_read_u32(tpl, 2, &freq);
std::cout << "orderkey: " << orderkey << ", pos: " << pos << ", freq: " << freq << std::endl;
err = ib_cursor_next(index_crsr);
tpl = ib_tuple_clear(tpl);
break;
}
ib_cursor_close(index_crsr);
ib_cursor_close(crsr);
ib_trx_commit(ib_trx);
fin_db();
return 0;
}
bool fin_db(void)
{
ib_err_t err = ib_shutdown(IB_SHUTDOWN_NORMAL);
if (err == DB_SUCCESS) {
printf("InnoDB shutdown succeed!\n");
} else {
printf("InnoDB shutdown failed: err %s\n", ib_strerror(err));
}
}
/**
* Store an item in the databse
*
* @param item the item to store
*/
void put_item(struct item* item)
{
ib_trx_t transaction= ib_trx_begin(IB_TRX_SERIALIZABLE);
if (do_put_item(transaction, item))
{
ib_err_t error= ib_trx_commit(transaction);
if (error != DB_SUCCESS)
{
fprintf(stderr, "Failed to store key:\n\t%s\n",
ib_strerror(error));
}
}
else
{
ib_err_t error= ib_trx_rollback(transaction);
if (error != DB_SUCCESS)
fprintf(stderr, "Failed to roll back the transaction:\n\t%s\n",
ib_strerror(error));
}
}
static void do_start(void* arg)
{
PortState* state = (PortState*)arg;
erl_drv_mutex_lock(G_ENGINE_STATE_LOCK);
if (G_ENGINE_STATE == ENGINE_STOPPED)
{
// Engine was stopped; set the flag, unlock and run the start
G_ENGINE_STATE = ENGINE_STARTING;
erl_drv_mutex_unlock(G_ENGINE_STATE_LOCK);
// Run the startup without holding any lock -- this can take a while
// if we are recovering from previous errors
ib_err_t error = ib_startup("barracuda");
if (error == DB_SUCCESS)
{
// Make sure the innokeystore database exists
// TODO: Avoid hard-coding the db name here...
if (ib_database_create("innokeystore") != IB_TRUE)
{
error = DB_ERROR;
}
}
// Relock and sort out results
erl_drv_mutex_lock(G_ENGINE_STATE_LOCK);
if (error == DB_SUCCESS)
{
G_ENGINE_STATE = ENGINE_STARTED;
send_ok(state);
}
else
{
G_ENGINE_STATE = ENGINE_STOPPED;
send_error_str(state, ib_strerror(error));
}
}
else if (G_ENGINE_STATE == ENGINE_STARTED)
{
// another thread has already completed startup
send_ok(state);
}
else
{
// Engine was not in stopped state when do_start was called.
// Probably due to multiple threads trying to start at the
// same time.
assert(G_ENGINE_STATE == ENGINE_STARTING);
send_error_atom(state, "starting");
}
erl_drv_mutex_unlock(G_ENGINE_STATE_LOCK);
}
/**
* Get an item from the engine
* @param key the key to grab
* @param nkey number of bytes in the key
* @return pointer to the item if found
*/
struct item* get_item(const void* key, size_t nkey)
{
ib_trx_t transaction= ib_trx_begin(IB_TRX_SERIALIZABLE);
struct item* ret= do_get_item(transaction, key, nkey);
ib_err_t error= ib_trx_rollback(transaction);
if (error != DB_SUCCESS)
fprintf(stderr, "Failed to roll back the transaction:\n\t%s\n",
ib_strerror(error));
return ret;
}
std::vector<uint32_t>
apply_l5_predicate_scan(std::vector<im_val_t> pos_custkey_array)
{
ib_err_t err;
std::vector<uint32_t> pos_array;
ib_crsr_t index_crsr;
err = ib_cursor_open_index_using_name(crsr_l5_c, "SECONDARY_KEY", &index_crsr);
if (err != DB_SUCCESS) { std::cout << ib_strerror(err) << std::endl; }
ib_cursor_set_cluster_access(index_crsr);
ib_tpl_t key_l5_c = ib_sec_search_tuple_create(index_crsr);
ib_tpl_t key_l5_n = ib_clust_search_tuple_create(crsr_l5_n);
size_t size = pos_custkey_array.size();
int res;
// TODO: first fetches all l5_pos for custkeys
std::vector<uint32_t> l5_pos_array;
for (int i = 0; i < size; ++i) {
// custkey
ib_tuple_write_u32(key_l5_c, 0, pos_custkey_array[i].val);
err = ib_cursor_moveto(index_crsr, key_l5_c, IB_CUR_GE, &res);
if (err != DB_SUCCESS) {
