Commit dedf125d authored by unknown's avatar unknown

Merge nusphere@work.mysql.com:/home/bk/mysql

into nslinux.bedford.progress.com:/usr1/mikef/MYSQL/there


BitKeeper/etc/logging_ok:
  auto-union
configure.in:
  Auto merged
Docs/manual.texi:
  Auto merged
sql/share/english/errmsg.txt:
  Auto merged
parents 97d52615 9d523813
......@@ -15,6 +15,10 @@ $make -k clean || true
aclocal && autoheader && aclocal && automake && autoconf
(cd bdb/dist && sh s_all)
(cd innobase && aclocal && autoheader && aclocal && automake && autoconf)
if [ -d gemini ]
then
(cd gemini && aclocal && autoheader && aclocal && automake && autoconf)
fi
CFLAGS=\"$cflags\" CXX=gcc CXXFLAGS=\"$cxxflags\" $configure"
......
mwagner@evoq.mwagner.org
sasha@mysql.sashanet.com
heikki@donna.mysql.fi
miguel@linux.local
monty@tik.mysql.fi
mikef@nslinux.bedford.progress.com
monty@donna.mysql.fi
monty@tik.mysql.fi
mwagner@evoq.mwagner.org
sasha@mysql.sashanet.com
......@@ -528,10 +528,25 @@ BDB or Berkeley_DB Tables
GEMINI Tables
* GEMINI overview::
* GEMINI start::
* GEMINI features::
* GEMINI TODO::
* GEMINI Overview::
* Using GEMINI Tables::
GEMINI Overview
* GEMINI Features::
* GEMINI Concepts::
* GEMINI Limitations::
Using GEMINI Tables
* Startup Options::
* Creating GEMINI Tables::
* Backing Up GEMINI Tables::
* Restoring GEMINI Tables::
* Using Auto_Increment Columns With GEMINI Tables::
* Performance Considerations::
* Sample Configurations::
* When To Use GEMINI Tables::
InnoDB Tables
......@@ -10096,7 +10111,7 @@ If you are using BDB (Berkeley DB) tables, you should familiarize
yourself with the different BDB specific startup options. @xref{BDB start}.
If you are using Gemini tables, refer to the Gemini-specific startup options.
@xref{GEMINI start}.
@xref{Using GEMINI Tables}.
If you are using InnoDB tables, refer to the InnoDB-specific startup
options. @xref{InnoDB start}.
......@@ -18845,7 +18860,7 @@ When you insert a value of @code{NULL} (recommended) or @code{0} into an
If you delete the row containing the maximum value for an
@code{AUTO_INCREMENT} column, the value will be reused with an
@code{ISAM}, @code{BDB} or @code{INNODB} table but not with a
@code{ISAM}, @code{GEMINI}, @code{BDB} or @code{INNODB} table but not with a
@code{MyISAM} table. If you delete all rows in the table with
@code{DELETE FROM table_name} (without a @code{WHERE}) in
@code{AUTOCOMMIT} mode, the sequence starts over for both table types.
......@@ -24535,87 +24550,849 @@ not in @code{auto_commit} mode, until this problem is fixed (the fix is
not trivial).
@end itemize
@cindex tables, @code{GEMINI}
@cindex GEMINI tables
@node GEMINI, InnoDB, BDB, Table types
@section GEMINI Tables
@cindex GEMINI tables, overview
@menu
* GEMINI overview::
* GEMINI start::
* GEMINI features::
* GEMINI TODO::
* GEMINI Overview::
* Using GEMINI Tables::
@end menu
@node GEMINI overview, GEMINI start, GEMINI, GEMINI
@subsection Overview of GEMINI tables
@node GEMINI Overview, Using GEMINI Tables, GEMINI, GEMINI
@subsection GEMINI Overview
@code{GEMINI} is a transaction-safe table handler for @strong{MySQL}. It
provides row-level locking, robust transaction support and reliable
crash recovery. It is targeted for databases that need to handle heavy
multi-user updates typical of transaction processing applications while
still providing excellent performance for read-intensive operations. The
@code{GEMINI} table type is developed and supported by NuSphere
Corporation (see @url{http://www.nusphere.com}).
@code{GEMINI} provides full ACID transaction properties (Atomic,
Consistent, Independent, and Durable) with a programming model that
includes support for statement atomicity and all four standard isolation
levels (Read Uncommitted, Read Committed, Repeatable Read, and
Serializable) defined in the SQL standard.
The @code{GEMINI} tables support row-level and table-level locking to
increase concurrency in applications and allow reading of tables without
locking for maximum concurrency in a heavy update environment. The
transaction, locking, and recovery mechanisms are tightly integrated to
eliminate unnecessary administration overhead.
In general, if @code{GEMINI} tables are selected for an application, it
is recommended that all tables updated in the application be
@code{GEMINI} tables to provide well-defined system behavior. If
non-@code{GEMINI} tables are mixed into the application then, ACID
transaction properties cannot be maintained. While there are clearly
cases where mixing table types is appropriate, it should always be done
with careful consideration of the impact on transaction consistency and
recoverability needs of the application and underlying database.
The @code{GEMINI} table type is derived from a successful commercial
database and uses the storage kernel technology tightly integrated with
@strong{MySQL} server. The basic @code{GEMINI} technology is in use by
millions of users worldwide in production environments today. This
maturity allows @code{GEMINI} tables to provide a solution for those
users who require transaction-based behavior as part of their
applications.
The @code{GEMINI} table type is developed and supported by NuSphere Corporation
(@uref{http://www.nusphere.com}). It features row-level locking, transaction
support (@code{COMMIT} and @code{ROLLBACK}), and automatic crash recovery.
The @code{GEMINI} table handler supports a configurable data cache that
allows a significant portion of any database to be maintained in memory
while still allowing durable updates.
@code{GEMINI} tables will be included in some future @strong{MySQL} 3.23.X
source distribution.
@cindex GEMINI tables, features
@menu
* GEMINI Features::
* GEMINI Concepts::
* GEMINI Limitations::
@end menu
@node GEMINI start, GEMINI features, GEMINI overview, GEMINI
@subsection GEMINI startup options
@node GEMINI Features, GEMINI Concepts, GEMINI Overview, GEMINI Overview
@subsubsection GEMINI Features
If you are running with @code{AUTOCOMMIT=0} then your changes in @code{GEMINI}
tables will not be updated until you execute @code{COMMIT}. Instead of commit
you can execute @code{ROLLBACK} to forget your changes. @xref{COMMIT}.
The following summarizes the major features provided by @code{GEMINI}
tables.
If you are running with @code{AUTOCOMMIT=1} (the default), your changes
will be committed immediately. You can start an extended transaction with
the @code{BEGIN WORK} SQL command, after which your changes will not be
committed until you execute @code{COMMIT} (or decide to @code{ROLLBACK}
the changes).
@itemize @bullet
@item
Supports all optimization statistics used by the @strong{MySQL} optimizer
including table cardinality, index range estimates and multi-component
selectivity to insure optimal query performance.
The following options to @code{mysqld} can be used to change the behavior of
GEMINI tables:
@item
Maintains exact cardinality information for each table so @code{SELECT
COUNT(*) FROM} table-name always returns an answer immediately.
@item
Supports index-only queries; when index data is sufficient to resolve a
query no record data is read (for non character types).
@item
@code{GEMINI} uses block based I/O for better performance. There is no
performance penalty for using @code{VARCHAR} fields. The maximum record size is
currently 32K.
@item
The number of rows in a single @code{GEMINI} table can be 4 quintillion
(full use of 64 bits).
@item
Individual tables can be as large as 16 petabytes.
@item
Locking is done at a record or row level rather than at table level
unless table locks are explicitly requested. When a row is inserted into
a table, other rows can be updated, inserted or deleted without waiting
for the inserted row to be committed.
@item
Provides durable transactions backed by a crash recovery mechanism that
returns the database to a known consistent state in the event of an
unexpected failure.
@item
Support for all isolation levels and statement atomicity defined in the
SQL standard.
@item
Reliable Master Replication; the master database can survive system
failure and recover all committed transactions.
@end itemize
@cindex GEMINI tables, concepts
@node GEMINI Concepts, GEMINI Limitations, GEMINI Features, GEMINI Overview
@subsubsection GEMINI Concepts
This section highlights some of the important concepts behind
@code{GEMINI} and the @code{GEMINI} programming model, including:
@itemize @bullet
@item
ACID Transactions
@item
Transaction COMMIT/ROLLBACK
@item
Statement Atomicity
@item
Recovery
@item
Isolation Levels
@item
Row-Level Locking
@end itemize
These features are described below.
@cindex GEMINI tables, ACID transactions
@noindent
@strong{ACID Transactions}
ACID in the context of transactions is an acronym which stands for
@emph{Atomicity}, @emph{Consistency}, @emph{Isolation}, @emph{Durability}.
@multitable @columnfractions .25 .75
@item @sc{Attribute} @tab @sc{Description}
@item
@strong{Atomicity}
@tab A transaction allows for the grouping of one or more changes to
tables and rows in the database to form an atomic or indivisible
operation. That is, either all of the changes occur or none of them
do. If for any reason the transaction cannot be completed, everything
this transaction changed can be restored to the state it was in prior to
the start of the transaction via a rollback operation.
@item
@strong{Consistency}
@tab
Transactions always operate on a consistent view of the data and when
they end always leave the data in a consistent state. Data may be said to
be consistent as long as it conforms to a set of invariants, such as no
two rows in the customer table have the same customer ID and all orders
have an associated customer row. While a transaction executes, these
invariants may be violated, but no other transaction will be allowed to
see these inconsistencies, and all such inconsistencies will have been
eliminated by the time the transaction ends.
@item
@strong{Isolation}
@tab To a given transaction, it should appear as though it is running
all by itself on the database. The effects of concurrently running
transactions are invisible to this transaction, and the effects of this
transaction are invisible to others until the transaction is committed.
@item
@strong{Durability}
@tab Once a transaction is committed, its effects are guaranteed to
persist even in the event of subsequent system failures. Until the
transaction commits, not only are any changes made by that transaction
not durable, but are guaranteed not to persist in the face of a system
failures, as crash recovery will rollback their effects.
@end multitable
@cindex GEMINI tables, COMMIT/ROLLBACK
@noindent
@strong{Transaction COMMIT/ROLLBACK}
As stated above, a transaction is a group of work being done to
data. Unless otherwise directed, @strong{MySQL} considers each statement
a transaction in itself. Multiple updates can be accomplished by placing
them in a single statement, however they are limited to a single table.
Applications tend to require more robust use of transaction
concepts. Take, for example, a system that processes an order: A row may
be inserted in an order table, additional rows may be added to an
order-line table, updates may be made to inventory tables, etc. It is
important that if the order completes, all the changes are made to all
the tables involved; likewise if the order fails, none of the changes to
the tables must occur. To facilitate this requirement, @strong{MySQL}
has syntax to start a transaction called @code{BEGIN WORK}. All
statements that occur after the @code{BEGIN WORK} statement are grouped
into a single transaction. The end of this transaction occurs when a
@code{COMMIT} or @code{ROLLBACK} statement is encountered. After the
@code{COMMIT} or @code{ROLLBACK} the system returns back to the behavior
before the @code{BEGIN WORK} statement was encountered where every
statement is a transaction.
To permanently turn off the behavior where every statement is a
transaction, @strong{MySQL} added a variable called
@code{AUTOCOMMIT}. The @code{AUTOCOMMIT} variable can have two values,
@code{1} and @code{0}. The mode where every statement is a transaction
is when @code{AUTOCOMMIT} is set to @code{1} (@code{AUTOCOMMIT=1}). When
@code{AUTOCOMMIT} is set to @code{0} (@code{AUTOCOMMIT=0}), then every
statement is part of the same transaction until the transaction end by
either @code{COMMIT} or @code{ROLLBACK}. Once a transaction completes, a
new transaction is immediately started and the process repeats.
Here is an example of the SQL statements that you may find in a typical
order:
@example
BEGIN WORK;
INSERT INTO order VALUES ...;
INSERT INTO order-lines VALUES ...;
INSERT INTO order-lines VALUES ...;
INSERT INTO order-lines VALUES ...;
UPDATE inventory WHERE ...;
COMMIT;
@end example
This example shows how to use the @code{BEGIN WORK} statement to start a
transaction. If the variable @code{AUTOCOMMIT} is set to @code{0}, then
a transaction would have been started already. In this case, the
@code{BEGIN WORK} commits the current transaction and starts a new one.
@cindex GEMINI tables, statement atomicity
@noindent
@strong{Statement Atomicity}
As mentioned above, when running with @code{AUTOCOMMIT} set to @code{1},
each statement executes as a single transaction. When a statement has an
error, then all changes make by the statement must be
undone. Transactions support this behavior. Non-transaction safe table
handlers would have a partial statement update where some of the changes
from the statement would be contained in the database and other changes
from the statement would not. Work would need to be done to manually
recover from the error.
@cindex GEMINI tables, recovery
@noindent
@strong{Recovery}
Transactions are the basis for database recovery. Recovery is what
supports the Durability attribute of the ACID transaction.
@code{GEMINI} uses a separate file called the Recovery Log located in
the @code{$DATADIR} directory named @code{gemini.rl}. This file
maintains the integrity of all the @code{GEMINI} tables. @code{GEMINI}
can not recover any data from non-@code{GEMINI} tables. In addition, the
@code{gemini.rl} file is used to rollback transactions in support of the
@code{ROLLBACK} statement.
In the event of a system failure, the next time the @strong{MySQL}
server is started, @code{GEMINI} will automatically go through its
crash recovery process. The result of crash recovery is that all the
@code{GEMINI} tables will contain the latest changes made to them, and
all transactions that were open at the time of the crash will have been
rolled back.
The @code{GEMINI} Recovery Log reuses space when it can. Space can be
reused when information in the Recovery Log is no longer needed for
crash recovery or rollback.
@cindex GEMINI tables, isolation levels
@noindent
@strong{Isolation Levels}
There are four isolation levels supported by @code{GEMINI}:
@itemize @bullet
@item
READ UNCOMMITTED
@item
READ COMMITTED
@item
REPEATABLE READ
@item
SERIALIZABLE
@end itemize
These isolation levels apply only to shared locks obtained by select
statements, excluding select for update. Statements that get exclusive
locks always retain those locks until the transaction commits or rolls
back.
By default, @code{GEMINI} operates at the @code{READ COMMITTED}
level. You can override the default using the following command:
@example
SET [GLOBAL | SESSION] TRANSACTION ISOLATION LEVEL [READ UNCOMMITTED |
READ COMMITTED | REPEATABLE READ | SERIALIZABLE ]
@end example
If the @code{SESSION} qualifier used, the specified isolation level
persists for the entire session. If the @code{GLOBAL} qualifier is used,
the specified isolation level is applied to all new connections from
this point forward. Note that the specified isolation level will not
change the behavior for existing connections including the connection
that exectues the @code{SET GLOBAL TRANSACTION ISOLATION LEVEL}
statement.
