Author: Vlad Lesin
Revision Date: 2014-03-20 15:39:27 UTC

Facebook readahead.

Author: Vlad Lesin
Revision Date: 2014-11-05 22:13:34 UTC

Consider statement as 'write' if restrictive enough
(MDL_SHARED_WRITE and above) MDL lock is acquired.

Author: Vlad Lesin
Revision Date: 2014-10-29 21:50:36 UTC

If a statement would be cancelled when --read-only option is on consider
this statement as 'write'.

Author: Vlad Lesin
Revision Date: 2014-09-11 11:20:10 UTC

Bug #1322218 fix.

Merged from 5.1.

The difference between this and 5.1 fix is in preventing floating point error
in the case if "users" table is not created.

Author: Vlad Lesin
Revision Date: 2014-09-08 10:23:19 UTC

Bug #1322218 fix.

In the case if mysqld returns error don't continue mysql_install_db script
execution.

The second case when floating point error happens due to dividing by zero when
the number of records in "users" table is null or there is no such table is not
fixed as 5.1 does not contain such code.

Author: Vlad Lesin
Revision Date: 2014-09-09 12:21:53 UTC

Implement this
https://blueprints.launchpad.net/percona-server/+spec/binary-answer-5.6
blueprint.

The task can be splitted on server and client parts.

Server-side.

If we look into mysqld_stmt_prepare() function we can see the code like this:

Protocol *save_protocol= thd->protocol;
thd->protocol= &thd->protocol_binary;
stmt->prepare(...);
thd->protocol= save_protocol;

The THD class contains this members:

Protocol *protocol; // Current protocol
Protocol_text protocol_text; // Normal protocol
Protocol_binary protocol_binary; // Binary protocol

So for server-side is enough to add new command COM_STMT_EXECUTE_IMMEDIATE which
would do absolutely the same things as COM_QUERY does but with replacing
thd->protocol with &thd->protocol_binary before processing and restore protocol
pointer after processing.

Client side.

Standard C API has already have all necessary infrastructure to parse binary
protocol. So the task is to use this infrastructure. The idea is to add new
function mysql_stmt_execute_immediate() which would send
COM_STMT_EXECUTE_IMMEDIATE to server, initialize MYSQL_STMT object to use it
for parsing result with standard functions for parsing the result of
"execute prepared statement". As well mysql_stmt_bind_immediate() is necessary
to initialize certain MYSQL_STMT fields which are usually initialized inside of
mysql_stmt_prepare() and to invoke standard mysql_stmt_bind_result() function
to describe the result data.

There are two functions to execute prepared and non-prepared queries
correspondingly: mysql_stmt_execute() and mysql_real_query(). Both functions
use the same function cli_read_query_result() to get information about fields
count, their names. So this part of result parsing is the same. But rows parsing
is absolutely different.

For non-prepared query mysql_fetch_row()->read_one_row() calls just set pointers
in array of rows to the certain values in read packet. As all values are in the
text format we have the array of pointers to null-terminated strings as the
output of the function.

But when binary data are parsed in mysql_stmt_fetch()->stmt_fetch_row() the
functions use information about field types from MYSQL_STMT::bind array.
For each element of this array
mysql_stmt_bind_result()->setup_one_fetch_function() sets the pointer to the
function which parses binary data of corresponding type to the buffer. For
example fetch_result_int32() is set up for MYSQL_TYPE_LONG data type.

So we have to develop new function mysql_stmt_execute_immediate() which would
mix mysql_real_query() and mysql_stmt_execute().

Use tests/stmt_execute_immediate.cc as the source of information about the
sequence of C API calls to send query and receive and parse binary answer.

stmt_execute_immediate.cc could be the part of mysql_client_test.c but the
customer disabled this test in his tree.

The new parameter --binary-protocol is added to mysqltest to allow communication
between client and server with binary protocol without "prepare" using
mysql_stmt_execute_immedate() function to send query.

Author: Vlad Lesin
Revision Date: 2014-08-12 10:01:57 UTC

Fix for bug #1197042.

Author: Vlad Lesin
Revision Date: 2014-05-15 11:04:53 UTC

Add possibility to change the number of connections randomly for
thread_pool plugin.

The new options is added:
--thread-pool-change-delay
  If this option is not null then the number of connections will be
  changed from 0 to thread-pool-threads-count with a period of
  thread-pool-change-delay ms.