std::cout << "failed moving cursor for l5_c" << std::endl;
}
err = ib_cursor_read_row(index_crsr, tpl_l5_c);
if (err == DB_RECORD_NOT_FOUND || err == DB_END_OF_INDEX) {
}
ib_u32_t l5_pos;
ib_tuple_read_u32(tpl_l5_c, 1, &l5_pos);
l5_pos_array.push_back(l5_pos);
tpl_l5_c = ib_tuple_clear(tpl_l5_c);
}
for (int i = 0; i < size; ++i) {
ib_tuple_write_u32(key_l5_n, 0, l5_pos_array[i]);
err = ib_cursor_moveto(crsr_l5_n, key_l5_n, IB_CUR_GE, &res);
if (err != DB_SUCCESS) {
std::cout << "failed moving cursor for l5_c" << std::endl;
}
if (res == 0) {
err = ib_cursor_read_row(crsr_l5_n, tpl_l5_n);
ib_u32_t c_nationkey;
ib_tuple_read_u32(tpl_l5_n, 1, &c_nationkey);
if (c_nationkey == pred_c_nationkey) {
pos_array.push_back(pos_custkey_array[i].pos);
}
}
tpl_l5_n = ib_tuple_clear(tpl_l5_n);
}
ib_tuple_delete(key_l5_c);
ib_tuple_delete(key_l5_n);
ib_cursor_close(index_crsr);
return pos_array;
}
/**
* Try to locate an item in the database. Return a cursor and the tuple to
* the item if I found it in the database.
*
* @param trx the transaction to use
* @param key the key of the item to look up
* @param nkey the size of the key
* @param cursor where to store the cursor (OUT)
* @param tuple where to store the tuple (OUT)
* @return true if I found the object, false otherwise
*/
static bool do_locate_item(ib_trx_t trx,
const void* key,
size_t nkey,
ib_crsr_t *cursor)
{
int res;
ib_tpl_t tuple= NULL;
*cursor= NULL;
checked(ib_cursor_open_table(tablename, trx, cursor));
tuple= ib_clust_search_tuple_create(*cursor);
if (tuple == NULL)
{
fprintf(stderr, "Failed to allocate tuple object\n");
goto error_exit;
}
checked(ib_col_set_value(tuple, key_col_idx, key, nkey));
ib_err_t err= ib_cursor_moveto(*cursor, tuple, IB_CUR_GE, &res);
if (err == DB_SUCCESS && res == 0)
{
ib_tuple_delete(tuple);
return true;
}
else if (err != DB_SUCCESS &&
err != DB_RECORD_NOT_FOUND &&
err != DB_END_OF_INDEX)
{
fprintf(stderr, "ERROR: ib_cursor_moveto(): %s\n", ib_strerror(err));
}
/* FALLTHROUGH */
error_exit:
if (tuple != NULL)
ib_tuple_delete(tuple);
if (*cursor != NULL)
{
ib_err_t cursor_error;
cursor_error= ib_cursor_close(*cursor);
(void) cursor_error;
}
*cursor= NULL;
return false;
}
static void do_get(void* arg)
{
PortState* state = (PortState*)arg;
// Unpack key from work buffer
// table - 8 bytes
// keysz - 1 byte
// key - variable
ib_id_t table ; UNPACK_INT(state->work_buffer, 0, &table);
unsigned char keysz = UNPACK_BYTE(state->work_buffer, sizeof(table));
char* key = UNPACK_BLOB(state->work_buffer, sizeof(table)+1);
ib_trx_t txn = ib_trx_begin(IB_TRX_REPEATABLE_READ);
ib_crsr_t cursor;
FAIL_ON_ERROR(ib_cursor_open_table_using_id(table, txn, &cursor),
ROLLBACK_RETURN(txn));
ib_tpl_t key_tuple = ib_clust_search_tuple_create(cursor);
ib_col_set_value(key_tuple, KEY_COL, key, keysz);
int searchloc;
ib_err_t error = ib_cursor_moveto(cursor, key_tuple, IB_CUR_GE, &searchloc);
// Drop the key tuple -- cursor is at desired location
ib_tuple_delete(key_tuple);
// If we encountered an error, bail
if (error != DB_SUCCESS && error != DB_RECORD_NOT_FOUND && error != DB_END_OF_INDEX)
{
ib_cursor_close(cursor);
send_error_str(state, ib_strerror(error));
ROLLBACK_RETURN(txn);
}
// Found it, read the value and send back to caller
if (searchloc == 0)
{
ib_tpl_t tuple = ib_clust_read_tuple_create(cursor);
// TODO: May need better error handling here
FAIL_ON_ERROR(ib_cursor_read_row(cursor, tuple),
{
ib_tuple_delete(tuple);
ib_cursor_close(cursor);
ROLLBACK_RETURN(txn);
});
/**
* Create the database schema.