@multitable @columnfractions .30 .70
@item @strong{Option} @tab @strong{Meaning}
@item @code{--gemini-full-recovery} @tab Default.
@item @code{--gemini-no-recovery} @tab Turn off recovery logging. Not recommended.
@item @code{--gemini-lazy-commit} @tab Relaxes the flush log at commit rule.
@item @code{--gemini-unbuffered-io} @tab All database writes bypass OS cache.
@item @code{--skip-gemini} @tab Don't use Gemini.
@item @code{--O gemini_db_buffers=#} @tab Number of database buffers in database cache.
@item @code{--O gemini_connection_limit=#} @tab Maximum number of connections to Gemini.
@item @code{--O gemini_spin_retries=#} @tab Spin lock retries (optimization).
@item @code{--O gemini_io_threads=#} @tab Number of background I/O threads.
@item @code{--O gemini_lock_table_size=#} @tab Set the maximum number of locks. Default 4096.
@item @sc{Isolation Level} @tab @sc{Description}
@item
@strong{READ UNCOMMITTED}
@tab Does not obtain any locks when reading rows. This means that if a
row is locked by another process in a transaction that has a more strict
isolation level, the @code{READ UNCOMMITTED} query will not wait until
the locks are released before reading the row. You will get an error if
attempt any updates while running at this isolation level.
@item
@strong{READ COMMITTED}
@tab Locks the requested rows long enough to copy the row from the
database block to the client row buffer. If a @code{READ COMMITTED}
query finds that a row is locked exclusively by another process, it will
wait until either the row has been released, or the lock timeout value
has expired.
@item
@strong{REPEATABLE READ}
@tab Locks all the rows needed to satisfy the query. These locks are
held until the transaction ends (commits or rolls back). If a
@code{REPEATABLE READ} query finds that a row is locked exclusively by
another process, it will wait until either the row has been released, or
the lock timeout value has expired.
@item
@strong{SERIALIZABLE}
@tab Locks the table that contains the rows needed to satisfy the
query. This lock is held until the transaction ends (commits or rolls
back). If a @code{SERIALIZABLE} query finds that a row is exclusively
locked by another process, it will wait until either the row has been
released, or the lock timeout value has expired.
@end multitable
The statements that get exclusive locks are @code{INSERT},
@code{UPDATE}, @code{DELETE} and @code{SELECT ... FOR UPDATE}. Select
statements without the @code{FOR UPDATE} qualifier get shared locks
which allow other not ''for update'' select statements to read the same
rows but block anyone trying to update the row from accessing it. Rows
or tables with exclusive locks block all access to the row from other
transactions until the transaction ends.
In general terms, the higher the Isolation level the more likelihood of
having concurrent locks and therefore lock conflicts. In such cases,
adjust the @code{-O gemini_lock_table_size} accordingly.
@cindex GEMINI tables, row-level locking
@noindent
@strong{Row-Level Locking}
@code{GEMINI} uses row locks, which allows high concurrency for requests
on the same table.
In order to avoid lock table overflow, SQL statements that require
applying locks to a large number of rows should either be run at the
serializable isolation level or should be covered by a lock table
statement.
Memory must be pre-allocated for the lock table. The mysqld server
startup option @code{-0 gemini_lock_table_size} can be used to adjust
the number of concurrent locks.
@cindex GEMINI tables, limitations
@node GEMINI Limitations, , GEMINI Concepts, GEMINI Overview
@subsubsection GEMINI Limitations
The following limitations are in effect for the current version of
@code{GEMINI}:
@itemize @bullet
@item
@code{DROP DATABASE} does not work with @code{GEMINI} tables; instead,
drop all the tables in the database first, then drop the database.
@item
Maximum number of @code{GEMINI} tables is 1012.
@item
Maximum number of @code{GEMINI} files a server can manage is 1012. Each
table consumes one file; an additional file is consumed if the table has
any indexes defined on it.
@item
Maximum size of BLOBs is 16MB.
@item
@code{FULLTEXT} indexes are not supported with @code{GEMINI} tables.
@item
There is no support for multi-component @code{AUTO_INCREMENT} fields
that provide alternating values at the component level. If you try to
create such a field, @code{GEMINI} will refuse.
@item
@code{TEMPORARY TABLES} are not supported by @code{GEMINI}. The
statement @code{CREATE TEMPORARY TABLE ... TYPE=GEMINI} will generate
the response: @code{ERROR 1005: Can't create table '/tmp/#sqlxxxxx'
(errno: 0)}.
@item
@code{FLUSH TABLES} has not been implemented with @code{GEMINI} tables.
@end itemize
@cindex GEMINI tables, using
@node Using GEMINI Tables, , GEMINI Overview, GEMINI
@subsection Using GEMINI Tables
This section explains the various startup options you can use with
@code{GEMINI} tables, how to backup @code{GEMINI} tables, some
performance considerations and sample configurations, and a brief
discussion of when to use @code{GEMINI} tables.
Specifically, the topics covered in this section are:
@itemize @bullet
@item
Startup Options
@item
Creating @code{GEMINI} Tables
@item
Backing Up @code{GEMINI} Tables
@item
Using Auto_Increment Columns With @code{GEMINI} Tables
@item
Performance Considerations
@item
Sample Configurations
@item
When To Use @code{GEMINI} Tables
@end itemize
@cindex GEMINI tables, startup options
@menu
* Startup Options::
* Creating GEMINI Tables::
* Backing Up GEMINI Tables::
* Restoring GEMINI Tables::
* Using Auto_Increment Columns With GEMINI Tables::
* Performance Considerations::
* Sample Configurations::
* When To Use GEMINI Tables::
@end menu
@node Startup Options, Creating GEMINI Tables, Using GEMINI Tables, Using GEMINI Tables
@subsubsection Startup Options
The table below lists options to mysqld that can be used to change the
behavior of @code{GEMINI} tables.
@multitable @columnfractions .40 .60
@item @sc{Option} @tab @sc{Description}
@item
@code{--default-table-type=gemini}
@tab Sets the default table handler to be @code{GEMINI}. All create
table statements will create @code{GEMINI} tables unless otherwise
specified with @code{TYPE=@var{table-type}}. As noted above, there is
currently a limitation with @code{TEMPORARY} tables using @code{GEMINI}.
@item
@code{--gemini-flush-log-at-commit}
@tab Forces the recovery log buffers to be flushed after every
commit. This can have a serious performance penalty, so use with
caution.
@item
@code{--gemini-recovery=FULL | NONE | FORCE}
@tab Sets the recovery mode. Default is @code{FULL}. @code{NONE} is
useful for performing repeatable batch operations because the updates
are not recorded in the recovery log. @code{FORCE} skips crash recovery
upon startup; this corrupts the database, and should be used in
emergencies only.
@item
@code{--gemini-unbuffered-io}
@tab All database writes bypass the OS cache. This can provide a
performance boost on heavily updated systems where most of the dataset
being worked on is cached in memory with the @code{gemini_buffer_cache}
parameter.
@item
@code{--O gemini_buffer_cache=size}
@tab Amount of memory to allocate for database buffers, including Index
and Record information. It is recommended that this number be 10% of the
total size of all @code{GEMINI} tables. Do not exceed amount of memory
on the system!
@item
@code{--O gemini_connection_limit=#}
@tab Maximum number of connections to @code{GEMINI}; default is
@code{100}. Each connection consumes about 1K of memory.
@item
@code{--O gemini_io_threads=#}
@tab Number of background I/O threads; default is @code{2}. Increase the
number when using @code{--gemini-unbuffered-io}
@item
@code{--O gemini_lock_table_size=#}
@tab Sets the maximum number of concurrent locks; default is 4096. Using
@code{SET [ GLOBAL | SESSION ] TRANSACTION ISOLATION = ...} will
determine how long a program will hold row locks.
@item
@code{--O gemini_lock_wait_timeout=seconds}
@tab Number of seconds to wait for record locks when performing queries;
default is 10 seconds. Using @code{SET [ GLOBAL | SESSION ] TRANSACTION
ISOLATION = ...} will determine how long a program will hold row locks.
@item
@code{--skip-gemini}
@tab Do not use @code{GEMINI}. If you use @code{--skip-gemini}, @strong{MySQL}
will not initialize the @code{GEMINI} table handler, saving memory; you
cannot use @code{GEMINI} tables if you use @code{--skip-gemini}.
@item
@code{--transaction-isolation=READ-UNCOMMITTED | READ-COMMITTED | REPEATABLE-READ | SERIALIZABLE}
@tab Sets the GLOBAL transaction isolation level for all users that
connect to the server; can be overridden with the SET ISOLATION LEVEL
statement.
@end multitable
If you use @code{--skip-gemini}, @strong{MySQL} will not initialize the
Gemini table handler, saving memory; you cannot use Gemini tables if you
use @code{--skip-gemini}.
@cindex GEMINI tables, creating
@node Creating GEMINI Tables, Backing Up GEMINI Tables, Startup Options, Using GEMINI Tables
@subsubsection Creating GEMINI Tables
@node GEMINI features, GEMINI TODO, GEMINI start, GEMINI
@subsection Features of @code{GEMINI} tables:
@code{GEMINI} tables can be created by either using the @code{CREATE
TABLE} syntax or the @code{ALTER TABLE} syntax.
@itemize @bullet
@item
If a query result can be resolved solely from the index key, Gemini will
not read the actual row stored in the database.
The syntax for creating a @code{GEMINI} table is:
@example
CREATE TABLE @var{table-name} (....) TYPE=GEMINI;
@end example
@item
The syntax to convert a table to @code{GEMINI} is:
@example
ALTER TABLE @var{table-name} TYPE=GEMINI;
@end example
@end itemize
@xref{Tutorial}, for more information on how to create and use
@code{MySQL} tables.
@cindex GEMINI tables, backing up
@node Backing Up GEMINI Tables, Restoring GEMINI Tables, Creating GEMINI Tables, Using GEMINI Tables
@subsubsection Backing Up GEMINI Tables
@code{GEMINI} supports both @code{BACKUP TABLE} and @code{RESTORE TABLE}
syntax. To learn more about how to use @code{BACKUP} and @code{RESTORE},
see @ref{BACKUP TABLE} and @ref{RESTORE TABLE}.
To backup @code{GEMINI} tables outside of the @code{MySQL} environment,
you must first shut down the @code{MySQL} server. Once the server is
shut down, you can copy the files associated with @code{GEMINI} to a
different location. The files that make up the @code{GEMINI} table
handler are:
@itemize @bullet
@item
All files associated with a table with a @code{.gmd} extention below the
@code{$DATADIR} directory. Such files include @code{@var{table}.gmd},
@code{@var{table}.gmi}, and @code{@var{table}.frm}
@item
@code{gemini.db} in the @code{$DATADIR} directory
@item
@code{gemini.rl} in the @code{$DATADIR} directory
@item
@code{gemini.lg} in the @code{$DATADIR} directory
@end itemize
All the @code{GEMINI} files must be copied together. You can not copy
just the @code{.gmi} and @code{.gmd} files to a different
@code{$DATADIR} and have them become part of a new database. You can
copy an entire @code{$DATADIR} directory to another location and start a
@strong{MySQL} server using the new @code{$DATADIR}.
@cindex GEMINI tables, restoring
@node Restoring GEMINI Tables, Using Auto_Increment Columns With GEMINI Tables, Backing Up GEMINI Tables, Using GEMINI Tables
@subsubsection Restoring GEMINI Tables
To restore @code{GEMINI} tables outside of the @code{MySQL} environment,
you must first shut down the @code{MySQL} server. Once the server is
shut down, you can remove all @code{GEMINI} files in the target
@code{$DATADIR} and then copy the files previously backed up into the
@code{$DATADIR} directory.
As mentioned above, the files that make up the @code{GEMINI} table
handler are:
@itemize @bullet
@item
All files associated with a table with a @code{.gmd} extention below the
@code{$DATADIR} directory. Such files include @code{@var{table}.gmd},
@code{@var{table}.gmi}, and @code{@var{table}.frm}
@item
@code{gemini.db} in the @code{$DATADIR} directory
@item
@code{gemini.rl} in the @code{$DATADIR} directory
@item
@code{gemini.lg} in the @code{$DATADIR} directory
@end itemize
When restoring a table, all the @code{GEMINI} files must be copied
together. You can not restore just the @code{.gmi} and @code{.gmd}
files.
@cindex GEMINI tables, auto_increment
@node Using Auto_Increment Columns With GEMINI Tables, Performance Considerations, Restoring GEMINI Tables, Using GEMINI Tables
@subsubsection Using Auto_Increment Columns With GEMINI Tables
As mentioned previously, @code{GEMINI} tables support row-level and
table-level locking to increase concurrency in applications and to allow
reading of tables without locking for maximum concurrency in heavy
update environments. This feature has several implications when working
with @code{auto_increment} tables.
In @code{MySQL}, when a column is defined as an @code{auto_increment}
column, and a row is inserted into the table with a @code{NULL} for the
column, the @code{auto_increment} column is updated to be 1 higher than
the highest value in the column.
With @code{MyISAM} tables, the @code{auto_increment} function is
implemented by looking in the index and finding the highest value and
adding 1 to it. This is possible because the entire @code{ISAM} table is
locked during the update period and the increment value is therefore
guaranteed to not be changing.
With @code{GEMINI} tables, the @code{auto_increment} function is
implemented by maintaining a counter in a separate location from the
table data. Instead of looking at the highest value in the table index,
@code{GEMINI} tables look at this separately maintained counter. This
means that in a transactional model, unlike the bottleneck inherent in
the @code{MyISAM} approach, @code{GEMINI} users do @b{not} have to wait
until the transaction that added the last value either commits or
rollbacks before looking at the value.
Two side-effects of the @code{GEMINI} implementation are:
@itemize @bullet
@item
Locking on Gemini tables is done at row level.
If an insert is done where the column with the @code{auto_increment} is
specified, and this specified value is the highest value, @code{MyISAM}
uses it as its @code{auto_increment} value, and every subsequent insert
is based on this. By contrast, @code{GEMINI} does not use this value,
but instead uses the value maintained in the separate @code{GEMINI}
counter location.
@item
@code{SELECT COUNT(*) FROM table_name} is fast; Gemini maintains a count
of the number of rows in the table.
To set the counter to a specific value, you can use @code{SET
insert_id=#} and insert a new row in the table. However, as a general
rule, values should not be inserted into an @code{auto_increment}
column; the database manager should be maintaining this field, not the
application. @code{SET insert_id} is a recovery mechanism that should be
used in case of error only.
@end itemize
@node GEMINI TODO, , GEMINI features, GEMINI
@subsection Current limitations of @code{GEMINI} tables:
Note that if you delete the row containing the maximum value for an
@code{auto_increment} column, the value will be reused with a
@code{GEMINI} table but not with a @code{MyISAM} table.
See @ref{CREATE TABLE} for more information about creating
@code{auto_increment} columns.
