This patch introduces an internal aggregate operator SELECT_SOME that
is automatically added to columns that are not in GROUP BY. It
"computes" some plausible value of the column (actually, last one
passed).
Along the way it fixes incorrect handling of HAVING being transferred
into WHERE, window function handling and a bit of other inconsistencies.
There were numerous memory leaks in plugin's code and associated code.
During typical run of MTR tests it leaked around 65 megabytes of
objects. As a result they may severely affect long-lived connections.
This patch fixes (almost) all leaks found in the plugin. The exceptions
are two leaks associated with SHOW CREATE TABLE columnstore_table and
getting information of columns of columnstore-handled table. These
should be fixed on the server side and work is on the way.
* MCOL-4234: improve GROUP BY and ORDER BY interaction (#3194)
This patch fixes the problem in MCOL-4234 and also generally improves
behavior of GROUP BY.
It does so by introducing a "dummy" aggregate and by wrapping columns
into it. This allows for columns that are not in GROUP BY to be used
more freely, for example, in SELECT * FROM tbl GROUP BY col - all
columns that are not "col" will be wrapped into an aggregate and query
will proceed to execution.
The dummy aggregate itself does nothing more than remember last value
passed into it.
There also an additional error message that tries to explain what types
of expressions can be wrapped into an aggregate.
* MCOL-5772: incorrect ORDER BY ordering for a columns not in GROUP BY (#3214)
When ORDER BY column is not in GROUP BY, is not an aggregate and there
is a SELECT column that is also not an aggregate, there was a problem:
ordering happened on the SELECTed column, not ORDERed one.
This patch fixes that particular problem and also performs some tidying
around newly added aggregate.
---------
Co-authored-by: Leonid Fedorov <79837786+mariadb-LeonidFedorov@users.noreply.github.com>
This patch improves handling of NULLs in textual fields in ColumnStore.
Previously empty strings were considered NULLs and it could be a problem
if data scheme allows for empty strings. It was also one of major
reasons of behavior difference between ColumnStore and other engines in
MariaDB family.
Also, this patch fixes some other bugs and incorrect behavior, for
example, incorrect comparison for "column <= ''" which evaluates to
constant True for all purposes before this patch.
These seem to have all fallen out of a recent Boost update to 1.81 which
dropped some internal includes. All of these uses within columnstore
relied on these transitive includes, so explicitly include what we need
to fix build.
Signed-off-by: Sam James <sam@gentoo.org>
* MCOL-5074 CASE with IN and aggregate asserts
gwip-scsp wasn't set and buildPredicateItem() was called which assumes it is set. Added code to set properly in this case
Part 1:
As part of MCOL-3776 to address synchronization issue while accessing
the fTimeZone member of the Func class, mutex locks were added to the
accessor and mutator methods. However, this slows down processing
of TIMESTAMP columns in PrimProc significantly as all threads across
all concurrently running queries would serialize on the mutex. This
is because PrimProc only has a single global object for the functor
class (class derived from Func in utils/funcexp/functor.h) for a given
function name. To fix this problem:
(1) We remove the fTimeZone as a member of the Func derived classes
(hence removing the mutexes) and instead use the fOperationType
member of the FunctionColumn class to propagate the timezone values
down to the individual functor processing functions such as
FunctionColumn::getStrVal(), FunctionColumn::getIntVal(), etc.
(2) To achieve (1), a timezone member is added to the
execplan::CalpontSystemCatalog::ColType class.
Part 2:
Several functors in the Funcexp code call dataconvert::gmtSecToMySQLTime()
and dataconvert::mySQLTimeToGmtSec() functions for conversion between seconds
since unix epoch and broken-down representation. These functions in turn call
the C library function localtime_r() which currently has a known bug of holding
a global lock via a call to __tz_convert. This significantly reduces performance
in multi-threaded applications where multiple threads concurrently call
localtime_r(). More details on the bug:
https://sourceware.org/bugzilla/show_bug.cgi?id=16145
This bug in localtime_r() caused processing of the Functors in PrimProc to
slowdown significantly since a query execution causes Functors code to be
processed in a multi-threaded manner.