Author: Vlad Lesin
Revision Date: 2014-04-22 12:37:15 UTC

Add mysqldump support for logical read ahead

Summary:
Adds options to mysqldump:
 --lra-size=X
 --lra-sleep=X
 --lra-n-node-recs-before-sleep=X

These just inject SET statements to set these session variables.

The original implementation is here:
https://github.com/facebook/mysql-5.6/commit/f69a4ea522bce24e4cdcc7696d5fad29587cf87a

Author: Vlad Lesin
Revision Date: 2014-04-22 10:04:46 UTC

When the session variable innodb_lra_size is set to N, we issue async
read requests for the next M logical pages where the total size of the M
pages on disk is N megabytes. The max allowed value of innodb_lra_size
is is 16384 which corresponds to prefetching 16GB of data. We may choose
to use smaller values in production.

The original implementation can be found here:
https://github.com/facebook/mysql-5.6/commit/f8e361952612d00979f7cf744f487e48b15cb5a6

This implementation does not contain code for flashcahe.

Author: Vlad Lesin
Revision Date: 2014-02-02 20:31:50 UTC

Bug #1268735 fixing is merged from 5.5.

The difference between 5.5 and this patches is the following:

      if (is_network_error(mysql_errno(mysql)))
      {
        mi->report(WARNING_LEVEL, mysql_errno(mysql),
                   "SET @master_heartbeat_period to master failed with error: %s",
                   mysql_error(mysql));
        mysql_free_result(mysql_store_result(mysql));
        goto network_err;
      }
- else if (!check_io_slave_killed(mi->io_thd, mi, NULL)) {
+ else {
        errmsg= "The slave I/O thread stops because SET @master_heartbeat_period "
          "on master failed.";
        err_code= ER_SLAVE_FATAL_ERROR;
        sprintf(err_buff, "%s Error: %s", errmsg, mysql_error(mysql));
        mysql_free_result(mysql_store_result(mysql));
        goto err;
      }

The initial code checked if the slave thread was killed in the case if there
was error in mysql_real_query(). To save this logic non-network error in
mysql_real_query() is treated as fatal error only if the slave thread was
not killed in 5.5. But as 5.6 upstream fix (see upstream revno: 2661.723.1)
does not contain this check it was decided do not include it in PS-5.6.

Author: Vlad Lesin
Revision Date: 2014-02-02 18:08:08 UTC

The fix for bug #1268735.

Process the network error during setting @master_heartbeat_period in
get_master_version_and_clock() correctly.

Author: Vlad Lesin
Revision Date: 2013-12-16 15:08:07 UTC

Port "query response time" feature
(http://www.percona.com/doc/percona-server/5.5/diagnostics/response_time_distribution.html)
as audit plugin. The original task description and related code is here:
https://mariadb.atlassian.net/browse/MDEV-4568 .

Author: Vlad Lesin
Revision Date: 2013-11-13 23:49:54 UTC

Bug #1227240 fix.

my.cnf can contain innodb_log_arch_dir option for server. xtrabackup can
use servers my.cnf. As innodb_log_arch_dir option can be used only for backup
preparing xtrabackup exits with error message in the case of backup with
innodb_log_arch_dir option. Such behavior does not allow to make backup if
default my.cnf contains innodb_log_arch_dir.

This fix ignores --innodb_log_arch_dir and --to-archived-lsn and outputs warning
if --prepare is not set.

Author: Vlad Lesin
Revision Date: 2013-08-13 11:37:28 UTC

Fast fix for bug #1156368.

The limit for max captured packet is currently defined in the code.
This limit checking adds some insurance that pcap packet was parsed
right. I think more flexible approach would be in letting this parameter
be a command line option.

Author: Vlad Lesin
Revision Date: 2013-08-07 06:03:17 UTC

Merged with mysql-5.5.33. Revision 4397 of lp:mysql-server/5.5 contains
partial fix for bug 1049871 / CVE-2012-4414. The fix is for the following
issue: "A long SET expression in LOAD DATA is incorrectly truncated when
written to the binary log."

When SET clauses are specified as a part of LOAD DATA statement, these SET
clause user command strings need to be stored as it is inorder to reconstruct
the original user command. These strings are stored as a part of SET clause item
tree's top most Item node's name itself, but Item::name can be of MAX_ALIAS_NAME
size.

Percona's fix just removes limit for Item::name but upstream fix uses new
String list to store the SET expression values.

