/**
* @fn Datum repack_index_swap(PG_FUNCTION_ARGS)
* @brief Swap out an original index on a table with the newly-created one.
*
* repack_index_swap(index)
*
* @param index Oid of the *original* index.
* @retval void
*/
Datum
repack_index_swap(PG_FUNCTION_ARGS)
{
Oid orig_idx_oid = PG_GETARG_OID(0);
Oid repacked_idx_oid;
StringInfoData str;
SPITupleTable *tuptable;
TupleDesc desc;
HeapTuple tuple;
/* authority check */
must_be_superuser("repack_index_swap");
/* connect to SPI manager */
repack_init();
initStringInfo(&str);
/* Find the OID of our new index. */
appendStringInfo(&str, "SELECT oid FROM pg_class "
"WHERE relname = 'index_%u' AND relkind = 'i'",
orig_idx_oid);
execute(SPI_OK_SELECT, str.data);
if (SPI_processed != 1)
elog(ERROR, "Could not find index 'index_%u', found " UINT64_FORMAT " matches",
orig_idx_oid, (uint64) SPI_processed);
tuptable = SPI_tuptable;
desc = tuptable->tupdesc;
tuple = tuptable->vals[0];
repacked_idx_oid = getoid(tuple, desc, 1);
swap_heap_or_index_files(orig_idx_oid, repacked_idx_oid);
SPI_finish();
PG_RETURN_VOID();
}
void
remove_user(object user)
{
string datagram;
int userid;
userid = getoid(user); /* get number following # in file_name(user) */
datagram = header("Z") + TAB + mudname + TAB + userid;
send_data(datagram);
}
void
add_user(object user, int which)
{
string name, datagram;
int userid, login_time;
if (!user) {
return;
}
userid = getoid(user);
/* refresh approx. 1/PARTITIONS of list each time */
if ((userid % PARTITIONS) != which) {
return;
}
login_time = (int)user->QUERY_LOGIN_TIME;
name = (string)user->QUERY_NAME;
datagram =
header("A")+TAB+mudname+TAB+userid+TAB+login_time+TAB+GENERATION
+TAB+name;
send_data(datagram);
}
/**
* @fn Datum reorg_swap(PG_FUNCTION_ARGS)
* @brief Swapping relfilenode of tables and relation ids of toast tables
* and toast indexes.
*
* reorg_swap(oid, relname)
*
* TODO: remove useless CommandCounterIncrement().
*
* @param oid Oid of table of target.
* @retval None.
*/
Datum
reorg_swap(PG_FUNCTION_ARGS)
{
Oid oid = PG_GETARG_OID(0);
const char *relname = get_quoted_relname(oid);
const char *nspname = get_quoted_nspname(oid);
Oid argtypes[1] = { OIDOID };
bool nulls[1] = { 0 };
Datum values[1];
SPITupleTable *tuptable;
TupleDesc desc;
HeapTuple tuple;
uint32 records;
uint32 i;
Oid reltoastrelid1;
Oid reltoastidxid1;
Oid oid2;
Oid reltoastrelid2;
Oid reltoastidxid2;
Oid owner1;
Oid owner2;
/* authority check */
must_be_superuser("reorg_swap");
/* connect to SPI manager */
reorg_init();
/* swap relfilenode and dependencies for tables. */
values[0] = ObjectIdGetDatum(oid);
execute_with_args(SPI_OK_SELECT,
"SELECT X.reltoastrelid, TX.reltoastidxid, X.relowner,"
" Y.oid, Y.reltoastrelid, TY.reltoastidxid, Y.relowner"
" FROM pg_catalog.pg_class X LEFT JOIN pg_catalog.pg_class TX"
" ON X.reltoastrelid = TX.oid,"
" pg_catalog.pg_class Y LEFT JOIN pg_catalog.pg_class TY"
" ON Y.reltoastrelid = TY.oid"
" WHERE X.oid = $1"
" AND Y.oid = ('reorg.table_' || X.oid)::regclass",
1, argtypes, values, nulls);
tuptable = SPI_tuptable;
desc = tuptable->tupdesc;