@cindex GEMINI tables, peformance considerations
@node Performance Considerations, Sample Configurations, Using Auto_Increment Columns With GEMINI Tables, Using GEMINI Tables
@subsubsection Performance Considerations
In addition to designing the best possible application, configuration of
the data and the server startup parameters need to be considered. How
the hardware is being used can have a dramatic affect on how fast the
system will respond to queries. Disk Drives and Memory must both be
considered.
@noindent
@strong{Disk Drives}
For best performance, you want to spread the data out over as many disks
as possible. Using RAID 10 stripes work very well. If there are a lot of
updates then the recovery log (@code{gemini.rl}) should be on a
relatively quiet disk drive.
To spread the data out without using RAID 10, you can do the following:
@itemize @bullet
@item
BLOB columns are not supported in @code{GEMINI} tables.
Group all the tables into three categories: Heavy Use, Moderate Use,
Light Use.
@item
Take the number of disk drives available and use a round-robin approach
to the three categories grouping the tables on a disk drive. The result
will be an equal distribution of Heavy/Moderate/Light tables assigned to
each disk drive.
@item
Once the tables have been converted to @code{GEMINI} by using the
@code{ALTER TABLE <name> TYPE=GEMINI} statements, move (@code{mv}) the
@code{.gmd} and @code{.gmi} files to a different disk drive and link
(@code{ln -s}) them back to the original directory where the @code{.frm}
file resides.
@item
The maximum number of concurrent users accessing @code{GEMINI} tables is
limited by @code{gemini_connection_limit}. The default is 100 users.
Finally, move the @code{gemini.rl} file to its quiet disk location and link
the file back to the @code{$DATADIR} directory.
@end itemize
NuSphere is working on removing these limitations.
@noindent
@strong{Memory}
The more data that can be placed in memory the faster the access to the
data. Figure out how large the @code{GEMINI} data is by adding up the
@code{.gmd} and @code{.gmi} file sizes. If you can, put at least 10% of
the data into memory. You allocate memory for the rows and indexes by
using the @code{gemini_buffer_cache} startup parameter. For example:
@example
mysqld -O gemini_buffer_cache=800M
@end example
@noindent
would allocate 800 MB of memory for the @code{GEMINI} buffer cache.
@cindex GEMINI tables, sample configurations
@node Sample Configurations, When To Use GEMINI Tables, Performance Considerations, Using GEMINI Tables
@subsubsection Sample Configurations
Based on the performance considerations above, we can look at some
examples for how to get the best performance out of the system when
using @code{GEMINI} tables.
@multitable @columnfractions .30 .70
@item @sc{Hardware} @tab @sc{Configuration}
@item
One CPU, 128MB memory, one disk drive
@tab Allocate 80MB of memory for reading and updating @code{GEMINI}
tables by starting the mysqld server with the following option:
@example
-O gemini_buffer_cache=80M
@end example
@item
Two CPUs, 512MB memory, four disk drives
@tab Use RAID 10 to stripe the data across all available disks, or use
the method described in the performance considerations section,
above. Allocate 450MB of memory for reading/updating @code{GEMINI}
tables:
@example
-O gemini_buffer_cache=450M
@end example
@end multitable
@cindex GEMINI tables, when to use
@node When To Use GEMINI Tables, , Sample Configurations, Using GEMINI Tables
@subsubsection When To Use GEMINI Tables
Because the @code{GEMINI} table handler provides crash recovery and
transaction support, there is extra overhead that is not found in other
non-transaction safe table handlers. Here are some general guidelines
for when to employ @code{GEMINI} and when to use other non-transaction
safe tables (@code{NTST}).
@multitable @columnfractions .30 .25 .45
@item
@sc{Access Trends} @tab @sc{Table Type} @tab @sc{Reason}
@item
Read-only
@tab @code{NTST}
@tab Less overhead and faster
@item
Critical data
@tab @code{GEMINI}
@tab Crash recovery protection
@item
High concurrency
@tab @code{GEMINI}
@tab Row-level locking
@item
Heavy update
@tab @code{GEMINI}
@tab Row-level locking
@end multitable
The table below shows how a typical application schema could be defined.
@multitable @columnfractions .15 .30 .25 .30
@item
@sc{Table} @tab @sc{Contents} @tab @sc{Table Type} @tab @sc{Reason}
@item
account
@tab Customer account data
@tab @code{GEMINI}
@tab Critical data, heavy update
@item
order
@tab Orders for a customer
@tab @code{GEMINI}
@tab Critical data, heavy update
@item
orderline
@tab Orderline detail for an order
@tab @code{GEMINI}
@tab Critical data, heavy update
@item
invdesc
@tab Inventory description
@tab @code{NTST}
@tab Read-only, frequent access
@item
salesrep
@tab Sales rep information
@tab @code{NTST}
@tab Infrequent update
@item
inventory
@tab Inventory information
@tab @code{GEMINI}
@tab High concurrency, critical data
@item
config
@tab System configuration
@tab @code{NTST}
@tab Read-only
@end multitable
@node InnoDB, , GEMINI, Table types
@section InnoDB Tables
......@@ -999,10 +999,10 @@ dnl echo "DBG_GEM1: gemini='$gemini'"
gemini_includes=
gemini_libs=
case "$gemini" in
no | default | *)
no)
AC_MSG_RESULT([Not using Gemini DB])
;;
yes )
yes | default | *)
have_gemini_db="yes"
gemini_includes="-I../gemini/incl -I../gemini"
gemini_libs="\
......
......@@ -2020,6 +2020,17 @@ EOF
echo "END OF INNODB CONFIGURATION"
fi
if test "X$have_gemini_db" = "Xyes"; then
sql_server_dirs="gemini $sql_server_dirs"
echo "CONFIGURING FOR GEMINI DB"
(cd gemini && sh ./configure) \
|| AC_MSG_ERROR([could not configure Gemini DB])
echo "END OF GEMINI DB CONFIGURATION"
AC_DEFINE(HAVE_GEMINI_DB)
fi
if test "$with_posix_threads" = "no" -o "$with_mit_threads" = "yes"
then
# MIT user level threads
......
......@@ -213,6 +213,7 @@ enum ha_base_keytype {
#define HA_ERR_CRASHED_ON_USAGE 145 /* Table must be repaired */
#define HA_ERR_LOCK_WAIT_TIMEOUT 146
#define HA_ERR_LOCK_TABLE_FULL 147
#define HA_ERR_READ_ONLY_TRANSACTION 148 /* Updates not allowed */
/* Other constants */
......
......@@ -205,4 +205,7 @@
#define ER_SLAVE_THREAD 1202
#define ER_TOO_MANY_USER_CONNECTIONS 1203
#define ER_SET_CONSTANTS_ONLY 1204
#define ER_ERROR_MESSAGES 205
#define ER_LOCK_WAIT_TIMEOUT 1205
#define ER_LOCK_TABLE_FULL 1206
#define ER_READ_ONLY_TRANSACTION 1207
#define ER_ERROR_MESSAGES 208
......@@ -4087,6 +4087,59 @@ const char *Field_blob::unpack(char *to, const char *from)
}
#ifdef HAVE_GEMINI_DB
/* Blobs in Gemini tables are stored separately from the rows which contain
** them (except for tiny blobs, which are stored in the row). For all other
** blob types (blob, mediumblob, longblob), the row contains the length of
** the blob data and a blob id. These methods (pack_id, get_id, and
** unpack_id) handle packing and unpacking blob fields in Gemini rows.
*/
char *Field_blob::pack_id(char *to, const char *from, ulonglong id, uint max_length)
{
char *save=ptr;
ptr=(char*) from;
ulong length=get_length(); // Length of from string
if (length > max_length)
{
ptr=to;
length=max_length;
store_length(length); // Store max length
ptr=(char*) from;
}
else
memcpy(to,from,packlength); // Copy length
if (length)
{
int8store(to+packlength, id);
}
ptr=save; // Restore org row pointer
return to+packlength+sizeof(id);
}
ulonglong Field_blob::get_id(const char *from)
{
ulonglong id = 0;
ulong length=get_length(from);
if (length)
longlongget(id, from+packlength);
return id;
}
const char *Field_blob::unpack_id(char *to, const char *from, const char *bdata)
{
memcpy(to,from,packlength);
ulong length=get_length(from);
from+=packlength;
if (length)
memcpy_fixed(to+packlength, &bdata, sizeof(bdata));
else
bzero(to+packlength,sizeof(bdata));
return from+sizeof(ulonglong);
}
#endif /* HAVE_GEMINI_DB */
/* Keys for blobs are like keys on varchars */
int Field_blob::pack_cmp(const char *a, const char *b, uint key_length)
......
......@@ -869,6 +869,13 @@ class Field_blob :public Field_str {
}
char *pack(char *to, const char *from, uint max_length= ~(uint) 0);
const char *unpack(char *to, const char *from);
#ifdef HAVE_GEMINI_DB
char *pack_id(char *to, const char *from, ulonglong id,
uint max_length= ~(uint) 0);
ulonglong get_id(const char *from);
const char *unpack_id(char *to, const char *from, const char *bdata);
enum_field_types blobtype() { return (packlength == 1 ? FIELD_TYPE_TINY_BLOB : FIELD_TYPE_BLOB);}
#endif
char *pack_key(char *to, const char *from, uint max_length);
char *pack_key_from_key_image(char* to, const char *from, uint max_length);
int pack_cmp(const char *a, const char *b, uint key_length);
......
......@@ -19,10 +19,13 @@
#pragma implementation // gcc: Class implementation
#endif
#include "mysql_priv.h"
#ifdef HAVE_GEMINI_DB
#include <string.h>
#include "mysql_priv.h"
#include "my_pthread.h"
#ifdef HAVE_GEMINI_DB
#include "ha_gemini.h"
#include "dbconfig.h"
#include "dsmpub.h"
#include "recpub.h"
......@@ -34,7 +37,17 @@
#include <hash.h>
#include <stdarg.h>
#include "geminikey.h"
#include "ha_gemini.h"
#define gemini_msg MSGD_CALLBACK
pthread_mutex_t gem_mutex;
static HASH gem_open_tables;
static GEM_SHARE *get_share(const char *table_name, TABLE *table);
static int free_share(GEM_SHARE *share, bool mutex_is_locked);
static byte* gem_get_key(GEM_SHARE *share,uint *length,
my_bool not_used __attribute__((unused)));
static void gemini_lock_table_overflow_error(dsmContext_t *pcontext);
const char *ha_gemini_ext=".gmd";
const char *ha_gemini_idx_ext=".gmi";
......@@ -48,6 +61,7 @@ long gemini_locktablesize;
long gemini_lock_wait_timeout;
long gemini_spin_retries;
long gemini_connection_limit;
char *gemini_basedir;
const char gemini_dbname[] = "gemini";
dsmContext_t *pfirstContext = NULL;
......@@ -61,7 +75,7 @@ TYPELIB gemini_recovery_typelib= {array_elements(gemini_recovery_names),"",
const int start_of_name = 2; /* Name passed as ./<db>/<table-name>
and we're not interested in the ./ */
static const int keyBufSize = MYMAXKEYSIZE * 2;
static const int keyBufSize = MAXKEYSZ + FULLKEYHDRSZ + MAX_REF_PARTS + 16;
static int gemini_tx_begin(THD *thd);
static void print_msg(THD *thd, const char *table_name, const char *op_name,
......@@ -87,40 +101,56 @@ bool gemini_init(void)
goto badret;
