MDEV-22088 S3 partitioning support
All ALTER PARTITION commands should now work on S3 tables except
REBUILD PARTITION
TRUNCATE PARTITION
REORGANIZE PARTITION
In addition, PARTIONED S3 TABLES can also be replicated.
This is achived by storing the partition tables .frm and .par file on S3
for partitioned shared (S3) tables.
The discovery methods are enchanced by allowing engines that supports
discovery to also support of the partitioned tables .frm and .par file
Things in more detail
- The .frm and .par files of partitioned tables are stored in S3 and kept
in sync.
- Added hton callback create_partitioning_metadata to inform handler
that metadata for a partitoned file has changed
- Added back handler::discover_check_version() to be able to check if
a table's or a part table's definition has changed.
- Added handler::check_if_updates_are_ignored(). Needed for partitioning.
- Renamed rebind() -> rebind_psi(), as it was before.
- Changed CHF_xxx hadnler flags to an enum
- Changed some checks from using table->file->ht to use
table->file->partition_ht() to get discovery to work with partitioning.
- If TABLE_SHARE::init_from_binary_frm_image() fails, ensure that we
don't leave any .frm or .par files around.
- Fixed that writefrm() doesn't leave unusable .frm files around
- Appended extension to path for writefrm() to be able to reuse to function
for creating .par files.
- Added DBUG_PUSH("") to a a few functions that caused a lot of not
critical tracing.
Prototype change:
- virtual ha_rows records_in_range(uint inx, key_range *min_key,
- key_range *max_key)
+ virtual ha_rows records_in_range(uint inx, const key_range *min_key,
+ const key_range *max_key,
+ page_range *res)
The handler can ignore the page_range parameter. In the case the handler
updates the parameter, the optimizer can deduce the following:
- If previous range's last key is on the same block as next range's first
key
- If the current key range is in one block
- We can also assume that the first and last block read are cached!
This can be used for a better calculation of IO seeks when we
estimate the cost of a range index scan.
The parameter is fully implemented for MyISAM, Aria and InnoDB.
A separate patch will update handler::multi_range_read_info_const() to
take the benefits of this change and also remove the double
records_in_range() calls that are not anymore needed.
This was done to both simplify the code and also to be easier to handle
storage engines that are clustered on some other index than the primary
key.
As pk_is_clustering_key() and is_clustering_key now are using only
index_flags, these where removed from all storage engines.
ha_partition: Remove redundant 'virtual' keywords and add
missing 'override'.
FIXME: handler::table_type() is not declared virtual, yet ha_partition
and ha_sequence are seemingly trying to override it.
- Flag ALTER_STORED_COLUMN_TYPE set while doing varchar extension
for partition table. Basically all partition supports
can_be_converted_by_engine() then it should be set to
ALTER_COLUMN_TYPE_CHANGE_BY_ENGINE.
Partition table with the AUTO_INCREMENT column we ahve to check if the
max value is properly loaded. So we need to open all tables in INSERT
PARTITION statement if necessary. Also we need to check if some
tables are pruned away and not count the max autoincrement in this case.
MDEV-19486 and one more similar bug appeared because handler::write_row() interface
welcomes to modify buffer by storage engine. But callers are not ready for that
thus bugs are possible in future.
handler::write_row():
handler::ha_write_row(): make argument const
Reason for the change was that ha_notify_table_changed() was done
after table open when .frm had been replaced, which caused failure
in engines that checks on open if .frm matches the engines table
definition.
Other changes:
- Remove not needed open/close call at end of inline alter table.
Some test that depended on the table beeing in the table cache after
ALTER TABLE had to be updated.
make live checksum to be returned in handler::info(),
and slow table-scan checksum to be calculated in handler::checksum().
part of
MDEV-16249 CHECKSUM TABLE for a spider table is not parallel and saves all data in memory in the spider head by default
Fix partitioning for trx_id-versioned tables.
`partition by hash`, `range` and others now work.
`partition by system_time` is forbidden.
Currently we cannot use row_start and row_end in `partition by`, because
insertion of versioned field is done by engine's handler, as well as
row_start/row_end's value set up, which is a transaction id -- so it's
also forbidden.
The drawback is that it's now impossible to use `partition by key()`
without parameters for such tables, because it references row_start and
row_end implicitly.
* add handler::vers_can_native()
* drop Table_scope_and_contents_source_st::vers_native()
* drop partition_element::find_engine_flag as unused
* forbid versioning partitioning for trx_id as not supported
* adopt vers tests for trx_id partitioning
* forbid any row_end referencing in `partition by` clauses,
including implicit `by key()`
This patch contains a full implementation of the optimization
that allows to use in-memory rowid / primary filters built for range
conditions over indexes. In many cases usage of such filters reduce
the number of disk seeks spent for fetching table rows.
In this implementation the choice of what possible filter to be applied
(if any) is made purely on cost-based considerations.
This implementation re-achitectured the partial implementation of
the feature pushed by Galina Shalygina in the commit
8d5a11122c32f4d9eb87536886c6e893377bdd07.
Besides this patch contains a better implementation of the generic
handler function handler::multi_range_read_info_const() that
takes into account gaps between ranges when calculating the cost of
range index scans. It also contains some corrections of the
implementation of the handler function records_in_range() for MyISAM.
This patch supports the feature for InnoDB and MyISAM.