records = SPI_processed;
if (records == 0)
elog(ERROR, "reorg_swap : no swap target");
tuple = tuptable->vals[0];
reltoastrelid1 = getoid(tuple, desc, 1);
reltoastidxid1 = getoid(tuple, desc, 2);
owner1 = getoid(tuple, desc, 3);
oid2 = getoid(tuple, desc, 4);
reltoastrelid2 = getoid(tuple, desc, 5);
reltoastidxid2 = getoid(tuple, desc, 6);
owner2 = getoid(tuple, desc, 7);
/* change owner of new relation to original owner */
if (owner1 != owner2)
{
ATExecChangeOwner(oid2, owner1, true, AccessExclusiveLock);
CommandCounterIncrement();
}
/* swap tables. */
swap_heap_or_index_files(oid, oid2);
CommandCounterIncrement();
/* swap indexes. */
values[0] = ObjectIdGetDatum(oid);
execute_with_args(SPI_OK_SELECT,
"SELECT X.oid, Y.oid"
" FROM pg_catalog.pg_index I,"
" pg_catalog.pg_class X,"
" pg_catalog.pg_class Y"
" WHERE I.indrelid = $1"
" AND I.indexrelid = X.oid"
" AND I.indisvalid"
" AND Y.oid = ('reorg.index_' || X.oid)::regclass",
1, argtypes, values, nulls);
tuptable = SPI_tuptable;
desc = tuptable->tupdesc;
records = SPI_processed;
for (i = 0; i < records; i++)
{
Oid idx1, idx2;
tuple = tuptable->vals[i];
idx1 = getoid(tuple, desc, 1);
idx2 = getoid(tuple, desc, 2);
swap_heap_or_index_files(idx1, idx2);
CommandCounterIncrement();
}
/* swap names for toast tables and toast indexes */
if (reltoastrelid1 == InvalidOid)
{
if (reltoastidxid1 != InvalidOid ||
reltoastrelid2 != InvalidOid ||
reltoastidxid2 != InvalidOid)
elog(ERROR, "reorg_swap : unexpected toast relations (T1=%u, I1=%u, T2=%u, I2=%u",
reltoastrelid1, reltoastidxid1, reltoastrelid2, reltoastidxid2);
/* do nothing */
}
else if (reltoastrelid2 == InvalidOid)
{
char name[NAMEDATALEN];
if (reltoastidxid1 == InvalidOid ||
reltoastidxid2 != InvalidOid)
elog(ERROR, "reorg_swap : unexpected toast relations (T1=%u, I1=%u, T2=%u, I2=%u",
reltoastrelid1, reltoastidxid1, reltoastrelid2, reltoastidxid2);
/* rename X to Y */
snprintf(name, NAMEDATALEN, "pg_toast_%u", oid2);
RENAME_REL(reltoastrelid1, name);
snprintf(name, NAMEDATALEN, "pg_toast_%u_index", oid2);
RENAME_REL(reltoastidxid1, name);
CommandCounterIncrement();
}
else if (reltoastrelid1 != InvalidOid)
{
char name[NAMEDATALEN];
int pid = getpid();
/* rename X to TEMP */
snprintf(name, NAMEDATALEN, "pg_toast_pid%d", pid);
RENAME_REL(reltoastrelid1, name);
snprintf(name, NAMEDATALEN, "pg_toast_pid%d_index", pid);
RENAME_REL(reltoastidxid1, name);
CommandCounterIncrement();
/* rename Y to X */
snprintf(name, NAMEDATALEN, "pg_toast_%u", oid);
RENAME_REL(reltoastrelid2, name);
snprintf(name, NAMEDATALEN, "pg_toast_%u_index", oid);
RENAME_REL(reltoastidxid2, name);
CommandCounterIncrement();
/* rename TEMP to Y */
snprintf(name, NAMEDATALEN, "pg_toast_%u", oid2);
RENAME_REL(reltoastrelid1, name);
snprintf(name, NAMEDATALEN, "pg_toast_%u_index", oid2);
RENAME_REL(reltoastidxid1, name);
CommandCounterIncrement();
}
/* drop reorg trigger */
execute_with_format(
SPI_OK_UTILITY,
"DROP TRIGGER IF EXISTS z_reorg_trigger ON %s.%s CASCADE",
nspname, relname);
SPI_finish();
PG_RETURN_VOID();
}
/*
* Re-organize one table.