}
/* dsmContextCreate and dsmContextSetString(DSM_TAGDB_DBNAME) must
** be the first DSM calls we make so that we can log any errors which
** occur in subsequent DSM calls. DO NOT INSERT ANY DSM CALLS IN
** BETWEEN THIS COMMENT AND THE COMMENT THAT SAYS "END OF CODE..."
*/
/* Gotta connect to the database regardless of the operation */
rc = dsmContextCreate(&pfirstContext);
if( rc != 0 )
{
printf("dsmContextCreate failed %ld\n",rc);
gemini_msg(pfirstContext, "dsmContextCreate failed %l",rc);
goto badret;
}
/* This call will also open the log file */
rc = dsmContextSetString(pfirstContext, DSM_TAGDB_DBNAME,
strlen(gemini_dbname), (TEXT *)gemini_dbname);
if( rc != 0 )
{
printf("Dbname tag failed %ld\n", rc);
gemini_msg(pfirstContext, "Dbname tag failed %l", rc);
goto badret;
}
/* END OF CODE NOT TO MESS WITH */
fn_format(pmsgsfile, GEM_MSGS_FILE, language, ".db", 2 | 4);
rc = dsmContextSetString(pfirstContext, DSM_TAGDB_MSGS_FILE,
strlen(pmsgsfile), (TEXT *)pmsgsfile);
if( rc != 0 )
{
printf("MSGS_DIR tag failed %ld\n", rc);
gemini_msg(pfirstContext, "MSGS_DIR tag failed %l", rc);
goto badret;
}
strxmov(pmsgsfile, gemini_basedir, GEM_SYM_FILE, NullS);
rc = dsmContextSetString(pfirstContext, DSM_TAGDB_SYMFILE,
strlen(pmsgsfile), (TEXT *)pmsgsfile);
if( rc != 0 )
{
gemini_msg(pfirstContext, "SYMFILE tag failed %l", rc);
goto badret;
}
rc = dsmContextSetLong(pfirstContext,DSM_TAGDB_ACCESS_TYPE,DSM_ACCESS_STARTUP);
if ( rc != 0 )
{
printf("ACCESS TAG set failed %ld\n",rc);
gemini_msg(pfirstContext, "ACCESS TAG set failed %l",rc);
goto badret;
}
rc = dsmContextSetLong(pfirstContext,DSM_TAGDB_ACCESS_ENV, DSM_SQL_ENGINE);
if( rc != 0 )
{
printf("ACCESS_ENV set failed %ld",rc);
gemini_msg(pfirstContext, "ACCESS_ENV set failed %l",rc);
goto badret;
}
......@@ -129,7 +159,7 @@ bool gemini_init(void)
(TEXT *)mysql_real_data_home);
if( rc != 0 )
{
printf("Datadir tag failed %ld\n", rc);
gemini_msg(pfirstContext, "Datadir tag failed %l", rc);
goto badret;
}
......@@ -137,7 +167,7 @@ bool gemini_init(void)
gemini_connection_limit);
if(rc != 0)
{
printf("MAX_USERS tag set failed %ld",rc);
gemini_msg(pfirstContext, "MAX_USERS tag set failed %l",rc);
goto badret;
}
......@@ -145,7 +175,7 @@ bool gemini_init(void)
gemini_lock_wait_timeout);
if(rc != 0)
{
printf("MAX_LOCK_ENTRIES tag set failed %ld",rc);
gemini_msg(pfirstContext, "MAX_LOCK_ENTRIES tag set failed %l",rc);
goto badret;
}
......@@ -153,7 +183,7 @@ bool gemini_init(void)
gemini_locktablesize);
if(rc != 0)
{
printf("MAX_LOCK_ENTRIES tag set failed %ld",rc);
gemini_msg(pfirstContext, "MAX_LOCK_ENTRIES tag set failed %l",rc);
goto badret;
}
......@@ -161,7 +191,7 @@ bool gemini_init(void)
gemini_spin_retries);
if(rc != 0)
{
printf("SPIN_AMOUNT tag set failed %ld",rc);
gemini_msg(pfirstContext, "SPIN_AMOUNT tag set failed %l",rc);
goto badret;
}
......@@ -172,22 +202,22 @@ bool gemini_init(void)
gemini_buffer_cache);
if(rc != 0)
{
printf("DB_BUFFERS tag set failed %ld",rc);
gemini_msg(pfirstContext, "DB_BUFFERS tag set failed %l",rc);
goto badret;
}
rc = dsmContextSetLong(pfirstContext, DSM_TAGDB_FLUSH_AT_COMMIT,
((gemini_options & GEMOPT_FLUSH_LOG) ? 1 : 0));
((gemini_options & GEMOPT_FLUSH_LOG) ? 0 : 1));
if(rc != 0)
{
printf("FLush_Log_At_Commit tag set failed %ld",rc);
gemini_msg(pfirstContext, "FLush_Log_At_Commit tag set failed %l",rc);
goto badret;
}
rc = dsmContextSetLong(pfirstContext, DSM_TAGDB_DIRECT_IO,
((gemini_options & GEMOPT_UNBUFFERED_IO) ? 1 : 0));
if(rc != 0)
{
printf("DIRECT_IO tag set failed %ld",rc);
gemini_msg(pfirstContext, "DIRECT_IO tag set failed %l",rc);
goto badret;
}
......@@ -195,10 +225,20 @@ bool gemini_init(void)
((gemini_recovery_options & GEMINI_RECOVERY_FULL) ? 1 : 0));
if(rc != 0)
{
printf("CRASH_PROTECTION tag set failed %ld",rc);
gemini_msg(pfirstContext, "CRASH_PROTECTION tag set failed %l",rc);
goto badret;
}
if (gemini_recovery_options & GEMINI_RECOVERY_FORCE)
{
rc = dsmContextSetLong(pfirstContext, DSM_TAGDB_FORCE_ACCESS, 1);
if(rc != 0)
{
printf("CRASH_PROTECTION tag set failed %ld",rc);
goto badret;
}
}
/* cluster size will come in bytes, need to convert it to
16 K units. */
gemini_log_cluster_size = (gemini_log_cluster_size + 16383) / 16384;
......@@ -207,7 +247,7 @@ bool gemini_init(void)
if(rc != 0)
{
printf("CRASH_PROTECTION tag set failed %ld",rc);
gemini_msg(pfirstContext, "CRASH_PROTECTION tag set failed %l",rc);
goto badret;
}
......@@ -215,12 +255,20 @@ bool gemini_init(void)
DSM_DB_OPENDB | DSM_DB_OPENFILE);
if( rc != 0 )
{
printf("dsmUserConnect failed rc = %ld\n",rc);
/* Message is output in dbenv() */
goto badret;
}
/* Set access to shared for subsequent user connects */
rc = dsmContextSetLong(pfirstContext,DSM_TAGDB_ACCESS_TYPE,DSM_ACCESS_SHARED);
rc = gemini_helper_threads(pfirstContext);
(void) hash_init(&gem_open_tables,32,0,0,
(hash_get_key) gem_get_key,0,0);
pthread_mutex_init(&gem_mutex,NULL);
DBUG_RETURN(0);
badret:
......@@ -231,30 +279,40 @@ bool gemini_init(void)
static int gemini_helper_threads(dsmContext_t *pContext)
{
int rc = 0;
int i;
pthread_attr_t thr_attr;
pthread_t hThread;
DBUG_ENTER("gemini_helper_threads");
rc = pthread_create (&hThread, 0, gemini_watchdog, (void *)pContext);
(void) pthread_attr_init(&thr_attr);
#if !defined(HAVE_DEC_3_2_THREADS)
pthread_attr_setscope(&thr_attr,PTHREAD_SCOPE_SYSTEM);
(void) pthread_attr_setdetachstate(&thr_attr,PTHREAD_CREATE_DETACHED);
pthread_attr_setstacksize(&thr_attr,32768);
#endif
rc = pthread_create (&hThread, &thr_attr, gemini_watchdog, (void *)pContext);
if (rc)
{
printf("Can't create gemini watchdog thread");
gemini_msg(pContext, "Can't Create gemini watchdog thread");
goto done;
}
if(!gemini_io_threads)
goto done;
rc = pthread_create(&hThread, 0, gemini_rl_writer, (void *)pContext);
rc = pthread_create(&hThread, &thr_attr, gemini_rl_writer, (void *)pContext);
if(rc)
{
printf("Can't create gemini recovery log writer thread");
gemini_msg(pContext, "Can't create Gemini recovery log writer thread");
goto done;
}
for( int i = gemini_io_threads - 1;i;i--)
for(i = gemini_io_threads - 1;i;i--)
{
rc = pthread_create(&hThread, 0, gemini_apw, (void *)pContext);
rc = pthread_create(&hThread, &thr_attr, gemini_apw, (void *)pContext);
if(rc)
{
printf("Can't create gemini page writer thread");
gemini_msg(pContext, "Can't create Gemini database page writer thread");
goto done;
}
}
......@@ -273,7 +331,7 @@ pthread_handler_decl(gemini_watchdog,arg )
rc = dsmContextCopy(pcontext,&pmyContext, DSMCONTEXTDB);
if( rc != 0 )
{
printf("dsmContextCopy failed for watchdog %d\n",rc);
gemini_msg(pcontext, "dsmContextCopy failed for Gemini watchdog %d",rc);
return 0;
}
......@@ -281,7 +339,7 @@ pthread_handler_decl(gemini_watchdog,arg )
if( rc != 0 )
{
printf("dsmUserConnect failed for watchdog %d\n",rc);
gemini_msg(pcontext, "dsmUserConnect failed for Gemini watchdog %d",rc);
return 0;
}
......@@ -311,7 +369,7 @@ pthread_handler_decl(gemini_rl_writer,arg )
rc = dsmContextCopy(pcontext,&pmyContext, DSMCONTEXTDB);
if( rc != 0 )
{
printf("dsmContextCopy failed for recovery log writer %d\n",rc);
gemini_msg(pcontext, "dsmContextCopy failed for Gemini recovery log writer %d",rc);
return 0;
}
......@@ -319,7 +377,7 @@ pthread_handler_decl(gemini_rl_writer,arg )
if( rc != 0 )
{
printf("dsmUserConnect failed for recovery log writer %d\n",rc);
gemini_msg(pcontext, "dsmUserConnect failed for Gemini recovery log writer %d",rc);
return 0;
}
......@@ -348,7 +406,7 @@ pthread_handler_decl(gemini_apw,arg )
rc = dsmContextCopy(pcontext,&pmyContext, DSMCONTEXTDB);
if( rc != 0 )
{
printf("dsmContextCopy failed for gemini page writer %d\n",rc);
gemini_msg(pcontext, "dsmContextCopy failed for Gemini page writer %d",rc);
my_thread_end();
return 0;
}
......@@ -356,7 +414,7 @@ pthread_handler_decl(gemini_apw,arg )
if( rc != 0 )
{
printf("dsmUserConnect failed for gemini page writer %d\n",rc);
gemini_msg(pcontext, "dsmUserConnect failed for Gemini page writer %d",rc);
my_thread_end();
return 0;
}
......@@ -388,7 +446,7 @@ int gemini_set_option_long(int optid, long optval)
}
if (rc)
{
printf("SPIN_AMOUNT tag set failed %ld",rc);
gemini_msg(pfirstContext, "SPIN_AMOUNT tag set failed %l",rc);
}
else
{
......@@ -410,7 +468,7 @@ static int gemini_connect(THD *thd)
DSMCONTEXTDB);
if( rc != 0 )
{
printf("dsmContextCopy failed %ld\n",rc);
gemini_msg(pfirstContext, "dsmContextCopy failed %l",rc);
return(rc);
}
......@@ -418,7 +476,7 @@ static int gemini_connect(THD *thd)
if( rc != 0 )
{
printf("dsmUserConnect failed %ld\n",rc);
gemini_msg(pfirstContext, "dsmUserConnect failed %l",rc);
return(rc);
}
......@@ -444,6 +502,9 @@ bool gemini_end(void)
THD *thd;
DBUG_ENTER("gemini_end");
hash_free(&gem_open_tables);
pthread_mutex_destroy(&gem_mutex);
if(pfirstContext)
{
rc = dsmShutdownSet(pfirstContext, DSM_SHUTDOWN_NORMAL);
......@@ -534,6 +595,24 @@ int gemini_rollback_to_savepoint(THD *thd)
DBUG_RETURN(rc);
}
int gemini_recovery_logging(THD *thd, bool on)
{
int error;
int noLogging;
if(!thd->gemini.context)
return 0;
if(on)
noLogging = 0;
else
noLogging = 1;
error = dsmContextSetLong((dsmContext_t *)thd->gemini.context,
DSM_TAGCONTEXT_NO_LOGGING,noLogging);
return error;
}
/* gemDataType - translates from mysql data type constant to gemini
key services data type contstant */
int gemDataType ( int mysqlType )
......@@ -599,8 +678,13 @@ int ha_gemini::open(const char *name, int mode, uint test_if_locked)
DBUG_ENTER("ha_gemini::open");
thd = current_thd;
thr_lock_init(&alock);
thr_lock_data_init(&alock,&lock,(void*)0);
/* Init shared structure */
if (!(share=get_share(name,table)))
{
DBUG_RETURN(1); /* purecov: inspected */
}
thr_lock_data_init(&share->lock,&lock,(void*) 0);
ref_length = sizeof(dsmRecid_t);
if(thd->gemini.context == NULL)
......@@ -610,7 +694,7 @@ int ha_gemini::open(const char *name, int mode, uint test_if_locked)
if(rc)
return rc;
}
if (!(rec_buff=my_malloc(table->rec_buff_length,
if (!(rec_buff=(byte*)my_malloc(table->rec_buff_length,
MYF(MY_WME))))
{
DBUG_RETURN(1);
......@@ -635,6 +719,12 @@ int ha_gemini::open(const char *name, int mode, uint test_if_locked)
rc = dsmObjectNameToNum((dsmContext_t *)thd->gemini.context,
(dsmText_t *)name_buff,
&tableId);
if (rc)
{
gemini_msg((dsmContext_t *)thd->gemini.context,
"Unable to find table number for %s", name_buff);
DBUG_RETURN(rc);
}
}
tableNumber = tableId;
......@@ -649,8 +739,33 @@ int ha_gemini::open(const char *name, int mode, uint test_if_locked)
crashed while being in the midst of a repair operation */
rc = dsmTableStatus((dsmContext_t *)thd->gemini.context,
tableNumber,&tableStatus);
if(tableStatus)
if(tableStatus == DSM_OBJECT_IN_REPAIR)
tableStatus = HA_ERR_CRASHED;
pthread_mutex_lock(&share->mutex);
share->use_count++;
pthread_mutex_unlock(&share->mutex);
if (table->blob_fields)