* @return true if the database schema was created without any problems
* false otherwise.
*/
static bool create_schema(void)
{
ib_tbl_sch_t schema= NULL;
ib_idx_sch_t dbindex= NULL;
if (ib_database_create("memcached") != IB_TRUE)
{
fprintf(stderr, "Failed to create database\n");
return false;
}
ib_trx_t transaction= ib_trx_begin(IB_TRX_SERIALIZABLE);
ib_id_t table_id;
checked(ib_table_schema_create(tablename, &schema, IB_TBL_COMPACT, 0));
checked(ib_table_schema_add_col(schema, "key", IB_BLOB,
IB_COL_NOT_NULL, 0, 32767));
checked(ib_table_schema_add_col(schema, "data", IB_BLOB,
IB_COL_NONE, 0, 1024*1024));
checked(ib_table_schema_add_col(schema, "flags", IB_INT,
IB_COL_UNSIGNED, 0, 4));
checked(ib_table_schema_add_col(schema, "cas", IB_INT,
IB_COL_UNSIGNED, 0, 8));
checked(ib_table_schema_add_col(schema, "exp", IB_INT,
IB_COL_UNSIGNED, 0, 4));
checked(ib_table_schema_add_index(schema, "PRIMARY_KEY", &dbindex));
checked(ib_index_schema_add_col(dbindex, "key", 0));
checked(ib_index_schema_set_clustered(dbindex));
checked(ib_schema_lock_exclusive(transaction));
checked(ib_table_create(transaction, schema, &table_id));
checked(ib_trx_commit(transaction));
ib_table_schema_delete(schema);
return true;
error_exit:
/* @todo release resources! */
{
ib_err_t error= ib_trx_rollback(transaction);
if (error != DB_SUCCESS)
fprintf(stderr, "Failed to roll back the transaction:\n\t%s\n",
ib_strerror(error));
}
return false;
}
int main(int argc,char **argv){
if(argc != 2){
printf("give me a database\table name\n");
return 1;
}
char *dtname=argv[1];
ib_err_t err;
err=ib_init();
if(err != DB_SUCCESS){
puts(ib_strerror(err));
return err;
}
err = ib_cfg_set_int("log_buffer_size", 8*1024*1024);
if(err != DB_SUCCESS){
puts(ib_strerror(err));
return err;
}
err = ib_cfg_set_int("force_recovery", 1);
if(err != DB_SUCCESS){
puts(ib_strerror(err));
return err;
}
err = ib_cfg_set_int("log_file_size", 128*1024*1024);
if(err != DB_SUCCESS){
puts(ib_strerror(err));
return err;
}
err = ib_cfg_set_int("log_files_in_group", 3);
if(err != DB_SUCCESS){
puts(ib_strerror(err));
return err;
}
err = ib_cfg_set_text("log_group_home_dir", "./");
//err = ib_cfg_set_text("log_group_home_dir", "/var/lib/mysql/");
if(err != DB_SUCCESS){
puts(ib_strerror(err));
return err;
}
err = ib_cfg_set_text("data_home_dir", "./");
if(err != DB_SUCCESS){
puts(ib_strerror(err));
return err;
}
err = ib_cfg_set_text("data_file_path", "ibdata1:500M:autoextend");
if(err != DB_SUCCESS){
puts(ib_strerror(err));
return err;
}
err = ib_cfg_set_bool_on("file_per_table");
if(err != DB_SUCCESS){
puts(ib_strerror(err));
return err;
}
err=ib_startup("Antelope");
if(err != DB_SUCCESS){
puts(ib_strerror(err));
return err;
}
ib_trx_t trx;
ib_crsr_t crsr;
trx=ib_trx_begin(IB_TRX_REPEATABLE_READ);
assert(trx != NULL);
err=ib_cursor_open_table(dtname,trx,&crsr);
if(err != DB_SUCCESS){
puts(ib_strerror(err));
return err;
}
ib_tpl_t tpl;
tpl=ib_clust_read_tuple_create(crsr);
assert(tpl != NULL);
err=ib_cursor_first(crsr);
while(err == DB_SUCCESS){
err=ib_cursor_read_row(crsr,tpl);
print_tuple(stdout,tpl);
err=ib_cursor_next(crsr);
tpl=ib_tuple_clear(tpl);
}
ib_tuple_delete(tpl);
err=ib_cursor_close(crsr);
if(err != DB_SUCCESS){
puts(ib_strerror(err));
return err;
}
err=ib_trx_commit(trx);
if(err != DB_SUCCESS){
puts(ib_strerror(err));
return err;
}
err=ib_shutdown(IB_SHUTDOWN_NORMAL);
if(err != DB_SUCCESS){
puts(ib_strerror(err));
return err;
}
return 0;
}
static void do_set_cfg(void* arg)
{
PortState* state = (PortState*)arg;
erl_drv_mutex_lock(G_ENGINE_STATE_LOCK);
if (G_ENGINE_STATE == ENGINE_STOPPED)
{
char* key = UNPACK_STRING(state->work_buffer, 0);
const char* value = UNPACK_STRING(state->work_buffer, strlen(key)+1);
if (strcmp(key, "error_log") == 0)
{
if (set_log_file(value) == 0)
{
send_ok(state);
}
else
{
send_error_atom(state, "einval");
}
}
else
{
// Check the expected type of the provided key so as to 1. validate it's a good key
// and 2. know what setter to use.