*/
static void
reorg_one_table(const reorg_table *table, const char *orderby)
{
PGresult *res;
const char *params[1];
int num;
int i;
int num_waiting = 0;
char *vxid;
char buffer[12];
StringInfoData sql;
initStringInfo(&sql);
elog(DEBUG2, "---- reorg_one_table ----");
elog(DEBUG2, "target_name : %s", table->target_name);
elog(DEBUG2, "target_oid : %u", table->target_oid);
elog(DEBUG2, "target_toast : %u", table->target_toast);
elog(DEBUG2, "target_tidx : %u", table->target_tidx);
elog(DEBUG2, "pkid : %u", table->pkid);
elog(DEBUG2, "ckid : %u", table->ckid);
elog(DEBUG2, "create_pktype : %s", table->create_pktype);
elog(DEBUG2, "create_log : %s", table->create_log);
elog(DEBUG2, "create_trigger : %s", table->create_trigger);
elog(DEBUG2, "alter_table : %s", table->alter_table);
elog(DEBUG2, "create_table : %s", table->create_table);
elog(DEBUG2, "drop_columns : %s", table->drop_columns ? table->drop_columns : "(skipped)");
elog(DEBUG2, "delete_log : %s", table->delete_log);
elog(DEBUG2, "lock_table : %s", table->lock_table);
elog(DEBUG2, "sql_peek : %s", table->sql_peek);
elog(DEBUG2, "sql_insert : %s", table->sql_insert);
elog(DEBUG2, "sql_delete : %s", table->sql_delete);
elog(DEBUG2, "sql_update : %s", table->sql_update);
elog(DEBUG2, "sql_pop : %s", table->sql_pop);
/*
* 1. Setup workspaces and a trigger.
*/
elog(DEBUG2, "---- setup ----");
lock_exclusive(utoa(table->target_oid, buffer), table->lock_table);
/*
* Check z_reorg_trigger is the trigger executed at last so that
* other before triggers cannot modify triggered tuples.
*/
params[0] = utoa(table->target_oid, buffer);
res = execute("SELECT reorg.conflicted_triggers($1)", 1, params);
if (PQntuples(res) > 0)
ereport(ERROR,
(errcode(E_PG_COMMAND),
errmsg("trigger %s conflicted for %s",
PQgetvalue(res, 0, 0), table->target_name)));
PQclear(res);
command(table->create_pktype, 0, NULL);
command(table->create_log, 0, NULL);
command(table->create_trigger, 0, NULL);
command(table->alter_table, 0, NULL);
printfStringInfo(&sql, "SELECT reorg.disable_autovacuum('reorg.log_%u')", table->target_oid);
command(sql.data, 0, NULL);
command("COMMIT", 0, NULL);
/*
* Register the table to be dropped on error. We use pktype as
* an advisory lock. The registration should be done after
* the first command succeeds.
*/
pgut_atexit_push(&reorg_cleanup, (void *) table);
/*
* 2. Copy tuples into temp table.