When using buffered sort in `UPDATE`, keyread is used. In this case,
`TABLE::update_virtual_field` should be aborted, but it actually isn't,
because it is called not with a top-level handler, but with the one that
is actually going to access the disk. Here the problemm is issued with
partitioning, so the solution is to recursively mark for keyread all the
underlying partition handlers.
* ha_partition: update keyread state for child partitions
Closes#800
We don't have many bits left, no need to add another InnoDB-specific flag.
Instead, we say that HA_REQUIRE_PRIMARY_KEY does not apply to SEQUENCE.
Meaning, if the engine declares HA_CAN_TABLES_WITHOUT_ROLLBACK (required
for SEQUENCE) it *must* support tables without a primary key.
main.derived_cond_pushdown: Move all 10.3 tests to the end,
trim trailing white space, and add an "End of 10.3 tests" marker.
Add --sorted_result to tests where the ordering is not deterministic.
main.win_percentile: Add --sorted_result to tests where the
ordering is no longer deterministic.
Fixed by adding table flag HA_WANTS_PRIMARY_KEY, which is like
HA_REQUIRE_PRIMARY_KEY but tells SQL upper layer that the storage engine
internally can handle tables without primary keys (for example for
sequences or trough user variables)
When using buffered sort in `UPDATE`, keyread is used. In this case,
`TABLE::update_virtual_field` should be aborted, but it actually isn't,
because it is called not with a top-level handler, but with the one that
is actually going to access the disk. Here the problemm is issued with
partitioning, so the solution is to recursively mark for keyread all the
underlying partition handlers.
* ha_partition: update keyread state for child partitions
Closes#800
The problem occurs in 10.2 and earlier releases of MariaDB Server because the
Partition Engine was not pushing the engine conditions to the underlying
storage engine of each partition. This caused Spider to return the first 5
rows in the table with the data provided by the customer. 2 of the 5 rows
did not qualify the WHERE clause, so they were removed from the result set by
the server.
To fix the problem, I have back-ported support for engine condition pushdown
in the Partition Engine from MariaDB Server 10.3.
Author:
Jacob Mathew.
Reviewer:
Kentoku Shiba.
Cherry-Picked:
Commit eb2ca3d on branch bb-10.2-MDEV-16912
The problem occurs in 10.2 and earlier releases of MariaDB Server because the
Partition Engine was not pushing the engine conditions to the underlying
storage engine of each partition. This caused Spider to return the first 5
rows in the table with the data provided by the customer. 2 of the 5 rows
did not qualify the WHERE clause, so they were removed from the result set by
the server.
To fix the problem, I have back-ported support for engine condition pushdown
in the Partition Engine from MariaDB Server 10.3.
Author:
Jacob Mathew.
Reviewer:
Kentoku Shiba.
The problem occurred because the Spider node was incorrectly handling
timestamp values sent to and received from the data nodes.
The problem has been corrected as follows:
- Added logic to set and maintain the UTC time zone on the data nodes.
To prevent timestamp ambiguity, it is necessary for the data nodes to use
a time zone such as UTC which does not have daylight savings time.
- Removed the spider_sync_time_zone configuration variable, which did not
solve the problem and which interfered with the solution.
- Added logic to convert to the UTC time zone all timestamp values sent to
and received from the data nodes. This is done for both unique and
non-unique timestamp columns. It is done for WHERE clauses, applying to
SELECT, UPDATE and DELETE statements, and for UPDATE columns.
- Disabled Spider's use of direct update when any of the columns to update is
a timestamp column. This is necessary to prevent false duplicate key value
errors.
- Added a new test spider.timestamp to thoroughly test Spider's handling of
timestamp values.
Author:
Jacob Mathew.
Reviewer:
Kentoku Shiba.
Cherry-Picked:
Commit 97cc9d3 on branch bb-10.3-MDEV-16246
Main reason was to make it easier to print the above structures in
a debugger. Additional benefits is that I was able to use same
defines for both structures, which simplifes some code.
Most of the code is just removing Alter_info:: and Alter_inplace_info::
from alter table flags.
Following renames was done:
HA_ALTER_FLAGS -> alter_table_operations
CHANGE_CREATE_OPTION -> ALTER_CHANGE_CREATE_OPTION
Alter_info::ADD_INDEX -> ALTER_ADD_INDEX
DROP_INDEX -> ALTER_DROP_INDEX
ADD_UNIQUE_INDEX -> ALTER_ADD_UNIQUE_INDEX
DROP_UNIQUE_INDEx -> ALTER_DROP_UNIQUE_INDEX
ADD_PK_INDEX -> ALTER_ADD_PK_INDEX
DROP_PK_INDEX -> ALTER_DROP_PK_INDEX
Alter_info:ALTER_ADD_COLUMN -> ALTER_PARSE_ADD_COLUMN
Alter_info:ALTER_DROP_COLUMN -> ALTER_PARSE_DROP_COLUMN
Alter_inplace_info::ADD_INDEX -> ALTER_ADD_NON_UNIQUE_NON_PRIM_INDEX
Alter_inplace_info::DROP_INDEX -> ALTER_DROP_NON_UNIQUE_NON_PRIM_INDEX
Other things:
- Added typedef alter_table_operatons for alter table flags
- DROP CHECK CONSTRAINT can now be done online
- Added checks for Aria tables in alter_table_online.test
- alter_table_flags now takes an ulonglong as argument.
- Don't support online operations if checksum option is used.
- sql_lex.cc doesn't add ALTER_ADD_INDEX if index is not created