{
/* Allocate room for the blob ids from an unpacked row. Note that
** we may not actually need all of this space because tiny blobs
** are stored in the packed row, not in a separate storage object
** like larger blobs. But we allocate an entry for all blobs to
** keep the code simpler.
*/
pBlobDescs = (gemBlobDesc_t *)my_malloc(
table->blob_fields * sizeof(gemBlobDesc_t),
MYF(MY_WME | MY_ZEROFILL));
}
else
{
pBlobDescs = 0;
}
get_index_stats(thd);
info(HA_STATUS_CONST);
DBUG_RETURN (rc);
}
......@@ -680,6 +795,12 @@ int ha_gemini::index_open(char *tableName)
rc = dsmObjectNameToNum((dsmContext_t *)thd->gemini.context,
(dsmText_t *)tableName,
&objectNumber);
if (rc)
{
gemini_msg((dsmContext_t *)thd->gemini.context,
"Unable to file Index number for %s", tableName);
DBUG_RETURN(rc);
}
pindexNumbers[i] = objectNumber;
}
}
......@@ -692,12 +813,22 @@ int ha_gemini::index_open(char *tableName)
int ha_gemini::close(void)
{
DBUG_ENTER("ha_gemini::close");
thr_lock_delete(&alock);
my_free(rec_buff,MYF(MY_ALLOW_ZERO_PTR));
my_free((char*)rec_buff,MYF(MY_ALLOW_ZERO_PTR));
rec_buff = 0;
my_free((char *)pindexNumbers,MYF(MY_ALLOW_ZERO_PTR));
pindexNumbers = 0;
DBUG_RETURN(0);
if (pBlobDescs)
{
for (uint i = 0; i < table->blob_fields; i++)
{
my_free((char*)pBlobDescs[i].pBlob, MYF(MY_ALLOW_ZERO_PTR));
}
my_free((char *)pBlobDescs, MYF(0));
pBlobDescs = 0;
}
DBUG_RETURN(free_share(share, 0));
}
......@@ -709,7 +840,7 @@ int ha_gemini::write_row(byte * record)
DBUG_ENTER("write_row");
if(tableStatus)
if(tableStatus == HA_ERR_CRASHED)
DBUG_RETURN(tableStatus);
thd = current_thd;
......@@ -737,10 +868,11 @@ int ha_gemini::write_row(byte * record)
/* A set insert-id statement so set the auto-increment value if this
value is higher than it's current value */
error = dsmTableAutoIncrement((dsmContext_t *)thd->gemini.context,
tableNumber, (ULONG64 *)&nr);
tableNumber, (ULONG64 *)&nr,1);
if(thd->next_insert_id > nr)
{
error = dsmTableAutoIncrementSet((dsmContext_t *)thd->gemini.context,tableNumber,
error = dsmTableAutoIncrementSet((dsmContext_t *)thd->gemini.context,
tableNumber,
(ULONG64)thd->next_insert_id);
}
}
......@@ -749,11 +881,13 @@ int ha_gemini::write_row(byte * record)
}
dsmRecord.table = tableNumber;
dsmRecord.maxLength = table->reclength;
dsmRecord.maxLength = table->rec_buff_length;
if ((error=pack_row((byte **)&dsmRecord.pbuffer, (int *)&dsmRecord.recLength,
record)))
record, FALSE)))
{
DBUG_RETURN(error);
}
error = dsmRecordCreate((dsmContext_t *)thd->gemini.context,
&dsmRecord,0);
......@@ -769,6 +903,8 @@ int ha_gemini::write_row(byte * record)
thd->gemini.needSavepoint = 1;
}
}
if(error == DSM_S_RQSTREJ)
error = HA_ERR_LOCK_WAIT_TIMEOUT;
DBUG_RETURN(error);
}
......@@ -777,10 +913,17 @@ longlong ha_gemini::get_auto_increment()
{
longlong nr;
int error;
int update;
THD *thd=current_thd;
if(thd->lex.sql_command == SQLCOM_SHOW_TABLES)
update = 0;
else
update = 1;
error = dsmTableAutoIncrement((dsmContext_t *)thd->gemini.context,
tableNumber, (ULONG64 *)&nr);
tableNumber, (ULONG64 *)&nr,
update);
return nr;
}
......@@ -828,8 +971,8 @@ int ha_gemini::handleIndexEntry(const byte * record, dsmRecid_t recid,
expects that the three lead bytes of the header are
not counted in this length -- But cxKeyPrepare also
expects that these three bytes are present in the keystr */
theKey.akey.keyLen = (COUNT)keyStringLen - 3;
theKey.akey.unknown_comp = thereIsAnull;
theKey.akey.keyLen = (COUNT)keyStringLen - FULLKEYHDRSZ;
theKey.akey.unknown_comp = (dsmBoolean_t)thereIsAnull;
theKey.akey.word_index = 0;
theKey.akey.descending_key =0;
if(option == KEY_CREATE)
......@@ -880,6 +1023,7 @@ int ha_gemini::createKeyString(const byte * record, KEY *pkeyinfo,
int componentLen;
int fieldType;
int isNull;
uint key_part_length;
KEY_PART_INFO *key_part;
......@@ -892,21 +1036,35 @@ int ha_gemini::createKeyString(const byte * record, KEY *pkeyinfo,
unsigned char *pos;
key_part = pkeyinfo->key_part + i;
key_part_length = key_part->length;
fieldType = gemDataType(key_part->field->type());
if(fieldType == GEM_CHAR)
switch (fieldType)
{
case GEM_CHAR:
{
/* Save the current ptr to the field in case we're building a key
to remove an old key value when an indexed character column
gets updated. */
char *ptr = key_part->field->ptr;
key_part->field->ptr = (char *)record + key_part->offset;
key_part->field->sort_string(rec_buff, key_part->length);
key_part->field->sort_string((char*)rec_buff, key_part->length);
key_part->field->ptr = ptr;
pos = (unsigned char *)rec_buff;
}
else
{
}
break;
case GEM_TINYBLOB:
case GEM_BLOB:
case GEM_MEDIUMBLOB:
case GEM_LONGBLOB:
((Field_blob*)key_part->field)->get_ptr((char**)&pos);
key_part_length = ((Field_blob*)key_part->field)->get_length(
(char*)record + key_part->offset);
break;
default:
pos = (unsigned char *)record + key_part->offset;
break;
}
isNull = record[key_part->null_offset] & key_part->null_bit;
......@@ -914,7 +1072,7 @@ int ha_gemini::createKeyString(const byte * record, KEY *pkeyinfo,
*thereIsAnull = true;
rc = gemFieldToIdxComponent(pos,
(unsigned long) key_part->length,
(unsigned long) key_part_length,
fieldType,
isNull ,
key_part->field->flags & UNSIGNED_FLAG,
......@@ -951,7 +1109,7 @@ int ha_gemini::update_row(const byte * old_record, byte * new_record)
}
for (uint keynr=0 ; keynr < table->keys ; keynr++)
{
if(key_cmp(keynr,old_record, new_record))
if(key_cmp(keynr,old_record, new_record,false))
{
error = handleIndexEntry(old_record,lastRowid,KEY_DELETE,keynr);
if(error)
......@@ -973,10 +1131,10 @@ int ha_gemini::update_row(const byte * old_record, byte * new_record)
dsmRecord.table = tableNumber;
dsmRecord.recid = lastRowid;
dsmRecord.maxLength = table->reclength;
dsmRecord.maxLength = table->rec_buff_length;
if ((error=pack_row((byte **)&dsmRecord.pbuffer, (int *)&dsmRecord.recLength,
new_record)))
new_record, TRUE)))
{
DBUG_RETURN(error);
}
......@@ -992,6 +1150,7 @@ int ha_gemini::delete_row(const byte * record)
int error = 0;
dsmRecord_t dsmRecord;
THD *thd = current_thd;
dsmContext_t *pcontext = (dsmContext_t *)thd->gemini.context;
DBUG_ENTER("delete_row");
statistic_increment(ha_delete_count,&LOCK_status);
......@@ -999,9 +1158,7 @@ int ha_gemini::delete_row(const byte * record)
if(thd->gemini.needSavepoint)
{
thd->gemini.savepoint++;
error = dsmTransaction((dsmContext_t *)thd->gemini.context,
&thd->gemini.savepoint,
DSMTXN_SAVE, 0, 0);
error = dsmTransaction(pcontext, &thd->gemini.savepoint, DSMTXN_SAVE, 0, 0);
if (error)
DBUG_RETURN(error);
thd->gemini.needSavepoint = 0;
......@@ -1013,8 +1170,27 @@ int ha_gemini::delete_row(const byte * record)
error = handleIndexEntries(record, dsmRecord.recid,KEY_DELETE);
if(!error)
{
error = dsmRecordDelete((dsmContext_t *)thd->gemini.context,
&dsmRecord, 0, NULL);
error = dsmRecordDelete(pcontext, &dsmRecord, 0, NULL);
}
/* Delete any blobs associated with this row */
if (table->blob_fields)
{
dsmBlob_t gemBlob;
gemBlob.areaType = DSMOBJECT_BLOB;
gemBlob.blobObjNo = tableNumber;
for (uint i = 0; i < table->blob_fields; i++)
{
if (pBlobDescs[i].blobId)
{
gemBlob.blobId = pBlobDescs[i].blobId;
my_free((char *)pBlobDescs[i].pBlob, MYF(MY_ALLOW_ZERO_PTR));
dsmBlobStart(pcontext, &gemBlob);
dsmBlobDelete(pcontext, &gemBlob, NULL);
/* according to DSM doc, no need to call dsmBlobEnd() */
}
}
}
DBUG_RETURN(error);
......@@ -1023,7 +1199,6 @@ int ha_gemini::delete_row(const byte * record)
int ha_gemini::index_init(uint keynr)
{
int error = 0;
int keyStringLen;
THD *thd;
DBUG_ENTER("index_init");
thd = current_thd;
......@@ -1046,19 +1221,9 @@ int ha_gemini::index_init(uint keynr)
}
pbracketBase->index = 0;
pbracketLimit->index = (dsmIndex_t)pindexNumbers[keynr];
pbracketLimit->keycomps = 1;
keyStringLen = 0;
error = gemKeyHigh(pbracketLimit->keystr, &keyStringLen,
pbracketLimit->index);
/* We have to subtract three here since cxKeyPrepare
expects that the three lead bytes of the header are
not counted in this length -- But cxKeyPrepare also
expects that these three bytes are present in the keystr */
pbracketLimit->keyLen = (COUNT)keyStringLen - 3;
pbracketBase->descending_key = pbracketLimit->descending_key = 0;
pbracketBase->ksubstr = pbracketLimit->ksubstr = 0;
pbracketLimit->keycomps = pbracketBase->keycomps = 1;
pfoundKey = (dsmKey_t *)my_malloc(sizeof(dsmKey_t) + keyBufSize,MYF(MY_WME));
if(!pfoundKey)
......@@ -1130,6 +1295,7 @@ int ha_gemini::pack_key( uint keynr, dsmKey_t *pkey,
{
uint offset=0;
unsigned char *pos;
uint key_part_length = key_part->length;
int fieldType;
if (key_part->null_bit)
......@@ -1141,7 +1307,7 @@ int ha_gemini::pack_key( uint keynr, dsmKey_t *pkey,
key_ptr+= key_part->store_length;
rc = gemFieldToIdxComponent(
(unsigned char *)key_ptr + offset,
(unsigned long) key_part->length,
(unsigned long) key_part_length,
0,
1 , /* Tells it to build a null component */
key_part->field->flags & UNSIGNED_FLAG,
......@@ -1153,20 +1319,31 @@ int ha_gemini::pack_key( uint keynr, dsmKey_t *pkey,
}
}
fieldType = gemDataType(key_part->field->type());
if(fieldType == GEM_CHAR)
switch (fieldType)
{
key_part->field->store(key_ptr + offset, key_part->length);
key_part->field->sort_string(rec_buff, key_part->length);
case GEM_CHAR:
key_part->field->store((char*)key_ptr + offset, key_part->length);
key_part->field->sort_string((char*)rec_buff, key_part->length);
pos = (unsigned char *)rec_buff;
}
else
{
break;
case GEM_TINYBLOB:
case GEM_BLOB:
case GEM_MEDIUMBLOB:
case GEM_LONGBLOB:
((Field_blob*)key_part->field)->get_ptr((char**)&pos);
key_part_length = ((Field_blob*)key_part->field)->get_length(
(char*)key_ptr + offset);
break;
default:
pos = (unsigned char *)key_ptr + offset;
break;
}
rc = gemFieldToIdxComponent(
pos,
(unsigned long) key_part->length,
(unsigned long) key_part_length,
fieldType,
0 ,
key_part->field->flags & UNSIGNED_FLAG,
......@@ -1189,7 +1366,7 @@ void ha_gemini::unpack_key(char *record, dsmKey_t *key, uint index)
int fieldIsNull, fieldType;
int rc = 0;
char unsigned *pos= &key->keystr[7];
char unsigned *pos= &key->keystr[FULLKEYHDRSZ+4/* 4 for the index number*/];
for ( ; key_part != end; key_part++)
{
......@@ -1202,7 +1379,8 @@ void ha_gemini::unpack_key(char *record, dsmKey_t *key, uint index)
}
rc = gemIdxComponentToField(pos, fieldType,
(unsigned char *)record + key_part->field->offset(),
key_part->field->field_length,
//key_part->field->field_length,
key_part->length,
key_part->field->decimals(),
&fieldIsNull);
if(fieldIsNull)
......@@ -1266,12 +1444,12 @@ int ha_gemini::index_read(byte * buf, const byte * key,
pbracketLimit->keyLen = componentLen;
}
/* We have to subtract three here since cxKeyPrepare
/* We have to subtract the header size here since cxKeyPrepare
expects that the three lead bytes of the header are
not counted in this length -- But cxKeyPrepare also
expects that these three bytes are present in the keystr */
pbracketBase->keyLen -= 3;
pbracketLimit->keyLen -= 3;
pbracketBase->keyLen -= FULLKEYHDRSZ;
pbracketLimit->keyLen -= FULLKEYHDRSZ;
thd = current_thd;
......@@ -1294,7 +1472,7 @@ int ha_gemini::index_next(byte * buf)
dsmMask_t findMode;
DBUG_ENTER("index_next");
if(tableStatus)
if(tableStatus == HA_ERR_CRASHED)
DBUG_RETURN(tableStatus);
thd = current_thd;
......@@ -1304,9 +1482,12 @@ int ha_gemini::index_next(byte * buf)
error = gemKeyLow(pbracketBase->keystr, &keyStringLen,
pbracketLimit->index);
pbracketBase->keyLen = (COUNT)keyStringLen - 3;
pbracketBase->keyLen = (COUNT)keyStringLen - FULLKEYHDRSZ;
pbracketBase->index = pbracketLimit->index;
pbracketBase->keycomps = 1;
error = gemKeyHigh(pbracketLimit->keystr, &keyStringLen,
pbracketLimit->index);
pbracketLimit->keyLen = (COUNT)keyStringLen - FULLKEYHDRSZ;
findMode = DSMFINDFIRST;
}
else
......@@ -1369,24 +1550,20 @@ int ha_gemini::index_last(byte * buf)
error = gemKeyLow(pbracketBase->keystr, &keyStringLen,
pbracketLimit->index);
if(error)
goto errorReturn;
pbracketBase->keyLen = (COUNT)keyStringLen - 3;
pbracketBase->keyLen = (COUNT)keyStringLen - FULLKEYHDRSZ;
pbracketBase->index = pbracketLimit->index;
pbracketBase->keycomps = 1;
error = gemKeyHigh(pbracketLimit->keystr, &keyStringLen,
pbracketLimit->index);
pbracketLimit->keyLen = (COUNT)keyStringLen - FULLKEYHDRSZ;
error = findRow(thd,DSMFINDLAST,buf);
errorReturn:
if (error == DSM_S_ENDLOOP)
error = HA_ERR_END_OF_FILE;
table->status = error ? STATUS_NOT_FOUND : 0;
DBUG_RETURN(error);
table->status = error ? STATUS_NOT_FOUND : 0;
DBUG_RETURN(error);
}
int ha_gemini::rnd_init(bool scan)
......@@ -1414,7 +1591,7 @@ int ha_gemini::rnd_next(byte *buf)
DBUG_ENTER("rnd_next");
if(tableStatus)
if(tableStatus == HA_ERR_CRASHED)
DBUG_RETURN(tableStatus);
thd = current_thd;
......@@ -1429,7 +1606,7 @@ int ha_gemini::rnd_next(byte *buf)
dsmRecord.recid = lastRowid;
dsmRecord.pbuffer = (dsmBuffer_t *)rec_buff;
dsmRecord.recLength = table->reclength;
dsmRecord.maxLength = table->reclength;
dsmRecord.maxLength = table->rec_buff_length;
error = dsmTableScan((dsmContext_t *)thd->gemini.context,
&dsmRecord, DSMFINDNEXT, lockMode, 0);
......@@ -1437,17 +1614,23 @@ int ha_gemini::rnd_next(byte *buf)
if(!error)
{
lastRowid = dsmRecord.recid;
unpack_row((char *)buf,(char *)dsmRecord.pbuffer);
error = unpack_row((char *)buf,(char *)dsmRecord.pbuffer);
}
if(!error)
;
else if (error == DSM_S_ENDLOOP)
error = HA_ERR_END_OF_FILE;
else if (error == DSM_S_RQSTREJ)
error = HA_ERR_LOCK_WAIT_TIMEOUT;
else if (error == DSM_S_LKTBFULL)
error = HA_ERR_LOCK_TABLE_FULL;
else
{
lastRowid = 0;
if (error == DSM_S_ENDLOOP)
error = HA_ERR_END_OF_FILE;
else if (error == DSM_S_RQSTREJ)
error = HA_ERR_LOCK_WAIT_TIMEOUT;
else if (error == DSM_S_LKTBFULL)
{
error = HA_ERR_LOCK_TABLE_FULL;
gemini_lock_table_overflow_error((dsmContext_t *)thd->gemini.context);
}
}
table->status = error ? STATUS_NOT_FOUND : 0;
DBUG_RETURN(error);
}
......@@ -1500,14 +1683,14 @@ int ha_gemini::fetch_row(void *gemini_context,const byte *buf)
dsmRecord.recid = lastRowid;
dsmRecord.pbuffer = (dsmBuffer_t *)rec_buff;
dsmRecord.recLength = table->reclength;
dsmRecord.maxLength = table->reclength;
dsmRecord.maxLength = table->rec_buff_length;
rc = dsmRecordGet((dsmContext_t *)gemini_context,
&dsmRecord, 0);
if(!rc)
{
unpack_row((char *)buf,(char *)dsmRecord.pbuffer);
rc = unpack_row((char *)buf,(char *)dsmRecord.pbuffer);
}
DBUG_RETURN(rc);
......@@ -1544,7 +1727,7 @@ int ha_gemini::findRow(THD *thd, dsmMask_t findMode, byte *buf)
if(key_read)
{
unpack_key(buf, pkey, active_index);
unpack_key((char*)buf, pkey, active_index);
}
if(!key_read) /* unpack_key may have turned off key_read */
{
......@@ -1554,10 +1737,17 @@ int ha_gemini::findRow(THD *thd, dsmMask_t findMode, byte *buf)
errorReturn:
if(!rc)
;
else if(rc == DSM_S_RQSTREJ)
rc = HA_ERR_LOCK_WAIT_TIMEOUT;
else if (rc == DSM_S_LKTBFULL)
rc = HA_ERR_LOCK_TABLE_FULL;
else
{
lastRowid = 0;
if(rc == DSM_S_RQSTREJ)
rc = HA_ERR_LOCK_WAIT_TIMEOUT;
else if (rc == DSM_S_LKTBFULL)
{
rc = HA_ERR_LOCK_TABLE_FULL;
gemini_lock_table_overflow_error((dsmContext_t *)thd->gemini.context);
}
}
DBUG_RETURN(rc);
}
......@@ -1578,25 +1768,47 @@ void ha_gemini::info(uint flag)
dsmStatus_t error;
ULONG64 rows;
if(thd->gemini.context == NULL)
{
/* Need to get this thread a connection into the database */
error = gemini_connect(thd);
if(error)
DBUG_VOID_RETURN;
}
error = dsmRowCount((dsmContext_t *)thd->gemini.context,tableNumber,&rows);
records = (ha_rows)rows;
deleted = 0;
}
else if ((flag & HA_STATUS_CONST))
if ((flag & HA_STATUS_CONST))
{
;
ha_rows *rec_per_key = share->rec_per_key;
for (uint i = 0; i < table->keys; i++)
for(uint k=0;
k < table->key_info[i].key_parts; k++,rec_per_key++)
table->key_info[i].rec_per_key[k] = *rec_per_key;
}
else if ((flag & HA_STATUS_ERRKEY))
if ((flag & HA_STATUS_ERRKEY))
{
errkey=last_dup_key;
}
else if ((flag & HA_STATUS_TIME))
if ((flag & HA_STATUS_TIME))
{
;
}
else if ((flag & HA_STATUS_AUTO))
if ((flag & HA_STATUS_AUTO))
{
;
THD *thd = current_thd;
dsmStatus_t error;
error = dsmTableAutoIncrement((dsmContext_t *)thd->gemini.context,
tableNumber,
(ULONG64 *)&auto_increment_value,
0);
/* Should return the next auto-increment value that
will be given -- so we need to increment the one dsm
currently reports. */
auto_increment_value++;
}
DBUG_VOID_RETURN;
......@@ -1658,7 +1870,22 @@ int ha_gemini::external_lock(THD *thd, int lock_type)
thd->gemini.lock_count = 1;
thd->gemini.tx_isolation = thd->tx_isolation;
}
// lockMode has already been set in store_lock
// If the statement about to be executed calls for
// exclusive locks and we're running at read uncommitted
// isolation level then raise an error.