ib_cfg_type_t key_type;
ib_err_t error = ib_cfg_var_get_type(key, &key_type);
if (error == DB_SUCCESS)
{
if (key_type == IB_CFG_TEXT)
{
// HACK: Semantics of setting a text configuration value for innodb changed
// to be pointer assignment (from copy) for vsn 1.0.6.6750. So, we strdup the
// value to ensure everything works as expected.
// TODO: Setup some sort of list of strdup'd values to ensure they all get
// cleaned up properly. In typical usage, this isn't going to be a problem
// as you only initialize once per run, but it bothers me just the same.
error = ib_cfg_set(key, strdup(value));
}
else
{
ErlDrvUInt value_i;
UNPACK_INT(state->work_buffer, strlen(key)+1, &value_i);
error = ib_cfg_set(key, value_i);
}
}
if (error == DB_SUCCESS)
{
send_ok(state);
}
else
{
send_error_str(state, ib_strerror(error));
}
}
}
else
{
send_error_atom(state, "starting");
}
erl_drv_mutex_unlock(G_ENGINE_STATE_LOCK);
}
static void do_init_table(void* arg)
{
PortState* state = (PortState*)arg;
unsigned char fmt_code = UNPACK_BYTE(state->work_buffer, 0);
// Unpack the table name (pre-formatted to be Database/TableName)
char* table = UNPACK_STRING(state->work_buffer, 1);
ib_id_t table_id;
ib_tbl_fmt_t table_fmt;
switch(fmt_code)
{
case FORMAT_REDUNDANT:
table_fmt = IB_TBL_REDUNDANT;
break;
case FORMAT_COMPACT:
table_fmt = IB_TBL_COMPACT;
break;
case FORMAT_DYNAMIC:
table_fmt = IB_TBL_DYNAMIC;
break;
case FORMAT_COMPRESSED:
table_fmt = IB_TBL_COMPRESSED;
break;
default:
send_error_atom(state, "bad_format");
return;
}
// If the table doesn't exist, create it
if (ib_table_get_id(table, &table_id) != DB_SUCCESS)
{
// Start a txn for schema access and be sure to make it serializable
ib_trx_t txn = ib_trx_begin(IB_TRX_SERIALIZABLE);
ib_schema_lock_exclusive(txn);
// Create the table schema
ib_tbl_sch_t schema;
ib_table_schema_create(table, &schema, table_fmt, 0);
ib_table_schema_add_col(schema, "key", IB_VARBINARY, IB_COL_NONE, 0, 255);
ib_table_schema_add_col(schema, "value", IB_BLOB, IB_COL_NONE, 0, 0);
// Create primary index on key
ib_idx_sch_t index;
ib_table_schema_add_index(schema, "PRIMARY_KEY", &index);
ib_index_schema_add_col(index, "key", 0);
ib_index_schema_set_clustered(index);
// Create the actual table
ib_err_t rc = ib_table_create(txn, schema, &table_id);
// Release the schema -- doesn't matter if table was created or not at this point
ib_schema_unlock(txn);
if (rc == DB_SUCCESS)
{
// Commit changes to schema (if any)
ib_trx_commit(txn);
}
else
{
// Failed to create table -- rollback and exit
ib_trx_rollback(txn);
send_error_str(state, ib_strerror(rc));
return;
}
}
// Guaranteed at this point to have a valid table_id
ErlDrvTermData response[] = { ERL_DRV_ATOM, driver_mk_atom("innostore_ok"),
ERL_DRV_BUF2BINARY, (ErlDrvTermData)&table_id,
(ErlDrvUInt)sizeof(table_id),
ERL_DRV_TUPLE, 2};
driver_send_term(state->port, state->port_owner, response,
sizeof(response) / sizeof(response[0]));
}
/*********************************************************************
CREATE TABLE T(
vchar VARCHAR(128),
blob VARCHAR(n),
count INT,
PRIMARY KEY(vchar); */
static
ib_err_t
create_table(
/*=========*/
const char* dbname, /*!< in: database name */