This branch contains Percona's fix because upstream's fix cause rpl_mdev382
test failure, i.e. the last argument in SET argiments list is repeated twice
in binlog. This is an example from rpl_mdev382 result file diff:

-LOAD DATA LOCAL INFILE '<name>' INTO TABLE `t``1` FIELDS TERMINATED BY ',' ENCLOSED BY '\'' ESCAPED BY '\\' LINES TERMINATED BY '\n' (`a``1`, @`b```) SET `b``2`= @`b```, `c``3`= concat('|', "b""a'z", "!")
+LOAD DATA LOCAL INFILE '<name>' INTO TABLE `t``1` FIELDS TERMINATED BY ',' ENCLOSED BY '\'' ESCAPED BY '\\' LINES TERMINATED BY '\n' (`a``1`, @`b```) SET `b``2`= @`b```, `c``3` = concat('|', "b""a'z", "!"), `c``3`= concat('|', "b""a'z", "!")

Author: Vlad Lesin
Revision Date: 2013-07-31 12:43:26 UTC

Use PS-5.6.12 with static openssl build to test archived logs.

Author: Vlad Lesin
Revision Date: 2013-05-14 13:24:56 UTC

Bug #1179193 fix.

Remove from innodb56.patch fix which disabled invoking
of fts_optimize_start_shutdown() in the case if srv_apply_log_only
is set.

Author: Vlad Lesin
Revision Date: 2013-05-14 13:20:39 UTC

Bug #1179193 fix.

Remove from innodb56.patch fix which disabled invoking
of fts_optimize_start_shutdown() in the case if srv_apply_log_only
is set.

Author: Vlad Lesin
Revision Date: 2013-04-04 02:48:47 UTC

Archived logs applying draft.

Author: Vlad Lesin
Revision Date: 2013-02-24 23:04:05 UTC

As the number of open tables is increased, table lookup
(testing if a table is already open) and (in particular)
the case when a table is not open, became increasingly more
expensive.

The problem was caused by the open table lookup mechanism,
which was based on traversing a linked list comparing the
file names.

As the list was replaced by a hash table, the lookup
time dropped significantly when used on systems with
a large number of open tables.

The original patch can be found here:
http://lists.mysql.com/commits/121507

Author: Vlad Lesin
Revision Date: 2013-02-24 22:52:02 UTC

As the number of open tables is increased, table lookup
(testing if a table is already open) and (in particular)
the case when a table is not open, became increasingly more
expensive.

The problem was caused by the open table lookup mechanism,
which was based on traversing a linked list comparing the
file names.

As the list was replaced by a hash table, the lookup
time dropped significantly when used on systems with
a large number of open tables.

The original patch can be found here:
http://lists.mysql.com/commits/121507

Author: Vlad Lesin
Revision Date: 2013-02-24 21:28:18 UTC

As the number of open tables is increased, table lookup
(testing if a table is already open) and (in particular)
the case when a table is not open, became increasingly more
expensive.

The problem was caused by the open table lookup mechanism,
which was based on traversing a linked list comparing the
file names.

As the list was replaced by a hash table, the lookup
time dropped significantly when used on systems with
a large number of open tables.

The original patch can be found here:
http://lists.mysql.com/commits/121507

Author: Vlad Lesin
Revision Date: 2013-02-01 16:26:58 UTC

Bug #1111215 fix.

There is no need to concatenate multiline queries as mysql C API supports
for multiline statements.

Author: Vlad Lesin
Revision Date: 2013-01-18 15:38:46 UTC

Digest tests fixes.

Author: Vlad Lesin
Revision Date: 2013-01-09 13:37:53 UTC

New system variable "SQL_TIMEOUT" allows to set limit for query execution time.
If this limit is exceeded the query is killed. SQL_TIMEOUT can be global or
session variable.

The timeout expiration is checked on each db-thread killed state check. If timeout
expires the function that returns thread killed status will return the value which
corresponds to "killed query" status.

Author: Vlad Lesin
Revision Date: 2013-01-07 16:05:50 UTC

Check timeout expiration on each db-thread killed status check.

Author: Vlad Lesin
Revision Date: 2012-09-21 15:07:48 UTC

Fix for bug #1049871

Author: Vlad Lesin
Revision Date: 2013-01-01 20:11:04 UTC

Fix for bug #1038940.

As the number of open tables is increased, table lookup
(testing if a table is already open) and (in particular)
the case when a table is not open, became increasingly more
expensive.

The problem was caused by the open table lookup mechanism,
which was based on traversing a linked list comparing the
file names.

As the list was replaced by a hash table, the lookup
time dropped significantly when used on systems with
a large number of open tables.