As a fix, we remove the calls to localtime_r() from gmtSecToMySQLTime()
and mySQLTimeToGmtSec() by performing the timezone-to-offset conversion
(done in dataconvert::timeZoneToOffset()) during the execution plan
creation in the plugin. Note that localtime_r() is only called when the
time_zone system variable is set to "SYSTEM".
This fix also required changing the timezone type from a std::string to
a long across the system.
DML statements executed on the primary node in a ColumnStore
cluster do not need to be written to the primary's binlog. This
is due to ColumnStore's distributed storage architecture.
With this patch, we disable writing to binlog when a DML statement
(INSERT/DELETE/UPDATE/LDI/INSERT..SELECT) is performed on a ColumnStore
table. HANDLER::external_lock() calls are used to
1. Turn OFF the OPTION_BIN_LOG flag
2. Turn ON the OPTION_BIN_TMP_LOG_OFF flag
in THD::variables.option_bits during a WRITE lock call.
THD::variables.option_bits is restored back to the original state
during the UNLOCK call in HANDLER::external_lock().
Further, isDMLStatement() function is added to reduce code verbosity
to check if a given statement is a DML statement.
Note that with this patch, not writing to primary's binlog means
DML replication from a ColumnStore cluster to another ColumnStore
cluster or to another foreign engine will not work.
on a non-ColumnStore table does not work.
As part of MCOL-4617, we moved the in-to-exists predicate creation
and injection from the server into the engine. However, when query
with an IN Subquery contains a non-ColumnStore table, the server
still performs the in-to-exists predicate transformation for the
foreign engine table. This caused ColumnStore's execution plan to
contain incorrect WHERE predicates. As a fix, we call
mutate_optimizer_flags() for the WRITE lock, in addition to the READ
table lock. And in mutate_optimizer_flags(), we change the optimizer
flag from OPTIMIZER_SWITCH_IN_TO_EXISTS to OPTIMIZER_SWITCH_MATERIALIZATION.
After an AggreateColumn corresponding to SUM(1+1) is created,
it is pushed to the list:
gwi.count_asterisk_list.push_back(ac)
Later, in getSelectPlan(), the expression SUM(1+1) was erroneously
treated as a constant:
if (!hasNonSupportItem && !nonConstFunc(ifp) && !(parseInfo & AF_BIT) && tmpVec.size() == 0)
{
srcp.reset(buildReturnedColumn(item, gwi, gwi.fatalParseError));
This code freed the original AggregateColumn and replaced to a ConstantColumn.
But gwi.count_asterisk_list still pointer to the freed AggregateColumn().
The expression SUM(1+1) was treated as a constant because tmpVec
was empty due to a bug in this code:
// special handling for count(*). This should not be treated as constant.
if (isp->argument_count() == 1 &&
( sfitempp[0]->type() == Item::CONST_ITEM &&
(sfitempp[0]->cmp_type() == INT_RESULT ||
sfitempp[0]->cmp_type() == STRING_RESULT ||
sfitempp[0]->cmp_type() == REAL_RESULT ||
sfitempp[0]->cmp_type() == DECIMAL_RESULT)
)
)
{
field_vec.push_back((Item_field*)item); //dummy
Notice, it handles only aggregate functions with explicit literals
passed as an argument, while it does not handle constant expressions
such as 1+1.
Fix:
- Adding new classes ConstantColumnNull, ConstantColumnString,
ConstantColumnNum, ConstantColumnUInt, ConstantColumnSInt,
ConstantColumnReal, ValStrStdString, to reuse the code easier.
- Moving a part of the code from the case branch handling CONST_ITEM
in buildReturnedColumn() into a new function
newConstantColumnNotNullUsingValNativeNoTz(). This
makes the code easier to read and to reuse in the future.
- Adding a new function newConstantColumnMaybeNullFromValStrNoTz().
Removing dulplicate code from !!!four!!! places, using the new
function instead.
- Adding a function isSupportedAggregateWithOneConstArg() to
properly catch all constant expressions. Using the new function parse_item()
in the code commented as "special handling for count(*)".
Now it pushes all constant expressions to field_vec, not only
explicit literals.
- Moving a part of the code from buildAggregateColumn()
to a helper function processAggregateColumnConstArg().