if(thd->gemini.tx_isolation == ISO_READ_UNCOMMITTED)
{
if(lockMode == DSM_LK_EXCL)
{
DBUG_RETURN(HA_ERR_READ_ONLY_TRANSACTION);
}
else
{
lockMode = DSM_LK_NOLOCK;
}
}
if(thd->gemini.context == NULL)
{
/* Need to get this thread a connection into the database */
......@@ -1678,6 +1905,8 @@ int ha_gemini::external_lock(THD *thd, int lock_type)
rc = dsmObjectLock((dsmContext_t *)thd->gemini.context,
(dsmObject_t)tableNumber,DSMOBJECT_TABLE,0,
lockMode, 1, 0);
if(rc == DSM_S_RQSTREJ)
rc = HA_ERR_LOCK_WAIT_TIMEOUT;
}
}
else /* lock_type == F_UNLK */
......@@ -1703,18 +1932,24 @@ THR_LOCK_DATA **ha_gemini::store_lock(THD *thd, THR_LOCK_DATA **to,
!thd->in_lock_tables)
lock_type = TL_WRITE_ALLOW_WRITE;
lock.type=lock_type;
if(thd->gemini.tx_isolation == ISO_READ_UNCOMMITTED)
lockMode = DSM_LK_NOLOCK;
else if(table->reginfo.lock_type > TL_WRITE_ALLOW_READ)
lockMode = DSM_LK_EXCL;
else
lockMode = DSM_LK_SHARE;
}
if(table->reginfo.lock_type > TL_WRITE_ALLOW_READ)
lockMode = DSM_LK_EXCL;
else
lockMode = DSM_LK_SHARE;
*to++= &lock;
return to;
}
void ha_gemini::update_create_info(HA_CREATE_INFO *create_info)
{
table->file->info(HA_STATUS_AUTO | HA_STATUS_CONST);
if (!(create_info->used_fields & HA_CREATE_USED_AUTO))
{
create_info->auto_increment_value=auto_increment_value;
}
}
int ha_gemini::create(const char *name, register TABLE *form,
HA_CREATE_INFO *create_info)
......@@ -1777,7 +2012,7 @@ int ha_gemini::create(const char *name, register TABLE *form,
(dsmText_t *)"gemini_data_area");
if( rc != 0 )
{
printf("dsmAreaNew failed %ld\n",rc);
gemini_msg(pcontext, "dsmAreaNew failed %l",rc);
return(rc);
}
......@@ -1787,7 +2022,7 @@ int ha_gemini::create(const char *name, register TABLE *form,
(dsmText_t *)&name_buff[start_of_name]);
if( rc != 0 )
{
printf("dsmExtentCreate failed %ld\n",rc);
gemini_msg(pcontext, "dsmExtentCreate failed %l",rc);
return(rc);
}
......@@ -1805,6 +2040,20 @@ int ha_gemini::create(const char *name, register TABLE *form,
(dsmText_t *)&name_buff[start_of_name],
&dummy,&dummy);
if (rc == 0 && table->blob_fields)
{
/* create a storage object record for blob fields */
rc = dsmObjectCreate(pcontext, areaNumber, &tableNumber,
DSMOBJECT_BLOB,0,0,0,
(dsmText_t *)&name_buff[start_of_name],
&dummy,&dummy);
if( rc != 0 )
{
gemini_msg(pcontext, "dsmObjectCreate for blob object failed %l",rc);
return(rc);
}
}
if(rc == 0 && form->keys)
{
fn_format(name_buff, name, "", ha_gemini_idx_ext, 2 | 4);
......@@ -1814,7 +2063,7 @@ int ha_gemini::create(const char *name, register TABLE *form,
(dsmText_t *)"gemini_index_area");
if( rc != 0 )
{
printf("dsmAreaNew failed %ld\n",rc);
gemini_msg(pcontext, "dsmAreaNew failed %l",rc);
return(rc);
}
/* Create an extent */
......@@ -1823,7 +2072,7 @@ int ha_gemini::create(const char *name, register TABLE *form,
(dsmText_t *)&name_buff[start_of_name]);
if( rc != 0 )
{
printf("dsmExtentCreate failed %ld\n",rc);
gemini_msg(pcontext, "dsmExtentCreate failed %l",rc);
return(rc);
}
......@@ -1859,10 +2108,11 @@ int ha_gemini::create(const char *name, register TABLE *form,
}
}
rc = dsmTableAutoIncrementSet(pcontext,tableNumber,
create_info->auto_increment_value);
/* The auto_increment value is the next one to be given
out so give dsm one less than this value */
if(create_info->auto_increment_value)
rc = dsmTableAutoIncrementSet(pcontext,tableNumber,
create_info->auto_increment_value-1);
/* Get a table lock on this table in case this table is being
created as part of an alter table statement. We don't want
......@@ -1950,26 +2200,25 @@ int ha_gemini::delete_table(const char *pname)
(dsmObject_t *)&tableNum);
if (rc)
{
printf("Cound not find table number for %s with string %s, %ld\n",
pname,name_buff,rc);
gemini_msg(pcontext, "Unable to find table number for %s", name_buff);
rc = gemini_rollback(thd);
if (rc)
{
printf("Error in rollback %ld\n",rc);
gemini_msg(pcontext, "Error in rollback %l",rc);
}
DBUG_RETURN(rc);
}
rc = dsmObjectInfo(pcontext, tableNum, DSMOBJECT_MIXTABLE, &tableArea,
&objectAttr, &associate, &associateType, &block, &root);
rc = dsmObjectInfo(pcontext, tableNum, DSMOBJECT_MIXTABLE, tableNum,
&tableArea, &objectAttr, &associateType, &block, &root);
if (rc)
{
printf("Failed to get area number for table %d, %s, return %ld\n",
gemini_msg(pcontext, "Failed to get area number for table %d, %s, return %l",
tableNum, pname, rc);
rc = gemini_rollback(thd);
if (rc)
{
printf("Error in rollback %ld\n",rc);
gemini_msg(pcontext, "Error in rollback %l",rc);
}
}
......@@ -1979,14 +2228,14 @@ int ha_gemini::delete_table(const char *pname)
rc = dsmObjectDeleteAssociate(pcontext, tableNum, &indexArea);
if (rc)
{
printf("Error deleting storage objects for table number %d, return %ld\n",
gemini_msg(pcontext, "Error deleting storage objects for table number %d, return %l",
(int)tableNum, rc);
/* roll back txn and return */
rc = gemini_rollback(thd);
if (rc)
{
printf("Error in rollback %ld\n",rc);
gemini_msg(pcontext, "Error in rollback %l",rc);
}
DBUG_RETURN(rc);
}
......@@ -1994,33 +2243,33 @@ int ha_gemini::delete_table(const char *pname)
if (indexArea != DSMAREA_INVALID)
{
/* Delete the extents for both Index and Table */
rc = dsmExtentDelete(pcontext, indexArea, 0);
rc = dsmExtentDelete(pcontext, indexArea);
rc = dsmAreaDelete(pcontext, indexArea);
if (rc)
{
printf("Error deleting Index Area %ld, return %ld\n", indexArea, rc);
gemini_msg(pcontext, "Error deleting Index Area %l, return %l", indexArea, rc);
/* roll back txn and return */
rc = gemini_rollback(thd);
if (rc)
{
printf("Error in rollback %ld\n",rc);
gemini_msg(pcontext, "Error in rollback %l",rc);
}
DBUG_RETURN(rc);
}
}
rc = dsmExtentDelete(pcontext, tableArea, 0);
rc = dsmExtentDelete(pcontext, tableArea);
rc = dsmAreaDelete(pcontext, tableArea);
if (rc)
{
printf("Error deleting table Area %ld, name %s, return %ld\n",
gemini_msg(pcontext, "Error deleting table Area %l, name %s, return %l",
tableArea, pname, rc);
/* roll back txn and return */
rc = gemini_rollback(thd);
if (rc)
{
printf("Error in rollback %ld\n",rc);
gemini_msg(pcontext, "Error in rollback %l",rc);
}
DBUG_RETURN(rc);
}
......@@ -2030,7 +2279,7 @@ int ha_gemini::delete_table(const char *pname)
rc = gemini_commit(thd);
if (rc)
{
printf("Failed to commit transaction %ld\n",rc);
gemini_msg(pcontext, "Failed to commit transaction %l",rc);
}
......@@ -2047,7 +2296,6 @@ int ha_gemini::rename_table(const char *pfrom, const char *pto)
THD *thd;
dsmContext_t *pcontext;
dsmStatus_t rc;
char tabname_buff[FN_REFLEN];
char dbname_buff[FN_REFLEN];
char name_buff[FN_REFLEN];
char newname_buff[FN_REFLEN];
......@@ -2056,6 +2304,7 @@ int ha_gemini::rename_table(const char *pfrom, const char *pto)
unsigned i, nameLen;
dsmObject_t tableNum;
dsmArea_t indexArea = 0;
dsmArea_t tableArea = 0;
DBUG_ENTER("ha_gemini::rename_table");
......@@ -2068,7 +2317,7 @@ int ha_gemini::rename_table(const char *pfrom, const char *pto)
{
if (gemini_is_vst(name_buff))
{
return 0;
return DSM_S_CANT_RENAME_VST;
}
}
}
......@@ -2113,21 +2362,51 @@ int ha_gemini::rename_table(const char *pfrom, const char *pto)
rc = dsmObjectNameToNum(pcontext, (dsmText_t *)name_buff, &tableNum);
if (rc)
{
gemini_msg(pcontext, "Unable to file Table number for %s", name_buff);
goto errorReturn;
}
rc = dsmObjectRename(pcontext, tableNum,
(dsmText_t *)newname_buff,
(dsmText_t *)&newidxextname_buff[start_of_name],
(dsmText_t *)&newextname_buff[start_of_name],
&indexArea);
&indexArea, &tableArea);
if (rc)
{
gemini_msg(pcontext, "Failed to rename %s to %s",name_buff,newname_buff);
goto errorReturn;
}
/* Rename the physical table and index files (if necessary).