The original patch can be found here:
http://lists.mysql.com/commits/121507

Author: Vlad Lesin
Revision Date: 2012-12-05 14:07:27 UTC

1) Apply logs only to certain LSN if corresponfing option is set
2) Skip already applied logs (segfault)

Author: Vlad Lesin
Revision Date: 2012-12-05 12:55:01 UTC

Process list thread-safe issues possible fixes.

Author: Vlad Lesin
Revision Date: 2012-11-14 14:29:43 UTC

Filter queries plugin is added. It is supported only be query-log input plugin.

Author: Vlad Lesin
Revision Date: 2012-11-08 15:24:56 UTC

Holes detection is added.

Author: Vlad Lesin
Revision Date: 2012-09-13 07:37:47 UTC

The bug #801989 fix.

The main idea of this fix is to return "true" on success and "false" otherwise
from os_file_set_nocache(). That allows to process O_DIRECT flag setting failures.

If setting O_DIRECT failes on log files the srv_unix_file_flush_method is set
to SRV_UNIX_O_DIRECT value to do fsync() on log files. If O_DIRECT fails on
other files the error is ignored to do buffered I/O on filesystems that do not
support direct I/O(for example tmpfs).

Author: Vlad Lesin
Revision Date: 2012-09-13 07:22:01 UTC

The bug #801989 fix.

The main idea of this fix is to return "true" on success and "false" otherwise
from os_file_set_nocache(). That allows to process O_DIRECT flag setting failures.

If setting O_DIRECT failes on log files the srv_unix_file_flush_method is set
to SRV_UNIX_O_DIRECT value to do fsync() on log files. If O_DIRECT fails on
other files the error is ignored to do buffered I/O on filesystems that do not
support direct I/O(for example tmpfs).

Author: Vlad Lesin
Revision Date: 2012-07-31 09:21:34 UTC

Virgin 5.0.33

Author: Vlad Lesin
Revision Date: 2012-07-31 09:00:46 UTC

Use rbtree instead of linked list to store flush list.

The original patch is here:

url: bzr+ssh://bazaar.launchpad.net/+branch/mysql-server/5.5
revno: 0.5.80

Author: Vlad Lesin
Revision Date: 2012-07-27 06:53:50 UTC

Innodb fast recovery patch

Author: Vlad Lesin
Revision Date: 2012-06-19 01:01:29 UTC

Add macros for libpcap.

Author: Vlad Lesin
Revision Date: 2012-06-07 19:25:10 UTC

refactoring

Author: Vlad Lesin
Revision Date: 2012-06-09 12:02:15 UTC

Slow log impruvements -
https://blueprints.launchpad.net/percona-server/+spec/log-slow-sp-statements

Author: Vlad Lesin
Revision Date: 2012-05-29 13:11:09 UTC

DRIZZLE_CON_OFFLINE option for libdrizzle. This option means don't read
data from network if there is truncated packet during parsing.

Author: Vlad Lesin
Revision Date: 2012-05-16 11:44:17 UTC

Read input from file.

Author: Vlad Lesin
Revision Date: 2012-05-14 18:32:59 UTC

Fix main.percona_query_response_time-stored and innodb.innodb_bug56680 tests
failure.

Author: Vlad Lesin
Revision Date: 2012-05-02 16:58:46 UTC

Merged with lp:~stewart/percona-server/bug758788

Author: Vlad Lesin
Revision Date: 2012-04-26 09:38:19 UTC

Get SQL_TIMEOUT=X variable to make queries automatically be killed
when they are taking too long time.
Example:
SET SQL_TIMEOUT=10;
SELECT SLEEP(20);

The query will be killed in 10 seconds.

Author: Vlad Lesin
Revision Date: 2012-04-26 09:23:44 UTC

Per query variable settings allow to assign some values to variables only for
the certain query after execution of which the variables restore their previous
values.
Example:
SET STATEMENT myisam_sort_buffer_size=40000, myisam_repair_threads=2
    FOR SELECT * FROM t;

Author: Vlad Lesin
Revision Date: 2012-02-22 11:49:22 UTC

Merged with trunk

Author: Vlad Lesin
Revision Date: 2012-02-06 17:46:36 UTC

Simplify sql_timeout test.

Author: Vlad Lesin
Revision Date: 2011-10-28 15:58:37 UTC

functional tests

Author: Vlad Lesin
Revision Date: 2011-07-13 13:02:42 UTC

Use colon to pass a number of pages to preload from the end of tablespace