Using processAggregateColumnConstArg() in the CONST_ITEM
and NULL_ITEM branches.
- Adding a new branch in buildReturnedColumn() handling FUNC_ITEM.
If a function has constant arguments, a ConstantColumn() is
immediately created, without going to
buildArithmeticColumn()/buildFunctionColumn().
- Reusing isSupportedAggregateWithOneConstArg()
and processAggregateColumnConstArg() in buildAggregateColumn().
A new branch catches aggregate function has only one constant argument
and immediately creates a single ConstantColumn without
traversing to the argument sub-components.
cross-engine join with a ColumnStore table errors out.
ColumnStore cannot directly update a foreign table. We detect whether
a multi-table UPDATE operation is performed on a foreign table, if so,
do not create the select_handler and let the server execute the UPDATE
operation instead.
This feature allows a query execution to fallback to the server,
in case query execution using the select_handler (SH) fails. In case
of fallback, a warning message containing the original reason for
query failure using SH is generated.
To accomplish this task, SH execution is moved to an earlier step when
we create the SH in create_columnstore_select_handler(), instead of the
previous call to SH execution in ha_columnstore_select_handler::init_scan().
This requires some pre-requisite steps that occur in the server in
JOIN::optimize() and JOIN::exec() to be performed before starting SH execution.
In addition, missing test cases from MCOL-424 are also added to the MTR suite,
and the corresponding fix using disable_indices_for_CEJ() is reverted back
since the original fix now appears to be redundant.
This is a subtask of MCOL-4525 Implement select_handler=AUTO.
Server performs outer join to inner join conversion using simplify_joins()
in sql/sql_select.cc, by updating the TABLE_LIST::outer_join variable.
In order to perform this conversion, permanent changes are made in some
cases to the SELECT_LEX::JOIN::conds and/or TABLE_LIST::on_expr.
This is undesirable for MCOL-4525 which will attemp to fallback and execute
the query inside the server, in case the query execution fails in ColumnStore
using the select_handler.
For a query such as:
SELECT * FROM t1 LEFT JOIN t2 ON expr1 LEFT JOIN t3 ON expr2
In some cases, server can update the original SELECT_LEX::JOIN::conds
and/or TABLE_LIST::on_expr and create new Item_cond_and objects
(e.g. with 2 Item's expr1 and expr2 in Item_cond_and::list).
Instead of making changes to the original query structs, we use
gp_walk_info::tableOnExprList and gp_walk_info::condList. 2 Item's,
expr1 and expr2, in the condList, mean Item_cond_and(expr1, expr2), and
hence avoid permanent transformations to the SELECT_LEX.
We also define a new member variable
ha_columnstore_select_handler::tableOuterJoinMap
which saves the original TABLE_LIST::outer_join values before they are
updated. This member variable will be used later on to restore to the original
state of TABLE_LIST::outer_join in case of a query fallback to server execution.
The original simplify_joins() implementation in the server also performs a
flattening of the JOIN nest, however we don't perform this operation in
convertOuterJoinToInnerJoin() since it is not required for ColumnStore.
Main theme of the patch is to fix joins processing in the plugin
code. We now use SELECT_LEX::top_join_list and process the nested
joins recursively, instead of SELECT_LEX::table_list struct which
we earlier used to build the join filters. The earlier approach
did not process certain nested join ON expressions, causing certain
queries to incorrectly error out such as that described in MCOL-4680.
In addition, some legacy code is also removed.
We earlier leveraged the server functionality provided by
Item_in_subselect::create_in_to_exists_cond and
Item_in_subselect::inject_in_to_exists_cond
to create and inject the in-to-exists predicate into an IN
subquery's JOIN struct. With this patch, we leave the IN subquery's
JOIN unaltered and instead directly perform this predicate creation
and injection into ColumnStore's select execution plan.
For CHAR/VARCHAR/TEXT fields, the buffer size of a field represents
the field size in bytes, which can be bigger than the field size in
number of characters, for multi-byte character sets such as utf8,
utf8mb4 etc. The buffer also contains a byte length prefix which can be
up to 65532 bytes for a VARCHAR field, and much higher for a TEXT
field (we process a maximum byte length for a TEXT field which fits in
4 bytes, which is 2^32 - 1 = 4GB!).