const char* name) /*!< in: table name */
{
ib_trx_t ib_trx;
ib_id_t table_id = 0;
ib_err_t err = DB_SUCCESS;
ib_tbl_sch_t ib_tbl_sch = NULL;
ib_idx_sch_t ib_idx_sch = NULL;
ib_tbl_fmt_t tbl_fmt = IB_TBL_COMPACT;
char table_name[IB_MAX_TABLE_NAME_LEN];
#ifdef __WIN__
sprintf(table_name, "%s/%s", dbname, name);
#else
snprintf(table_name, sizeof(table_name), "%s/%s", dbname, name);
#endif
if (page_size > 0) {
tbl_fmt = IB_TBL_COMPRESSED;
printf("Creating compressed table with page size %d\n",
page_size);
}
err = ib_table_schema_create(
table_name, &ib_tbl_sch, tbl_fmt, page_size);
assert(err == DB_SUCCESS);
err = ib_table_schema_add_col(
ib_tbl_sch, "vchar",
IB_VARCHAR, IB_COL_NONE, 0, 128);
assert(err == DB_SUCCESS);
err = ib_table_schema_add_col(
ib_tbl_sch, "blob",
IB_BLOB, IB_COL_NONE, 0, 0);
assert(err == DB_SUCCESS);
err = ib_table_schema_add_col(
ib_tbl_sch, "count",
IB_INT, IB_COL_UNSIGNED, 0, 4);
assert(err == DB_SUCCESS);
err = ib_table_schema_add_index(ib_tbl_sch, "PRIMARY", &ib_idx_sch);
assert(err == DB_SUCCESS);
/* Set prefix length to 0. */
err = ib_index_schema_add_col( ib_idx_sch, "vchar", 0);
assert(err == DB_SUCCESS);
err = ib_index_schema_set_clustered(ib_idx_sch);
assert(err == DB_SUCCESS);
err = ib_index_schema_set_unique(ib_idx_sch);
assert(err == DB_SUCCESS);
/* create table */
ib_trx = ib_trx_begin(IB_TRX_REPEATABLE_READ);
err = ib_schema_lock_exclusive(ib_trx);
assert(err == DB_SUCCESS);
err = ib_table_create(ib_trx, ib_tbl_sch, &table_id);
if (err != DB_SUCCESS) {
fprintf(stderr, "Warning: table create failed: %s\n",
ib_strerror(err));
err = ib_trx_rollback(ib_trx);
} else {
err = ib_trx_commit(ib_trx);
}
assert(err == DB_SUCCESS);
if (ib_tbl_sch != NULL) {
ib_table_schema_delete(ib_tbl_sch);
}
return(err);
}
void q3(void)
{
char *l1_shipdate_rle = "l1_shipdate_rle";
char *l1_extendedprice_uncompressed = "l1_extendedprice_uncompressed";
char *l3_orderdate_rle = "l3_orderdate_rle";
char *l3_custkey_uncompressed = "l3_custkey_uncompressed";
char *l3_orderkey_uncompressed = "l3_orderkey_uncompressed";
char *l5_customer_key_rle = "l5_customer_key_rle";
char *l5_nationkey_uncompressed = "l5_nationkey_uncompressed";
char *l6_orderkey_rle = "l6_orderkey_rle";
char *l6_extendedprice_uncompressed = "l6_extendedprice_uncompressed";
char *l6_shipdate_uncompressed = "l6_shipdate_uncompressed";
char *ji_l3_l1 = "ji_l3-l1";
char cstore_l1_shipdate_rle[64];
char cstore_l1_extendedprice_uncompressed[64];
char cstore_l3_orderdate_rle[64];
char cstore_l3_custkey_uncompressed[64];
char cstore_l3_orderkey_uncompressed[64];
char cstore_l5_customer_key_rle[64];
char cstore_l5_nationkey_uncompressed[64];
char cstore_l6_orderkey_rle[64];
char cstore_l6_extendedprice_uncompressed[64];
char cstore_l6_shipdate_uncompressed[64];
char cstore_ji_l3_l1[64];
sprintf(cstore_l1_shipdate_rle, "%s/%s", db_name, l1_shipdate_rle);
sprintf(cstore_l1_extendedprice_uncompressed, "%s/%s", db_name, l1_extendedprice_uncompressed);
sprintf(cstore_l3_orderdate_rle, "%s/%s", db_name, l3_orderdate_rle);
sprintf(cstore_l3_custkey_uncompressed, "%s/%s", db_name, l3_custkey_uncompressed);
sprintf(cstore_l3_orderkey_uncompressed, "%s/%s", db_name, l3_orderkey_uncompressed);