*/
elog(DEBUG2, "---- copy tuples ----");
command("BEGIN ISOLATION LEVEL SERIALIZABLE", 0, NULL);
/* SET work_mem = maintenance_work_mem */
command("SELECT set_config('work_mem', current_setting('maintenance_work_mem'), true)", 0, NULL);
if (orderby && !orderby[0])
command("SET LOCAL synchronize_seqscans = off", 0, NULL);
res = execute(SQL_XID_SNAPSHOT, 0, NULL);
vxid = strdup(PQgetvalue(res, 0, 0));
PQclear(res);
command(table->delete_log, 0, NULL);
command(table->create_table, 0, NULL);
printfStringInfo(&sql, "SELECT reorg.disable_autovacuum('reorg.table_%u')", table->target_oid);
if (table->drop_columns)
command(table->drop_columns, 0, NULL);
command(sql.data, 0, NULL);
command("COMMIT", 0, NULL);
/*
* 3. Create indexes on temp table.
*/
elog(DEBUG2, "---- create indexes ----");
params[0] = utoa(table->target_oid, buffer);
res = execute("SELECT indexrelid,"
" reorg.reorg_indexdef(indexrelid, indrelid),"
" indisvalid,"
" pg_get_indexdef(indexrelid)"
" FROM pg_index WHERE indrelid = $1", 1, params);
num = PQntuples(res);
for (i = 0; i < num; i++)
{
reorg_index index;
int c = 0;
const char *isvalid;
const char *indexdef;
index.target_oid = getoid(res, i, c++);
index.create_index = getstr(res, i, c++);
isvalid = getstr(res, i, c++);
indexdef = getstr(res, i, c++);
if (isvalid && isvalid[0] == 'f') {
elog(WARNING, "skipping invalid index: %s", indexdef);
continue;
}
elog(DEBUG2, "[%d]", i);
elog(DEBUG2, "target_oid : %u", index.target_oid);
elog(DEBUG2, "create_index : %s", index.create_index);
/*
* NOTE: If we want to create multiple indexes in parallel,
* we need to call create_index in multiple connections.
*/
command(index.create_index, 0, NULL);
}
PQclear(res);
/*
* 4. Apply log to temp table until no tuples are left in the log
* and all of the old transactions are finished.
*/
for (;;)
{
num = apply_log(table, APPLY_COUNT);
if (num > 0)
continue; /* there might be still some tuples, repeat. */
/* old transactions still alive ? */
params[0] = vxid;
res = execute(SQL_XID_ALIVE, 1, params);
num = PQntuples(res);
if (num > 0)
{
/* Wait for old transactions.
* Only display the message below when the number of
* transactions we are waiting on changes (presumably,
* num_waiting should only go down), so as not to
* be too noisy.
*/
if (num != num_waiting)
{
elog(NOTICE, "Waiting for %d transactions to finish. First PID: %s", num, PQgetvalue(res, 0, 0));
num_waiting = num;
}
PQclear(res);
sleep(1);
continue;
}
else
{
/* All old transactions are finished;
* go to next step. */
PQclear(res);
break;
}
}
/*
* 5. Swap.
*/
elog(DEBUG2, "---- swap ----");
lock_exclusive(utoa(table->target_oid, buffer), table->lock_table);
apply_log(table, 0);
params[0] = utoa(table->target_oid, buffer);
command("SELECT reorg.reorg_swap($1)", 1, params);
command("COMMIT", 0, NULL);
/*
* 6. Drop.
*/
elog(DEBUG2, "---- drop ----");
command("BEGIN ISOLATION LEVEL READ COMMITTED", 0, NULL);
params[0] = utoa(table->target_oid, buffer);
command("SELECT reorg.reorg_drop($1)", 1, params);
command("COMMIT", 0, NULL);
pgut_atexit_pop(&reorg_cleanup, (void *) table);
free(vxid);
/*
* 7. Analyze.
* Note that cleanup hook has been already uninstalled here because analyze
* is not an important operation; No clean up even if failed.
