/**
* @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();
}
/**
* @fn Datum reorg_trigger(PG_FUNCTION_ARGS)
* @brief Insert a operation log into log-table.
*
* reorg_trigger(sql)
*
* @param sql SQL to insert a operation log into log-table.
*/
Datum
reorg_trigger(PG_FUNCTION_ARGS)
{
TriggerData *trigdata = (TriggerData *) fcinfo->context;
TupleDesc desc;
HeapTuple tuple;
Datum values[2];
bool nulls[2] = { 0, 0 };
Oid argtypes[2];
const char *sql;
/* authority check */
must_be_superuser("reorg_trigger");
/* make sure it's called as a trigger at all */
if (!CALLED_AS_TRIGGER(fcinfo) ||
!TRIGGER_FIRED_BEFORE(trigdata->tg_event) ||
!TRIGGER_FIRED_FOR_ROW(trigdata->tg_event) ||
trigdata->tg_trigger->tgnargs != 1)
elog(ERROR, "reorg_trigger: invalid trigger call");
/* retrieve parameters */
sql = trigdata->tg_trigger->tgargs[0];
desc = RelationGetDescr(trigdata->tg_relation);
argtypes[0] = argtypes[1] = trigdata->tg_relation->rd_rel->reltype;
/* connect to SPI manager */
reorg_init();
if (TRIGGER_FIRED_BY_INSERT(trigdata->tg_event))
{
/* INSERT: (NULL, newtup) */
tuple = trigdata->tg_trigtuple;
nulls[0] = true;
values[1] = copy_tuple(tuple, desc);
}
else if (TRIGGER_FIRED_BY_DELETE(trigdata->tg_event))
{
/* DELETE: (oldtup, NULL) */
tuple = trigdata->tg_trigtuple;
values[0] = copy_tuple(tuple, desc);
nulls[1] = true;
}
else
{
/* UPDATE: (oldtup, newtup) */
tuple = trigdata->tg_newtuple;
values[0] = copy_tuple(trigdata->tg_trigtuple, desc);
values[1] = copy_tuple(tuple, desc);
}
/* INSERT INTO reorg.log VALUES ($1, $2) */
execute_with_args(SPI_OK_INSERT, sql, 2, argtypes, values, nulls);
SPI_finish();
PG_RETURN_POINTER(tuple);
}
/**
* @fn Datum reorg_drop(PG_FUNCTION_ARGS)
* @brief Delete temporarily objects.
*
* reorg_drop(oid, relname)
*
* @param oid Oid of target table.
* @retval None.
*/
Datum
reorg_drop(PG_FUNCTION_ARGS)
{
Oid oid = PG_GETARG_OID(0);
const char *relname = get_quoted_relname(oid);
const char *nspname = get_quoted_nspname(oid);
/* authority check */
must_be_superuser("reorg_drop");
/* connect to SPI manager */
reorg_init();
/*
* drop reorg trigger: We have already dropped the trigger in normal
* cases, but it can be left on error.
*/
execute_with_format(
SPI_OK_UTILITY,
"DROP TRIGGER IF EXISTS z_reorg_trigger ON %s.%s CASCADE",
nspname, relname);
#if PG_VERSION_NUM < 80400
/* delete autovacuum settings */
execute_with_format(
SPI_OK_DELETE,
"DELETE FROM pg_catalog.pg_autovacuum v"
" USING pg_class c, pg_namespace n"
" WHERE relname IN ('log_%u', 'table_%u')"
" AND n.nspname = 'reorg'"
" AND c.relnamespace = n.oid"
" AND v.vacrelid = c.oid",
oid, oid);
#endif
/* drop log table */
execute_with_format(
SPI_OK_UTILITY,
"DROP TABLE IF EXISTS reorg.log_%u CASCADE",
oid);
/* drop temp table */
execute_with_format(
SPI_OK_UTILITY,
"DROP TABLE IF EXISTS reorg.table_%u CASCADE",
oid);
/* drop type for log table */
execute_with_format(
SPI_OK_UTILITY,
"DROP TYPE IF EXISTS reorg.pk_%u CASCADE",
oid);
SPI_finish();
PG_RETURN_VOID();
}
/**
* @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();
}
/**
* @fn Datum reorg_apply(PG_FUNCTION_ARGS)
* @brief Apply operations in log table into temp table.