** Close the file, rename it, and reopen it (have to do it this
** way so rename works on Windows).
*/
if (!(rc = dsmAreaClose(pcontext, tableArea)))
{
if (!(rc = rename_file_ext(pfrom, pto, ha_gemini_ext)))
{
rc = dsmAreaOpen(pcontext, tableArea, 0);
if (rc)
{
gemini_msg(pcontext, "Failed to reopen area %d",tableArea);
}
}
}
/* rename the physical table and index files (if necessary) */
rc = rename_file_ext(pfrom, pto, ha_gemini_ext);
if (!rc && indexArea)
{
rc = rename_file_ext(pfrom, pto, ha_gemini_idx_ext);
if (!(rc = dsmAreaClose(pcontext, indexArea)))
{
if (!(rc = rename_file_ext(pfrom, pto, ha_gemini_idx_ext)))
{
rc = dsmAreaOpen(pcontext, indexArea, 0);
if (rc)
{
gemini_msg(pcontext, "Failed to reopen area %d",tableArea);
}
}
}
}
errorReturn:
......@@ -2143,17 +2422,38 @@ int ha_gemini::rename_table(const char *pfrom, const char *pto)
double ha_gemini::scan_time()
{
return records / (gemini_blocksize / table->reclength);
return (double)records /
(double)((gemini_blocksize / (double)table->reclength));
}
int ha_gemini::check(THD* thd, HA_CHECK_OPT* check_opt)
int ha_gemini::analyze(THD* thd, HA_CHECK_OPT* check_opt)
{
int error;
uint saveIsolation;
dsmMask_t saveLockMode;
check_opt->quick = true;
check_opt->optimize = true; // Tells check not to get table lock
saveLockMode = lockMode;
saveIsolation = thd->gemini.tx_isolation;
thd->gemini.tx_isolation = ISO_READ_UNCOMMITTED;
lockMode = DSM_LK_NOLOCK;
error = check(thd,check_opt);
lockMode = saveLockMode;
thd->gemini.tx_isolation = saveIsolation;
return (error);
}
int ha_gemini::check(THD* thd, HA_CHECK_OPT* check_opt)
{
int error = 0;
int checkStatus = HA_ADMIN_OK;
ha_rows indexCount;
byte *buf = 0, *indexBuf = 0;
byte *buf = 0, *indexBuf = 0, *prevBuf = 0;
int errorCount = 0;
info(HA_STATUS_VARIABLE); // Makes sure row count is up to date
/* Get a shared table lock */
if(thd->gemini.needSavepoint)
{
......@@ -2167,23 +2467,33 @@ int ha_gemini::check(THD* thd, HA_CHECK_OPT* check_opt)
return(error);
thd->gemini.needSavepoint = 0;
}
buf = my_malloc(table->rec_buff_length,MYF(MY_WME));
indexBuf = my_malloc(table->rec_buff_length,MYF(MY_WME));
buf = (byte*)my_malloc(table->rec_buff_length,MYF(MY_WME));
indexBuf = (byte*)my_malloc(table->rec_buff_length,MYF(MY_WME));
prevBuf = (byte*)my_malloc(table->rec_buff_length,MYF(MY_WME |MY_ZEROFILL ));
/* Lock the table */
error = dsmObjectLock((dsmContext_t *)thd->gemini.context,
(dsmObject_t)tableNumber,
DSMOBJECT_TABLE,0,
DSM_LK_SHARE, 1, 0);
if (!check_opt->optimize)
error = dsmObjectLock((dsmContext_t *)thd->gemini.context,
(dsmObject_t)tableNumber,
DSMOBJECT_TABLE,0,
DSM_LK_SHARE, 1, 0);
if(error)
{
gemini_msg((dsmContext_t *)thd->gemini.context,
"Failed to lock table %d, error %d",tableNumber, error);
return error;
}
info(HA_STATUS_VARIABLE);
ha_rows *rec_per_key = share->rec_per_key;
/* If quick option just scan along index converting and counting entries */
for (uint i = 0; i < table->keys; i++)
{
key_read = 1;
key_read = 1; // Causes data to be extracted from the keys
indexCount = 0;
// Clear the cardinality stats for this index
memset(table->key_info[i].rec_per_key,0,
sizeof(table->key_info[0].rec_per_key[0]) *
table->key_info[i].key_parts);
error = index_init(i);
error = index_first(indexBuf);
while(!error)
......@@ -2195,8 +2505,12 @@ int ha_gemini::check(THD* thd, HA_CHECK_OPT* check_opt)
error = fetch_row(thd->gemini.context,buf);
if(!error)
{
if(key_cmp(i,buf,indexBuf))
if(key_cmp(i,buf,indexBuf,false))
{
gemini_msg((dsmContext_t *)thd->gemini.context,
"Check Error! Key does not match row for rowid %d for index %s",
lastRowid,table->key_info[i].name);
print_msg(thd,table->real_name,"check","error",
"Key does not match row for rowid %d for index %s",
lastRowid,table->key_info[i].name);
......@@ -2209,6 +2523,9 @@ int ha_gemini::check(THD* thd, HA_CHECK_OPT* check_opt)
{
errorCount++;
checkStatus = HA_ADMIN_CORRUPT;
gemini_msg((dsmContext_t *)thd->gemini.context,
"Check Error! Key does not have a valid row pointer %d for index %s",
lastRowid,table->key_info[i].name);
print_msg(thd,table->real_name,"check","error",
"Key does not have a valid row pointer %d for index %s",
lastRowid,table->key_info[i].name);
......@@ -2218,10 +2535,27 @@ int ha_gemini::check(THD* thd, HA_CHECK_OPT* check_opt)
}
}
}
key_cmp(i,indexBuf,prevBuf,true);
bcopy((void *)indexBuf,(void *)prevBuf,table->rec_buff_length);
if(!error)
error = index_next(indexBuf);
}
for(uint j=1; j < table->key_info[i].key_parts; j++)
{
table->key_info[i].rec_per_key[j] += table->key_info[i].rec_per_key[j-1];
}
for(uint k=0; k < table->key_info[i].key_parts; k++)
{
if (table->key_info[i].rec_per_key[k])
table->key_info[i].rec_per_key[k] =
records / table->key_info[i].rec_per_key[k];
*rec_per_key = table->key_info[i].rec_per_key[k];
rec_per_key++;
}
if(error == HA_ERR_END_OF_FILE)
{
/* Check count of rows */
......@@ -2231,6 +2565,10 @@ int ha_gemini::check(THD* thd, HA_CHECK_OPT* check_opt)
/* Number of index entries does not agree with the number of
rows in the index. */
checkStatus = HA_ADMIN_CORRUPT;
gemini_msg((dsmContext_t *)thd->gemini.context,
"Check Error! Total rows %d does not match total index entries %d for %s",
records, indexCount,
table->key_info[i].name);
print_msg(thd,table->real_name,"check","error",
"Total rows %d does not match total index entries %d for %s",
records, indexCount,
......@@ -2248,23 +2586,61 @@ int ha_gemini::check(THD* thd, HA_CHECK_OPT* check_opt)
{
/* Now scan the table and for each row generate the keys
and find them in the index */
error = fullCheck(thd, buf);\
error = fullCheck(thd, buf);
if(error)
checkStatus = error;
}
// Store the key distribution information
error = saveKeyStats(thd);
error_return:
my_free(buf,MYF(MY_ALLOW_ZERO_PTR));
my_free((char*)buf,MYF(MY_ALLOW_ZERO_PTR));
my_free((char*)indexBuf,MYF(MY_ALLOW_ZERO_PTR));
my_free((char*)prevBuf,MYF(MY_ALLOW_ZERO_PTR));
index_end();
key_read = 0;
error = dsmObjectUnlock((dsmContext_t *)thd->gemini.context,
(dsmObject_t)tableNumber,
DSMOBJECT_TABLE,0,
DSM_LK_SHARE,0);
if(!check_opt->optimize)
{
error = dsmObjectUnlock((dsmContext_t *)thd->gemini.context,
(dsmObject_t)tableNumber,
DSMOBJECT_TABLE,0,
DSM_LK_SHARE,0);
if (error)
{
gemini_msg((dsmContext_t *)thd->gemini.context,
"Unable to unlock table %d", tableNumber);
}
}
return checkStatus;
}
int ha_gemini::saveKeyStats(THD *thd)
{
dsmStatus_t rc = 0;
/* Insert a row in the indexStats table for each column of
each index of the table */
for(uint i = 0; i < table->keys; i++)
{
for (uint j = 0; j < table->key_info[i].key_parts && !rc ;j++)
{
rc = dsmIndexStatsPut((dsmContext_t *)thd->gemini.context,
tableNumber, pindexNumbers[i],
j, (LONG64)table->key_info[i].rec_per_key[j]);
if (rc)
{
gemini_msg((dsmContext_t *)thd->gemini.context,
"Failed to update index stats for table %d, index %d",
tableNumber, pindexNumbers[i]);
}
}
}
return rc;
}
int ha_gemini::fullCheck(THD *thd,byte *buf)
{
int error;
......@@ -2319,7 +2695,12 @@ int ha_gemini::repair(THD* thd, HA_CHECK_OPT* check_opt)
&thd->gemini.savepoint,
DSMTXN_SAVE, 0, 0);
if (error)
{
gemini_msg((dsmContext_t *)thd->gemini.context,
"Error setting savepoint number %d, error %d",
thd->gemini.savepoint++, error);
return(error);
}
thd->gemini.needSavepoint = 0;
}
......@@ -2330,7 +2711,11 @@ int ha_gemini::repair(THD* thd, HA_CHECK_OPT* check_opt)
DSMOBJECT_TABLE,0,
DSM_LK_EXCL, 1, 0);
if(error)
{
gemini_msg((dsmContext_t *)thd->gemini.context,
"Failed to lock table %d, error %d",tableNumber, error);
return error;
}
error = dsmContextSetLong((dsmContext_t *)thd->gemini.context,
DSM_TAGCONTEXT_NO_LOGGING,1);
......@@ -2338,13 +2723,18 @@ int ha_gemini::repair(THD* thd, HA_CHECK_OPT* check_opt)
error = dsmTableReset((dsmContext_t *)thd->gemini.context,
(dsmTable_t)tableNumber, table->keys,
pindexNumbers);
if (error)
{
gemini_msg((dsmContext_t *)thd->gemini.context,
"dsmTableReset failed for table %d, error %d",tableNumber, error);
}
buf = my_malloc(table->rec_buff_length,MYF(MY_WME));
buf = (byte*)my_malloc(table->rec_buff_length,MYF(MY_WME));
dsmRecord.table = tableNumber;
dsmRecord.recid = 0;
dsmRecord.pbuffer = (dsmBuffer_t *)rec_buff;
dsmRecord.recLength = table->reclength;
dsmRecord.maxLength = table->reclength;
dsmRecord.maxLength = table->rec_buff_length;
while(!error)
{
error = dsmTableScan((dsmContext_t *)thd->gemini.context,
......@@ -2352,13 +2742,15 @@ int ha_gemini::repair(THD* thd, HA_CHECK_OPT* check_opt)
1);
if(!error)
{
unpack_row((char *)buf,(char *)dsmRecord.pbuffer);
error = handleIndexEntries(buf,dsmRecord.recid,KEY_CREATE);
if(error == HA_ERR_FOUND_DUPP_KEY)
if (!(error = unpack_row((char *)buf,(char *)dsmRecord.pbuffer)))
{
/* We don't want to stop on duplicate keys -- we're repairing
here so let's get as much repaired as possible. */
error = 0;
error = handleIndexEntries(buf,dsmRecord.recid,KEY_CREATE);
if(error == HA_ERR_FOUND_DUPP_KEY)
{
/* We don't want to stop on duplicate keys -- we're repairing
here so let's get as much repaired as possible. */
error = 0;
}
}
}
}
......@@ -2366,7 +2758,13 @@ int ha_gemini::repair(THD* thd, HA_CHECK_OPT* check_opt)
(dsmObject_t)tableNumber,
DSMOBJECT_TABLE,0,
DSM_LK_EXCL,0);
my_free(buf,MYF(MY_ALLOW_ZERO_PTR));
if (error)
{
gemini_msg((dsmContext_t *)thd->gemini.context,
"Unable to unlock table %d", tableNumber);
}
my_free((char*)buf,MYF(MY_ALLOW_ZERO_PTR));
error = dsmContextSetLong((dsmContext_t *)thd->gemini.context,
DSM_TAGCONTEXT_NO_LOGGING,0);
......@@ -2374,6 +2772,313 @@ int ha_gemini::repair(THD* thd, HA_CHECK_OPT* check_opt)
return error;
}
int ha_gemini::restore(THD* thd, HA_CHECK_OPT *check_opt)
{
dsmContext_t *pcontext = (dsmContext_t *)thd->gemini.context;
char* backup_dir = thd->lex.backup_dir;
char src_path[FN_REFLEN], dst_path[FN_REFLEN];
char* table_name = table->real_name;
int error = 0;
int errornum;
const char* errmsg = "";
dsmArea_t tableArea = 0;
dsmObjectAttr_t objectAttr;
dsmObject_t associate;
dsmObjectType_t associateType;
dsmDbkey_t block, root;
dsmStatus_t rc;
rc = dsmObjectInfo(pcontext, tableNumber, DSMOBJECT_MIXTABLE, tableNumber,
&tableArea, &objectAttr, &associateType, &block, &root);
if (rc)
{
error = HA_ADMIN_FAILED;
errmsg = "Failed in dsmObjectInfo (.gmd) (Error %d)";
errornum = rc;
gemini_msg(pcontext, errmsg ,errornum);
goto err;
}
rc = dsmAreaFlush(pcontext, tableArea, FLUSH_BUFFERS | FLUSH_SYNC);
if (rc)
{
error = HA_ADMIN_FAILED;
errmsg = "Failed in dsmAreaFlush (.gmd) (Error %d)";
errornum = rc;
gemini_msg(pcontext, errmsg ,errornum);
goto err;
}
rc = dsmAreaClose(pcontext, tableArea);
if (rc)
{
error = HA_ADMIN_FAILED;
errmsg = "Failed in dsmAreaClose (.gmd) (Error %d)";
errornum = rc;
gemini_msg(pcontext, errmsg ,errornum);
goto err;
}
/* Restore the data file */
if (!fn_format(src_path, table_name, backup_dir, ha_gemini_ext, 4 + 64))
{
return HA_ADMIN_INVALID;
}
if (my_copy(src_path, fn_format(dst_path, table->path, "",
ha_gemini_ext, 4), MYF(MY_WME)))
{
error = HA_ADMIN_FAILED;
errmsg = "Failed in my_copy (.gmd) (Error %d)";
errornum = errno;
gemini_msg(pcontext, errmsg ,errornum);
goto err;
}
rc = dsmAreaFlush(pcontext, tableArea, FREE_BUFFERS);
if (rc)
{
error = HA_ADMIN_FAILED;
errmsg = "Failed in dsmAreaFlush (.gmd) (Error %d)";
errornum = rc;
gemini_msg(pcontext, errmsg ,errornum);
goto err;
}
rc = dsmAreaOpen(pcontext, tableArea, 1);
if (rc)
{
error = HA_ADMIN_FAILED;
errmsg = "Failed in dsmAreaOpen (.gmd) (Error %d)";
errornum = rc;
gemini_msg(pcontext, errmsg ,errornum);
goto err;
}
#ifdef GEMINI_BACKUP_IDX
dsmArea_t indexArea = 0;
rc = dsmObjectInfo(pcontext, tableNumber, DSMOBJECT_MIXINDEX, &indexArea,
&objectAttr, &associate, &associateType, &block, &root);
if (rc)
{
error = HA_ADMIN_FAILED;
errmsg = "Failed in dsmObjectInfo (.gmi) (Error %d)";
errornum = rc;
gemini_msg(pcontext, errmsg ,errornum);
goto err;
}
rc = dsmAreaClose(pcontext, indexArea);
if (rc)
{
error = HA_ADMIN_FAILED;