*/
if (analyze)
{
elog(DEBUG2, "---- analyze ----");
command("BEGIN ISOLATION LEVEL READ COMMITTED", 0, NULL);
printfStringInfo(&sql, "ANALYZE %s", table->target_name);
command(sql.data, 0, NULL);
command("COMMIT", 0, NULL);
}
termStringInfo(&sql);
}
/*
* Call reorg_one_table for the target table or each table in a database.
*/
static bool
reorg_one_database(const char *orderby, const char *table)
{
bool ret = true;
PGresult *res;
int i;
int num;
StringInfoData sql;
initStringInfo(&sql);
reconnect(ERROR);
/* Disable statement timeout. */
command("SET statement_timeout = 0", 0, NULL);
/* Restrict search_path to system catalog. */
command("SET search_path = pg_catalog, pg_temp, public", 0, NULL);
/* To avoid annoying "create implicit ..." messages. */
command("SET client_min_messages = warning", 0, NULL);
/* acquire target tables */
appendStringInfoString(&sql, "SELECT * FROM reorg.tables WHERE ");
if (table)
{
appendStringInfoString(&sql, "relid = $1::regclass");
res = execute_elevel(sql.data, 1, &table, DEBUG2);
}
else
{
appendStringInfoString(&sql, "pkid IS NOT NULL");
if (!orderby)
appendStringInfoString(&sql, " AND ckid IS NOT NULL");
res = execute_elevel(sql.data, 0, NULL, DEBUG2);
}
if (PQresultStatus(res) != PGRES_TUPLES_OK)
{
if (sqlstate_equals(res, SQLSTATE_INVALID_SCHEMA_NAME))
{
/* Schema reorg does not exist. Skip the database. */
ret = false;
goto cleanup;
}
else
{
/* exit otherwise */
printf("%s", PQerrorMessage(connection));
PQclear(res);
exit(1);
}
}
num = PQntuples(res);
for (i = 0; i < num; i++)
{
reorg_table table;
const char *create_table;
const char *ckey;
int c = 0;
table.target_name = getstr(res, i, c++);
table.target_oid = getoid(res, i, c++);
table.target_toast = getoid(res, i, c++);
table.target_tidx = getoid(res, i, c++);
table.pkid = getoid(res, i, c++);
table.ckid = getoid(res, i, c++);
if (table.pkid == 0)
ereport(ERROR,
(errcode(E_PG_COMMAND),
errmsg("relation \"%s\" must have a primary key or not-null unique keys", table.target_name)));
table.create_pktype = getstr(res, i, c++);
table.create_log = getstr(res, i, c++);
table.create_trigger = getstr(res, i, c++);
table.alter_table = getstr(res, i, c++);
create_table = getstr(res, i, c++);
table.drop_columns = getstr(res, i, c++);
table.delete_log = getstr(res, i, c++);
table.lock_table = getstr(res, i, c++);
ckey = getstr(res, i, c++);
resetStringInfo(&sql);
if (!orderby)
{
/* CLUSTER mode */
if (ckey == NULL)
ereport(ERROR,
(errcode(E_PG_COMMAND),
errmsg("relation \"%s\" has no cluster key", table.target_name)));
appendStringInfo(&sql, "%s ORDER BY %s", create_table, ckey);
table.create_table = sql.data;
}
else if (!orderby[0])
{
/* VACUUM FULL mode */
table.create_table = create_table;
}
else
{
/* User specified ORDER BY */
appendStringInfo(&sql, "%s ORDER BY %s", create_table, orderby);
table.create_table = sql.data;
}
table.sql_peek = getstr(res, i, c++);
table.sql_insert = getstr(res, i, c++);
table.sql_delete = getstr(res, i, c++);
table.sql_update = getstr(res, i, c++);
table.sql_pop = getstr(res, i, c++);
reorg_one_table(&table, orderby);
}
cleanup:
PQclear(res);
disconnect();
termStringInfo(&sql);
return ret;
}