*
* reorg_apply(sql_peek, sql_insert, sql_delete, sql_update, sql_pop, count)
*
* @param sql_peek SQL to pop tuple from log table.
* @param sql_insert SQL to insert into temp table.
* @param sql_delete SQL to delete from temp table.
* @param sql_update SQL to update temp table.
* @param sql_pop SQL to delete tuple from log table.
* @param count Max number of operations, or no count iff <=0.
* @retval Number of performed operations.
*/
Datum
reorg_apply(PG_FUNCTION_ARGS)
{
#define DEFAULT_PEEK_COUNT 1000
const char *sql_peek = PG_GETARG_CSTRING(0);
const char *sql_insert = PG_GETARG_CSTRING(1);
const char *sql_delete = PG_GETARG_CSTRING(2);
const char *sql_update = PG_GETARG_CSTRING(3);
const char *sql_pop = PG_GETARG_CSTRING(4);
int32 count = PG_GETARG_INT32(5);
SPIPlanPtr plan_peek = NULL;
SPIPlanPtr plan_insert = NULL;
SPIPlanPtr plan_delete = NULL;
SPIPlanPtr plan_update = NULL;
SPIPlanPtr plan_pop = NULL;
uint32 n, i;
Oid argtypes_peek[1] = { INT4OID };
Datum values_peek[1];
bool nulls_peek[1] = { 0 };
/* authority check */
must_be_superuser("reorg_apply");
/* connect to SPI manager */
reorg_init();
/* peek tuple in log */
plan_peek = reorg_prepare(sql_peek, 1, argtypes_peek);
for (n = 0;;)
{
int ntuples;
SPITupleTable *tuptable;
TupleDesc desc;
Oid argtypes[3]; /* id, pk, row */
Datum values[3]; /* id, pk, row */
bool nulls[3]; /* id, pk, row */
/* peek tuple in log */
if (count == 0)
values_peek[0] = Int32GetDatum(DEFAULT_PEEK_COUNT);
else
values_peek[0] = Int32GetDatum(Min(count - n, DEFAULT_PEEK_COUNT));
execute_plan(SPI_OK_SELECT, plan_peek, values_peek, nulls_peek);
if (SPI_processed <= 0)
break;
/* copy tuptable because we will call other sqls. */
ntuples = SPI_processed;
tuptable = SPI_tuptable;
desc = tuptable->tupdesc;
argtypes[0] = SPI_gettypeid(desc, 1); /* id */
argtypes[1] = SPI_gettypeid(desc, 2); /* pk */
argtypes[2] = SPI_gettypeid(desc, 3); /* row */
for (i = 0; i < ntuples; i++, n++)
{
HeapTuple tuple;
tuple = tuptable->vals[i];
values[0] = SPI_getbinval(tuple, desc, 1, &nulls[0]);
values[1] = SPI_getbinval(tuple, desc, 2, &nulls[1]);
values[2] = SPI_getbinval(tuple, desc, 3, &nulls[2]);
if (nulls[1])
{
/* INSERT */
if (plan_insert == NULL)
plan_insert = reorg_prepare(sql_insert, 1, &argtypes[2]);
execute_plan(SPI_OK_INSERT, plan_insert, &values[2], &nulls[2]);
}
else if (nulls[2])
{
/* DELETE */
if (plan_delete == NULL)
plan_delete = reorg_prepare(sql_delete, 1, &argtypes[1]);
execute_plan(SPI_OK_DELETE, plan_delete, &values[1], &nulls[1]);
}
else
{
/* UPDATE */
if (plan_update == NULL)
plan_update = reorg_prepare(sql_update, 2, &argtypes[1]);
execute_plan(SPI_OK_UPDATE, plan_update, &values[1], &nulls[1]);
}
}
/* delete tuple in log */
if (plan_pop == NULL)
plan_pop = reorg_prepare(sql_pop, 1, argtypes);
execute_plan(SPI_OK_DELETE, plan_pop, values, nulls);
SPI_freetuptable(tuptable);
}
SPI_finish();
PG_RETURN_INT32(n);
}
/**
* @fn Datum repack_drop(PG_FUNCTION_ARGS)
* @brief Delete temporarily objects.