sprintf(cstore_l5_customer_key_rle, "%s/%s", db_name, l5_customer_key_rle);
sprintf(cstore_l5_nationkey_uncompressed, "%s/%s", db_name, l5_nationkey_uncompressed);
sprintf(cstore_l6_orderkey_rle, "%s/%s", db_name, l6_orderkey_rle);
sprintf(cstore_l6_extendedprice_uncompressed, "%s/%s", db_name, l6_extendedprice_uncompressed);
sprintf(cstore_l6_shipdate_uncompressed, "%s/%s", db_name, l6_shipdate_uncompressed);
sprintf(cstore_ji_l3_l1, "%s/%s", db_name, ji_l3_l1);
double t0 = gettimeofday_sec();
ib_err_t err;
//ib_crsr_t crsr_l1_s, crsr_l1_e, crsr_l3_o, crsr_l3_c, crsr_l5_c, crsr_l5_n, crsr_ji;
ib_trx_t ib_trx = ib_trx_begin(IB_TRX_REPEATABLE_READ);
err = ib_cursor_open_table(cstore_l1_shipdate_rle, ib_trx, &crsr_l1_s);
if (err != DB_SUCCESS) { std::cout << ib_strerror(err) << " " << cstore_l1_shipdate_rle << std::endl; }
err = ib_cursor_open_table(cstore_l1_extendedprice_uncompressed, ib_trx, &crsr_l1_e);
if (err != DB_SUCCESS) { std::cout << ib_strerror(err) << " " << cstore_l1_extendedprice_uncompressed << std::endl; }
err = ib_cursor_open_table(cstore_l3_orderdate_rle, ib_trx, &crsr_l3_o);
if (err != DB_SUCCESS) { std::cout << ib_strerror(err) << " " << cstore_l3_orderdate_rle << std::endl; }
err = ib_cursor_open_table(cstore_l3_custkey_uncompressed, ib_trx, &crsr_l3_c);
if (err != DB_SUCCESS) { std::cout << ib_strerror(err) << " " << cstore_l3_custkey_uncompressed << std::endl; }
err = ib_cursor_open_table(cstore_l3_orderkey_uncompressed, ib_trx, &crsr_l3_ok);
if (err != DB_SUCCESS) { std::cout << ib_strerror(err) << " " << cstore_l3_orderkey_uncompressed << std::endl; }
err = ib_cursor_open_table(cstore_l5_customer_key_rle, ib_trx, &crsr_l5_c);
if (err != DB_SUCCESS) { std::cout << ib_strerror(err) << " " << cstore_l5_customer_key_rle << std::endl; }
err = ib_cursor_open_table(cstore_l5_nationkey_uncompressed, ib_trx, &crsr_l5_n);
if (err != DB_SUCCESS) { std::cout << ib_strerror(err) << " " << cstore_l5_nationkey_uncompressed << std::endl; }
err = ib_cursor_open_table(cstore_l6_orderkey_rle, ib_trx, &crsr_l6_o);
if (err != DB_SUCCESS) { std::cout << ib_strerror(err) << " " << cstore_l6_orderkey_rle << std::endl; }
err = ib_cursor_open_table(cstore_l6_extendedprice_uncompressed, ib_trx, &crsr_l6_e);
if (err != DB_SUCCESS) { std::cout << ib_strerror(err) << " " << cstore_l6_extendedprice_uncompressed << std::endl; }
err = ib_cursor_open_table(cstore_l6_shipdate_uncompressed, ib_trx, &crsr_l6_s);
if (err != DB_SUCCESS) { std::cout << ib_strerror(err) << " " << cstore_l6_shipdate_uncompressed << std::endl; }
err = ib_cursor_open_table(cstore_ji_l3_l1, ib_trx, &crsr_ji);
if (err != DB_SUCCESS) { std::cout << ib_strerror(err) << " " << cstore_ji_l3_l1 << std::endl; }
tpl_l1_s = ib_clust_read_tuple_create(crsr_l1_s);
tpl_l1_e = ib_clust_read_tuple_create(crsr_l1_e);
tpl_l3_o = ib_clust_read_tuple_create(crsr_l3_o);
tpl_l3_c = ib_clust_read_tuple_create(crsr_l3_c);
tpl_l3_ok = ib_clust_read_tuple_create(crsr_l3_ok);
tpl_l5_c = ib_clust_read_tuple_create(crsr_l5_c);
tpl_l5_n = ib_clust_read_tuple_create(crsr_l5_n);
tpl_l6_o = ib_clust_read_tuple_create(crsr_l6_o);
tpl_l6_e = ib_clust_read_tuple_create(crsr_l6_e);