There is also special processing for a TEXT field defined with a default
length like so:
CREATE TABLE cs1 (a TEXT CHARACTER SET utf8)
Here, the byte length is a fixed 65535, irrespective of the character
set used. This is different from a case such as:
CREATE TABLE cs1 (a TEXT(65535) CHARACTER SET utf8), where the byte length
for the field will be 65535*3.
The change for MCOL-4264 erroneously added the "lock_type" member
to cal_connection_info, which is shared between multiple tables.
So some tables that were opened for write erroneously identified
themselves as read only.
Moving the member to ha_mcs instead.
Problem:
When processing cross-engine queries like:
update cstab1 set a=100 where a not in (select a from innotab1 where a=11);
delete from innotab1 where a not in (select a from cstab1 where a=1);
the ColumnStore plugin erroneously executed the whole query inside
ColumnStore.
Fix:
- Adding a new member cal_connection_info::lock_type and setting it
inside ha_mcs_impl_external_lock() to the value passed in the parameter
"lock_type".
- Adding a method cal_connection_info::isReadOnly() to test
if the last table lock made in ha_mcs_impl_external_lock()
for done for reading.
- Adding a new condition checking cal_connection_info::isReadOnly() inside
ha_mcs_impl_rnd_init(). If the current table was locked last time for reading,
then doUpdateDelete() should not be executed.
For certain queries, such as:
update cs1 set i = 41 where i = 42 or (i is null and 42 is null);
the SELECT_LEX.where does not contain the required where conditions.
Server sends the where conditions in the call to cond_push(), so
we are storing them in a handler data member, condStack, and later
push them down to getSelectPlan() for UPDATES/DELETEs.
Original SH implementation sends the result set back to the client
thus it can't be used in INSERT..SELECT, SELECT INTO OUTFILE,CREATE
TABLE AS SELECT etc.
CLX-77 feature has been backported into MDB to enable SH to run
query part of the mentioned queries.
Direct update/delete executed doUpdateDelete as well as the regular
execution route for doUpdateDelete.
This patch only executes doUpdateDelete the first time and direct
update/delete collects the counts.
Disabled 4th if block in buildOuterJoin to handle non-optimized MDB query
structures.
Broke getSelectPlan into pieces: processFrom, processWhere.
MCOL-3593 UNION processing depends on two flags isUnion that comes as
arg of getSelectPlan and unionSel that is a local variable in
getSelectPlan. Modularization of getSelectPlan broke the mechanizm.
This patch is supposed to partially fix it.
MCOL-3593 Removed unused if condition from buildOuterJoin that allows
unsupported construct subquery in ON expression.
Fixed an improper if condition that ignors tableMap entries w/o condition
in external_lock thus external_lock doesn't clean up when the query
finishes.
Fixed wrong logging for queries processed in tableMode. Now rnd_init
properly sends queryText down to ExeMgr to be logged.
MCOL-3593 Unused attribute FromSubQuery::fFromSub was removed.
getSelectPlan has been modularized into: setExecutionParams,
processFrom, processWhere. SELECT, HAVING, GROUP BY, ORDER BY
still lives in getSelectPlan.
Copied optimization function simplify_joins_ into our pushdown
code to provide the plugin code with some rewrites from MDB it
expects.
The columnstore_processing_handlers_fallback session variable
has been removed thus CS can't fallback from SH to partial
execution paths, e.g. DH, GBH or plugin API.
MCOL-3602 Moved MDB optimizer rewrites into a separate file.
Add SELECT_LEX::optimize_unflattened_subqueries() call to fix IN
into EXISTS rewrite for semi-JOINs with subqueries.
disable_indices_for_CEJ() add index related hints to disable
index access methods in Cross Engine Joins.
create_SH() now flattens JOIN that has both physical tables and
views. This fixes most of views related tests in the regression.
This patch fixes:
MCOL-3557 - Row Based Replication events to ColumnStore tables will no
longer cause MariaDB to crash, it will error instead.
MCOL-3556 - Remove the Columnstore.xml variable to turn on ColumnStore
tables applying replication events and instead make it a system variable
that can be set in my.cnf called "columnstore_replication_slave". This
allows it to be set per-UM.
