Details
-
Bug
-
Status: Open (View Workflow)
-
Major
-
Resolution: Unresolved
-
10.5
-
None
-
None
Description
When run with --view-protocol, these tests
spider/bg.ha
|
spider/bg.ha_part
|
spider.ha
|
spider/handler.ha
|
spider.ha_part
|
spider/handler.ha_part
|
spider/bugfix.mdev_30649
|
spider/bugfix.mdev_30727
|
fail with
mysqltest: At line 398: query '$MASTER_1_COPY_TABLES_2_1' failed: 12507: This UDF can't execute if other tables are opened 'thd->derived_tables'=0x52000002c128
|
|
|
The result from queries just before the failure was:
|
< snip >
|
# [... 19 lines elided]
|
SELECT spider_copy_tables('ta_l', '0', '1');
|
If we remove the corresponding check:
modified storage/spider/spd_copy_tables.cc
|
@@ -753,16 +753,20 @@ long long spider_copy_tables_body(
|
ER_SPIDER_UDF_CANT_USE_IF_OPEN_TABLE_STR_WITH_NUM, MYF(0),
|
"thd->handler_tables_hash.records",
|
(longlong) thd->handler_tables_hash.records);
|
- } else if (thd->derived_tables != 0)
|
- {
|
- my_printf_error(ER_SPIDER_UDF_CANT_USE_IF_OPEN_TABLE_NUM,
|
- ER_SPIDER_UDF_CANT_USE_IF_OPEN_TABLE_STR_WITH_PTR, MYF(0),
|
- "thd->derived_tables", thd->derived_tables); |
Then we get tripped at the next check of thd->lock.
If we remove both checks:
modified storage/spider/spd_copy_tables.cc
|
@@ -737,8 +737,10 @@ long long spider_copy_tables_body(
|
if (
|
thd->open_tables != 0 ||
|
thd->handler_tables_hash.records != 0 ||
|
- thd->derived_tables != 0 ||
|
- thd->lock != 0 ||
|
thd->locked_tables_list.locked_tables() ||
|
thd->locked_tables_mode != LTM_NONE
|
) {
|
@@ -753,16 +755,20 @@ long long spider_copy_tables_body(
|
ER_SPIDER_UDF_CANT_USE_IF_OPEN_TABLE_STR_WITH_NUM, MYF(0),
|
"thd->handler_tables_hash.records",
|
(longlong) thd->handler_tables_hash.records);
|
- } else if (thd->derived_tables != 0)
|
- {
|
- my_printf_error(ER_SPIDER_UDF_CANT_USE_IF_OPEN_TABLE_NUM,
|
- ER_SPIDER_UDF_CANT_USE_IF_OPEN_TABLE_STR_WITH_PTR, MYF(0),
|
- "thd->derived_tables", thd->derived_tables);
|
- } else if (thd->lock != 0)
|
- {
|
- my_printf_error(ER_SPIDER_UDF_CANT_USE_IF_OPEN_TABLE_NUM,
|
- ER_SPIDER_UDF_CANT_USE_IF_OPEN_TABLE_STR_WITH_PTR, MYF(0),
|
- "thd->lock", thd->lock); |
then the errors disappear, the tests pass normally but crashes in view protocol
Attachments
Issue Links
- is part of
-
-
- Open
-
Upon further analysis, I think the error reporting makes sense and we should disable view protocol for calls to this udf in spider tests.
More specifically, consider the statement SELECT spider_copy_tables('t', '0', '1').
Without view protocol, the query is simply SELECT spider_copy_tables('t', '0', '1'), and lock_tables() in the initial open-and-lock is a no-op because there are no tables involved (the server layer has no idea the string 't' refers to a table).
// lock_tables:DBUG_RETURN(0);With view protocol, the query is SELECT * FROM mysqltest_tmp_v, where mysqltest_tmp_v is a view created using the actual SELECT of the udf. So the initial lock_tables does lock tables:
// lock_tables:flags)))DBUG_RETURN(TRUE);With or without view protocol, during the execution of the udf, to copy from one table at the data node to another, spider needs to open the spider table t to access its fields so that it can construct queries to select from the source table and insert into the destination table.
// spider_copy_tables_body:insert_ct->append_insert_str(SPIDER_DB_INSERT_IGNORE) ||insert_ct->append_into_str() ||insert_ct->append_table_name(0) ||insert_ct->append_open_paren_str() ||insert_ct->append_table_columns(table_share) ||insert_ct->append_values_str()) {my_error(ER_OUT_OF_RESOURCES, MYF(0), HA_ERR_OUT_OF_MEM);}With view protocol, the attempt to lock t trips at an assertion failure:
// lock_tables:{BTW this is not a problem with other spider udfs such as spider_direct_sql, which simply passes the query to the data node without the need to lock any tables at the spider node.