tpl_l6_s = ib_clust_read_tuple_create(crsr_l6_s);
tpl_ji = ib_clust_read_tuple_create(crsr_ji);
ib_tpl_t key = ib_clust_search_tuple_create(crsr_l3_o);
ib_tuple_write_u32(key, 0, pred_o_orderdate);
int res;
// move cursor
//err = ib_cursor_moveto(l3_o, key, IB_CUR_L, &res);
err = ib_cursor_first(crsr_l3_o);
ib_tuple_delete(key);
if (err == DB_SUCCESS) {
//std::cout << "success" << std::endl;
} else {
std::cout << "failed" << std::endl;
ib_cursor_close(crsr_l3_o);
return;
}
while (err == DB_SUCCESS) {
err = ib_cursor_read_row(crsr_l3_o, tpl_l3_o);
/* Possible handle locking and timeout errors too in multi-threaded applications. */
if (err == DB_RECORD_NOT_FOUND || err == DB_END_OF_INDEX) {
break;
}
ib_u32_t orderdate, pos, freq;
ib_tuple_read_u32(tpl_l3_o, 0, &orderdate);
if (orderdate >= pred_o_orderdate) { break; }
ib_tuple_read_u32(tpl_l3_o, 1, &pos);
ib_tuple_read_u32(tpl_l3_o, 2, &freq);
double t1 = gettimeofday_sec();
// get L3 o_custkey from position filtering
std::vector<im_val_t> pos_custkey_array = get_l3_custkeys_from_sequences(pos, freq);
std::sort(pos_custkey_array.begin(), pos_custkey_array.end(), val_asc);
/*
for (int i = 0; i < pos_custkey_array.size(); ++i) {
std::cout << pos_custkey_array[i].pos << ": " << pos_custkey_array[i].val << std::endl;
}
*/
double t2 = gettimeofday_sec();
std::cout << "orderdate: " << pred_o_orderdate << " get_l3_custkeys_from_sequences: " << t2 - t1 << std::endl;
// join with L5 and apply c_nationkey predicate
//std::vector<uint32_t> pos_array = apply_l5_predicate(pos_custkey_array);
std::vector<uint32_t> pos_array = apply_l5_predicate_scan(pos_custkey_array);
/*
for (int i = 0; i < pos_array.size(); ++i) {
std::cout << "pos: " << pos_array[i] << std::endl;
}
*/
double t3 = gettimeofday_sec();
std::cout << "orderdate: " << pred_o_orderdate << " apply_l5_predicate: " << t3 - t2 << std::endl;
// get L3 o_orderkey from position filtring
std::vector<im_val_t> pos_orderkey_array = get_l3_orderkeys_from_positions(pos_array);
/*
for (int i = 0; i < pos_orderkey_array.size(); ++i) {
std::cout << pos_orderkey_array[i].pos << ": " << pos_orderkey_array[i].val << std::endl;
}
*/
double t4 = gettimeofday_sec();
std::cout << "orderdate: " << pred_o_orderdate << " get_l3_orderkeys_from_positions: " << t4 - t3 << std::endl;
// join with L1 apply L1 l_shipdate predicate
// TODO: we don't need orderkey if we have L3->L1 join-index (we only need positions)
if (use_ji) {
std::vector<im_agg_t> extendedprice_array = apply_l1_predicate_via_ji(pos_orderkey_array);
double t5 = gettimeofday_sec();
std::cout << "orderdate: " << pred_o_orderdate << " apply_l1_predicate(JI): " << t5 - t4 << std::endl;
} else {
std::sort(pos_orderkey_array.begin(), pos_orderkey_array.end(), val_asc);
std::vector<im_agg_t> extendedprice_array = apply_l6_predicate(pos_orderkey_array);
double t5 = gettimeofday_sec();
std::cout << "orderdate: " << pred_o_orderdate << " apply_l6_predicate: " << t5 - t4 << std::endl;
}
/*
for (int i = 0; i < extendedprice_array.size(); ++i) {
std::cout << "orderdate: " << orderdate << ", orderkey: " << pos_orderkey_array[i].val << ", SUM(extendedprice): " << extendedprice_array[i].agg << std::endl;