*
* repack_drop(oid, relname)
*
* @param oid Oid of target table.
* @retval None.
*/
Datum
repack_drop(PG_FUNCTION_ARGS)
{
Oid oid = PG_GETARG_OID(0);
int numobj = PG_GETARG_INT32(1);
const char *relname = get_quoted_relname(oid);
const char *nspname = get_quoted_nspname(oid);
if (!(relname && nspname))
{
elog(ERROR, "table name not found for OID %u", oid);
PG_RETURN_VOID();
}
/* authority check */
must_be_superuser("repack_drop");
/* connect to SPI manager */
repack_init();
/*
* To prevent concurrent lockers of the repack target table from causing
* deadlocks, take an exclusive lock on it. Consider that the following
* commands take exclusive lock on tables log_xxx and the target table
* itself when deleting the repack_trigger on it, while concurrent
* updaters require row exclusive lock on the target table and in
* addition, on the log_xxx table, because of the trigger.
*
* Consider how a deadlock could occur - if the DROP TABLE repack.log_%u
* gets a lock on log_%u table before a concurrent updater could get it
* but after the updater has obtained a lock on the target table, the
* subsequent DROP TRIGGER ... ON target-table would report a deadlock as
* it finds itself waiting for a lock on target-table held by the updater,
* which in turn, is waiting for lock on log_%u table.
*
* Fixes deadlock mentioned in the Github issue #55.
*
* Skip the lock if we are not going to do anything.
* Otherwise, if repack gets accidentally run twice for the same table
* at the same time, the second repack, in order to perform
* a pointless cleanup, has to wait until the first one completes.
* This adds an ACCESS EXCLUSIVE lock request into the queue
* making the table effectively inaccessible for any other backend.
*/
if (numobj > 0)
{
execute_with_format(
SPI_OK_UTILITY,
"LOCK TABLE %s.%s IN ACCESS EXCLUSIVE MODE",
nspname, relname);
}
/* drop log table: must be done before dropping the pk type,
* since the log table is dependent on the pk type. (That's
* why we check numobj > 1 here.)
*/
if (numobj > 1)
{
execute_with_format(
SPI_OK_UTILITY,
"DROP TABLE IF EXISTS repack.log_%u CASCADE",
oid);
--numobj;
}
/* drop type for pk type */
if (numobj > 0)
{
execute_with_format(
SPI_OK_UTILITY,
"DROP TYPE IF EXISTS repack.pk_%u",
oid);
--numobj;
}
/*
* drop repack trigger: We have already dropped the trigger in normal
* cases, but it can be left on error.
*/
if (numobj > 0)
{
execute_with_format(
SPI_OK_UTILITY,
"DROP TRIGGER IF EXISTS repack_trigger ON %s.%s CASCADE",
nspname, relname);
--numobj;
}
/* drop temp table */
if (numobj > 0)
{
execute_with_format(
SPI_OK_UTILITY,
"DROP TABLE IF EXISTS repack.table_%u CASCADE",
oid);
--numobj;
}
SPI_finish();
PG_RETURN_VOID();
}
/**
* @fn Datum repack_apply(PG_FUNCTION_ARGS)
* @brief Apply operations in log table into temp table.
*
* repack_apply(sql_peek, sql_insert, sql_delete, sql_update, sql_pop, count)
*
* @param sql_peek SQL to pop tuple from log table.
* @param sql_insert SQL to insert into temp table.
* @param sql_delete SQL to delete from temp table.
* @param sql_update SQL to update temp table.
* @param sql_pop SQL to bulk-delete tuples from log table.
* @param count Max number of operations, or no count iff <=0.
* @retval Number of performed operations.