}
*/
// here we have L3.o_orderdate, L1.l_extendedprice
// aggratation if needed (omitted)
err = ib_cursor_next(crsr_l3_o);
/* Possible handle locking and timeout errors too
* in multi-threaded applications. */
if (err == DB_RECORD_NOT_FOUND || err == DB_END_OF_INDEX) {
break;
}
tpl_l3_o = ib_tuple_clear(tpl_l3_o);
}
ib_tuple_delete(tpl_l3_o);
ib_cursor_close(crsr_l1_s);
ib_cursor_close(crsr_l1_e);
ib_cursor_close(crsr_l3_o);
ib_cursor_close(crsr_l3_c);
ib_cursor_close(crsr_l3_ok);
ib_cursor_close(crsr_l5_c);
ib_cursor_close(crsr_l5_n);
ib_cursor_close(crsr_l6_o);
ib_cursor_close(crsr_l6_e);
ib_cursor_close(crsr_l6_s);
ib_cursor_close(crsr_ji);
ib_trx_commit(ib_trx);
double t6 = gettimeofday_sec();
std::cout << "orderdate: " << pred_o_orderdate << " total:" << t6-t0 << std::endl;
}
std::vector<im_agg_t>
apply_l6_predicate(std::vector<im_val_t> pos_orderkey_array)
{
ib_err_t err;
std::vector<im_agg_t> extendedprice_array;
ib_crsr_t index_crsr;
err = ib_cursor_open_index_using_name(crsr_l6_o, "SECONDARY_KEY", &index_crsr);
if (err != DB_SUCCESS) { std::cout << ib_strerror(err) << std::endl; }
ib_cursor_set_cluster_access(index_crsr);
ib_tpl_t key_l6_o = ib_sec_search_tuple_create(index_crsr);
ib_tpl_t key_l6_s = ib_clust_search_tuple_create(crsr_l6_s);
ib_tpl_t key_l6_e = ib_clust_search_tuple_create(crsr_l6_e);
size_t size = pos_orderkey_array.size();
int res;
for (int i = 0; i < size; ++i) {
uint32_t sum_extendedprice = 0;
ib_tuple_write_u32(key_l6_o, 0, pos_orderkey_array[i].val);
err = ib_cursor_moveto(index_crsr, key_l6_o, IB_CUR_GE, &res);
if (err != DB_SUCCESS) {
std::cout << "failed moving cursor for index_crsr" << std::endl;
}
err = ib_cursor_read_row(index_crsr, tpl_l6_o);
if (err == DB_RECORD_NOT_FOUND || err == DB_END_OF_INDEX) {
}
ib_u32_t orderkey, pos, freq;
ib_tuple_read_u32(tpl_l6_o, 0, &orderkey);
ib_tuple_read_u32(tpl_l6_o, 1, &pos);
ib_tuple_read_u32(tpl_l6_o, 2, &freq);
//std::cout << "orderkey: " << orderkey << ", pos: " << pos << ", freq: " << freq << std::endl;
if (orderkey != pos_orderkey_array[i].val) {
break;
}
tpl_l6_o = ib_tuple_clear(tpl_l6_o);
for (int i = pos; i < pos + freq; ++i) {
ib_tuple_write_u32(key_l6_s, 0, i);
err = ib_cursor_moveto(crsr_l6_s, key_l6_s, IB_CUR_GE, &res);
if (err != DB_SUCCESS) {
std::cout << "failed moving cursor for l6_s" << std::endl;
}
err = ib_cursor_read_row(crsr_l6_s, tpl_l6_s);
ib_u32_t shipdate;
ib_tuple_read_u32(tpl_l6_s, 1, &shipdate);
//std::cout << "shipdate: " << shipdate << std::endl;
if (shipdate > pred_l_shipdate) {
ib_tuple_write_u32(key_l6_e, 0, i);
err = ib_cursor_moveto(crsr_l6_e, key_l6_e, IB_CUR_GE, &res);
if (err != DB_SUCCESS) {
std::cout << "failed moving cursor for l6_e" << std::endl;
}
err = ib_cursor_read_row(crsr_l6_e, tpl_l6_e);
ib_u32_t extendedprice;
ib_tuple_read_u32(tpl_l6_e, 1, &extendedprice);
sum_extendedprice += extendedprice;
tpl_l6_e = ib_tuple_clear(tpl_l6_e);
}
tpl_l6_s = ib_tuple_clear(tpl_l6_s);
}
im_agg_t im_agg = {sum_extendedprice};
extendedprice_array.push_back(im_agg);
}
ib_tuple_delete(key_l6_o);
ib_tuple_delete(key_l6_e);
ib_tuple_delete(key_l6_s);
ib_cursor_close(index_crsr);
return extendedprice_array;
}