*/
Datum
repack_apply(PG_FUNCTION_ARGS)
{
#define DEFAULT_PEEK_COUNT 1000
const char *sql_peek = PG_GETARG_CSTRING(0);
const char *sql_insert = PG_GETARG_CSTRING(1);
const char *sql_delete = PG_GETARG_CSTRING(2);
const char *sql_update = PG_GETARG_CSTRING(3);
/* sql_pop, the fourth arg, will be used in the loop below */
int32 count = PG_GETARG_INT32(5);
SPIPlanPtr plan_peek = NULL;
SPIPlanPtr plan_insert = NULL;
SPIPlanPtr plan_delete = NULL;
SPIPlanPtr plan_update = NULL;
uint32 n, i;
Oid argtypes_peek[1] = { INT4OID };
Datum values_peek[1];
const char nulls_peek[1] = { 0 };
StringInfoData sql_pop;
initStringInfo(&sql_pop);
/* authority check */
must_be_superuser("repack_apply");
/* connect to SPI manager */
repack_init();
/* peek tuple in log */
plan_peek = repack_prepare(sql_peek, 1, argtypes_peek);
for (n = 0;;)
{
int ntuples;
SPITupleTable *tuptable;
TupleDesc desc;
Oid argtypes[3]; /* id, pk, row */
Datum values[3]; /* id, pk, row */
bool nulls[3]; /* id, pk, row */
/* peek tuple in log */
if (count <= 0)
values_peek[0] = Int32GetDatum(DEFAULT_PEEK_COUNT);
else
values_peek[0] = Int32GetDatum(Min(count - n, DEFAULT_PEEK_COUNT));
execute_plan(SPI_OK_SELECT, plan_peek, values_peek, nulls_peek);
if (SPI_processed <= 0)
break;
/* copy tuptable because we will call other sqls. */
ntuples = SPI_processed;
tuptable = SPI_tuptable;
desc = tuptable->tupdesc;
argtypes[0] = SPI_gettypeid(desc, 1); /* id */
argtypes[1] = SPI_gettypeid(desc, 2); /* pk */
argtypes[2] = SPI_gettypeid(desc, 3); /* row */
resetStringInfo(&sql_pop);
appendStringInfoString(&sql_pop, PG_GETARG_CSTRING(4));
for (i = 0; i < ntuples; i++, n++)
{
HeapTuple tuple;
char *pkid;
tuple = tuptable->vals[i];
values[0] = SPI_getbinval(tuple, desc, 1, &nulls[0]);
values[1] = SPI_getbinval(tuple, desc, 2, &nulls[1]);
values[2] = SPI_getbinval(tuple, desc, 3, &nulls[2]);
pkid = SPI_getvalue(tuple, desc, 1);
Assert(pkid != NULL);
if (nulls[1])
{
/* INSERT */
if (plan_insert == NULL)
plan_insert = repack_prepare(sql_insert, 1, &argtypes[2]);
execute_plan(SPI_OK_INSERT, plan_insert, &values[2], (nulls[2] ? "n" : " "));
}
else if (nulls[2])
{
/* DELETE */
if (plan_delete == NULL)
plan_delete = repack_prepare(sql_delete, 1, &argtypes[1]);
execute_plan(SPI_OK_DELETE, plan_delete, &values[1], (nulls[1] ? "n" : " "));
}
else
{
/* UPDATE */
if (plan_update == NULL)
plan_update = repack_prepare(sql_update, 2, &argtypes[1]);
execute_plan(SPI_OK_UPDATE, plan_update, &values[1], (nulls[1] ? "n" : " "));
}
/* Add the primary key ID of each row from the log
* table we have processed so far to this
* DELETE ... IN (...) query string, so we
* can delete all the rows we have processed at-once.
*/
if (i == 0)
appendStringInfoString(&sql_pop, pkid);
else
appendStringInfo(&sql_pop, ",%s", pkid);
pfree(pkid);
}
/* i must be > 0 (and hence we must have some rows to delete)
* since SPI_processed > 0
*/
Assert(i > 0);
appendStringInfoString(&sql_pop, ");");
/* Bulk delete of processed rows from the log table */
execute(SPI_OK_DELETE, sql_pop.data);
SPI_freetuptable(tuptable);
}
SPI_finish();
PG_RETURN_INT32(n);
}