errmsg = "Failed in dsmAreaClose (.gmi) (Error %d)";
errornum = rc;
gemini_msg(pcontext, errmsg ,errornum);
goto err;
}
/* Restore the index file */
if (!fn_format(src_path, table_name, backup_dir, ha_gemini_idx_ext, 4 + 64))
{
return HA_ADMIN_INVALID;
}
if (my_copy(src_path, fn_format(dst_path, table->path, "",
ha_gemini_idx_ext, 4), MYF(MY_WME)))
{
error = HA_ADMIN_FAILED;
errmsg = "Failed in my_copy (.gmi) (Error %d)";
errornum = errno;
gemini_msg(pcontext, errmsg ,errornum);
goto err;
}
rc = dsmAreaOpen(pcontext, indexArea, 1);
if (rc)
{
error = HA_ADMIN_FAILED;
errmsg = "Failed in dsmAreaOpen (.gmi) (Error %d)";
errornum = rc;
gemini_msg(pcontext, errmsg ,errornum);
goto err;
}
return HA_ADMIN_OK;
#else /* #ifdef GEMINI_BACKUP_IDX */
HA_CHECK_OPT tmp_check_opt;
tmp_check_opt.init();
/* The following aren't currently implemented in ha_gemini::repair
** tmp_check_opt.quick = 1;
** tmp_check_opt.flags |= T_VERY_SILENT;
*/
return (repair(thd, &tmp_check_opt));
#endif /* #ifdef GEMINI_BACKUP_IDX */
err:
{
#if 0
/* mi_check_print_error is in ha_myisam.cc, so none of the informative
** error messages above is currently being printed
*/
MI_CHECK param;
myisamchk_init(&param);
param.thd = thd;
param.op_name = (char*)"restore";
param.table_name = table->table_name;
param.testflag = 0;
mi_check_print_error(&param,errmsg, errornum);
#endif
return error;
}
}
int ha_gemini::backup(THD* thd, HA_CHECK_OPT *check_opt)
{
dsmContext_t *pcontext = (dsmContext_t *)thd->gemini.context;
char* backup_dir = thd->lex.backup_dir;
char src_path[FN_REFLEN], dst_path[FN_REFLEN];
char* table_name = table->real_name;
int error = 0;
int errornum;
const char* errmsg = "";
dsmArea_t tableArea = 0;
dsmObjectAttr_t objectAttr;
dsmObject_t associate;
dsmObjectType_t associateType;
dsmDbkey_t block, root;
dsmStatus_t rc;
rc = dsmObjectInfo(pcontext, tableNumber, DSMOBJECT_MIXTABLE, tableNumber,
&tableArea, &objectAttr, &associateType, &block, &root);
if (rc)
{
error = HA_ADMIN_FAILED;
errmsg = "Failed in dsmObjectInfo (.gmd) (Error %d)";
errornum = rc;
goto err;
}
/* Flush the buffers before backing up the table */
dsmAreaFlush((dsmContext_t *)thd->gemini.context, tableArea,
FLUSH_BUFFERS | FLUSH_SYNC);
if (rc)
{
error = HA_ADMIN_FAILED;
errmsg = "Failed in dsmAreaFlush (.gmd) (Error %d)";
errornum = rc;
gemini_msg(pcontext, errmsg ,errornum);
goto err;
}
/* Backup the .FRM file */
if (!fn_format(dst_path, table_name, backup_dir, reg_ext, 4 + 64))
{
errmsg = "Failed in fn_format() for .frm file: errno = %d";
error = HA_ADMIN_INVALID;
errornum = errno;
gemini_msg(pcontext, errmsg ,errornum);
goto err;
}
if (my_copy(fn_format(src_path, table->path,"", reg_ext, 4),
dst_path,
MYF(MY_WME | MY_HOLD_ORIGINAL_MODES )))
{
error = HA_ADMIN_FAILED;
errmsg = "Failed copying .frm file: errno = %d";
errornum = errno;
gemini_msg(pcontext, errmsg ,errornum);
goto err;
}
/* Backup the data file */
if (!fn_format(dst_path, table_name, backup_dir, ha_gemini_ext, 4 + 64))
{
errmsg = "Failed in fn_format() for .GMD file: errno = %d";
error = HA_ADMIN_INVALID;
errornum = errno;
gemini_msg(pcontext, errmsg ,errornum);
goto err;
}
if (my_copy(fn_format(src_path, table->path,"", ha_gemini_ext, 4),
dst_path,
MYF(MY_WME | MY_HOLD_ORIGINAL_MODES )) )
{
errmsg = "Failed copying .GMD file: errno = %d";
error= HA_ADMIN_FAILED;
errornum = errno;
gemini_msg(pcontext, errmsg ,errornum);
goto err;
}
#ifdef GEMINI_BACKUP_IDX
dsmArea_t indexArea = 0;
rc = dsmObjectInfo(pcontext, tableNumber, DSMOBJECT_MIXINDEX, &indexArea,
&objectAttr, &associate, &associateType, &block, &root);
if (rc)
{
error = HA_ADMIN_FAILED;
errmsg = "Failed in dsmObjectInfo (.gmi) (Error %d)";
errornum = rc;
gemini_msg(pcontext, errmsg ,errornum);
goto err;
}
/* Backup the index file */
if (!fn_format(dst_path, table_name, backup_dir, ha_gemini_idx_ext, 4 + 64))
{
errmsg = "Failed in fn_format() for .GMI file: errno = %d";
error = HA_ADMIN_INVALID;
errornum = errno;
gemini_msg(pcontext, errmsg ,errornum);
goto err;
}
if (my_copy(fn_format(src_path, table->path,"", ha_gemini_idx_ext, 4),
dst_path,
MYF(MY_WME | MY_HOLD_ORIGINAL_MODES )) )
{
errmsg = "Failed copying .GMI file: errno = %d";
error= HA_ADMIN_FAILED;
errornum = errno;
gemini_msg(pcontext, errmsg ,errornum);
goto err;
}
#endif /* #ifdef GEMINI_BACKUP_IDX */
return HA_ADMIN_OK;
err:
{
#if 0
/* mi_check_print_error is in ha_myisam.cc, so none of the informative
** error messages above is currently being printed
*/
MI_CHECK param;
myisamchk_init(&param);
param.thd = thd;
param.op_name = (char*)"backup";
param.table_name = table->table_name;
param.testflag = 0;
mi_check_print_error(&param,errmsg, errornum);
#endif
return error;
}
}
int ha_gemini::optimize(THD* thd, HA_CHECK_OPT *check_opt)
{
return HA_ADMIN_ALREADY_DONE;
}
ha_rows ha_gemini::records_in_range(int keynr,
const byte *start_key,uint start_key_len,
enum ha_rkey_function start_search_flag,
......@@ -2412,7 +3117,7 @@ ha_rows ha_gemini::records_in_range(int keynr,
pbracketBase->keyLen = componentLen;
}
pbracketBase->keyLen -= 3;
pbracketBase->keyLen -= FULLKEYHDRSZ;
if(end_key)
{
......@@ -2431,9 +3136,10 @@ ha_rows ha_gemini::records_in_range(int keynr,
pbracketLimit->keyLen = componentLen;
}
pbracketLimit->keyLen -= 3;
pbracketLimit->keyLen -= FULLKEYHDRSZ;
error = dsmIndexRowsInRange((dsmContext_t *)current_thd->gemini.context,
pbracketBase,pbracketLimit,
tableNumber,
&pctInrange);
if(pctInrange >= 1)
rows = (ha_rows)pctInrange;
......@@ -2457,32 +3163,82 @@ ha_rows ha_gemini::records_in_range(int keynr,
may only happen in rows with blobs, as the default row length is
pre-allocated.
*/
int ha_gemini::pack_row(byte **pprow, int *ppackedLength, const byte *record)
int ha_gemini::pack_row(byte **pprow, int *ppackedLength, const byte *record,
bool update)
{
THD *thd = current_thd;
dsmContext_t *pcontext = (dsmContext_t *)thd->gemini.context;
gemBlobDesc_t *pBlobDesc = pBlobDescs;
if (fixed_length_row)
{
*pprow = (byte *)record;
*ppackedLength=(int)table->reclength;
return 0;
}
if (table->blob_fields)
{
return HA_ERR_WRONG_COMMAND;
}
/* Copy null bits */
memcpy(rec_buff, record, table->null_bytes);
byte *ptr=rec_buff + table->null_bytes;
for (Field **field=table->field ; *field ; field++)
ptr=(byte*) (*field)->pack((char*) ptr,record + (*field)->offset());
{
#ifdef GEMINI_TINYBLOB_IN_ROW
/* Tiny blobs (255 bytes or less) are stored in the row; larger
** blobs are stored in a separate storage object (see ha_gemini::create).
*/
if ((*field)->type() == FIELD_TYPE_BLOB &&
((Field_blob*)*field)->blobtype() != FIELD_TYPE_TINY_BLOB)
#else
if ((*field)->type() == FIELD_TYPE_BLOB)
#endif
{
dsmBlob_t gemBlob;
char *blobptr;
gemBlob.areaType = DSMOBJECT_BLOB;
gemBlob.blobObjNo = tableNumber;
gemBlob.blobId = 0;
gemBlob.totLength = gemBlob.segLength =
((Field_blob*)*field)->get_length((char*)record + (*field)->offset());
((Field_blob*)*field)->get_ptr((char**) &blobptr);
gemBlob.pBuffer = (dsmBuffer_t *)blobptr;
gemBlob.blobContext.blobOffset = 0;
if (gemBlob.totLength)
{
dsmBlobStart(pcontext, &gemBlob);
if (update && pBlobDesc->blobId)
{
gemBlob.blobId = pBlobDesc->blobId;
dsmBlobUpdate(pcontext, &gemBlob, NULL);
}
else
{
dsmBlobPut(pcontext, &gemBlob, NULL);
}
dsmBlobEnd(pcontext, &gemBlob);
}
ptr = (byte*)((Field_blob*)*field)->pack_id((char*) ptr,
(char*)record + (*field)->offset(), (longlong)gemBlob.blobId);
pBlobDesc++;
}
else
{
ptr=(byte*) (*field)->pack((char*) ptr, (char*)record + (*field)->offset());
}
}
*pprow=rec_buff;
*ppackedLength= (ptr - rec_buff);
return 0;
}
void ha_gemini::unpack_row(char *record, char *prow)
int ha_gemini::unpack_row(char *record, char *prow)
{
THD *thd = current_thd;
dsmContext_t *pcontext = (dsmContext_t *)thd->gemini.context;
gemBlobDesc_t *pBlobDesc = pBlobDescs;
if (fixed_length_row)
{
/* If the table is a VST, the row is in Gemini internal format.
......@@ -2568,38 +3324,129 @@ void ha_gemini::unpack_row(char *record, char *prow)
const char *ptr= (const char*) prow;
memcpy(record, ptr, table->null_bytes);
ptr+=table->null_bytes;
for (Field **field=table->field ; *field ; field++)
ptr= (*field)->unpack(record + (*field)->offset(), ptr);
{
#ifdef GEMINI_TINYBLOB_IN_ROW
/* Tiny blobs (255 bytes or less) are stored in the row; larger
** blobs are stored in a separate storage object (see ha_gemini::create).
*/
if ((*field)->type() == FIELD_TYPE_BLOB &&
((Field_blob*)*field)->blobtype() != FIELD_TYPE_TINY_BLOB)
#else
if ((*field)->type() == FIELD_TYPE_BLOB)
#endif
{
dsmBlob_t gemBlob;
gemBlob.areaType = DSMOBJECT_BLOB;
gemBlob.blobObjNo = tableNumber;
gemBlob.blobId = (dsmBlobId_t)(((Field_blob*)*field)->get_id(ptr));
if (gemBlob.blobId)
{
gemBlob.totLength =
gemBlob.segLength = ((Field_blob*)*field)->get_length(ptr);
/* Allocate memory to store the blob. This memory is freed
** the next time unpack_row is called for this table.
*/
gemBlob.pBuffer = (dsmBuffer_t *)my_malloc(gemBlob.totLength,
MYF(0));
if (!gemBlob.pBuffer)
{
return HA_ERR_OUT_OF_MEM;
}
gemBlob.blobContext.blobOffset = 0;
dsmBlobStart(pcontext, &gemBlob);
dsmBlobGet(pcontext, &gemBlob, NULL);
dsmBlobEnd(pcontext, &gemBlob);
}
else
{
gemBlob.pBuffer = 0;
}
ptr = ((Field_blob*)*field)->unpack_id(record + (*field)->offset(),
ptr, (char *)gemBlob.pBuffer);
pBlobDesc->blobId = gemBlob.blobId;
my_free((char*)pBlobDesc->pBlob, MYF(MY_ALLOW_ZERO_PTR));
pBlobDesc->pBlob = gemBlob.pBuffer;
pBlobDesc++;
}
else
{
ptr= (*field)->unpack(record + (*field)->offset(), ptr);
}
}
}
return 0;
}
int ha_gemini::key_cmp(uint keynr, const byte * old_row,
const byte * new_row)
const byte * new_row, bool updateStats)
{
KEY_PART_INFO *key_part=table->key_info[keynr].key_part;
KEY_PART_INFO *end=key_part+table->key_info[keynr].key_parts;
for ( ; key_part != end ; key_part++)
for ( uint i = 0 ; key_part != end ; key_part++, i++)
{
if (key_part->null_bit)
{
if ((old_row[key_part->null_offset] & key_part->null_bit) !=
(new_row[key_part->null_offset] & key_part->null_bit))
{
if(updateStats)
table->key_info[keynr].rec_per_key[i]++;
return 1;
}
else if((old_row[key_part->null_offset] & key_part->null_bit) &&
(new_row[key_part->null_offset] & key_part->null_bit))
/* Both are null */
continue;
}
if (key_part->key_part_flag & (HA_BLOB_PART | HA_VAR_LENGTH))
{
if (key_part->field->cmp_binary(old_row + key_part->offset,
new_row + key_part->offset,
if (key_part->field->cmp_binary((char*)(old_row + key_part->offset),
(char*)(new_row + key_part->offset),
(ulong) key_part->length))
{
if(updateStats)
table->key_info[keynr].rec_per_key[i]++;
return 1;
}
}
else
{
if (memcmp(old_row+key_part->offset, new_row+key_part->offset,
key_part->length))
{
/* Check for special case of -0 which causes table check
to find an invalid key when comparing the the index
value of 0 to the -0 stored in the row */
if(key_part->field->type() == FIELD_TYPE_DECIMAL)
{
double fieldValue;
char *ptr = key_part->field->ptr;
key_part->field->ptr = (char *)old_row + key_part->offset;
fieldValue = key_part->field->val_real();
if(fieldValue == 0)
{
key_part->field->ptr = (char *)new_row + key_part->offset;
fieldValue = key_part->field->val_real();
if(fieldValue == 0)
{
key_part->field->ptr = ptr;
continue;
}
}
key_part->field->ptr = ptr;
}
if(updateStats)
{
table->key_info[keynr].rec_per_key[i]++;
}
return 1;
}
}
}
return 0;
......@@ -2612,13 +3459,13 @@ int gemini_parse_table_name(const char *fullname, char *dbname, char *tabname)
/* separate out the name of the table and the database
*/
namestart = strchr(fullname + start_of_name, '/');
namestart = (char *)strchr(fullname + start_of_name, '/');
if (!namestart)
{
/* if on Windows, slashes go the other way */
namestart = strchr(fullname + start_of_name, '\\');
namestart = (char *)strchr(fullname + start_of_name, '\\');
}
nameend = strchr(fullname + start_of_name, '.');
nameend = (char *)strchr(fullname + start_of_name, '.');
/* sometimes fullname has an extension, sometimes it doesn't */
if (!nameend)
{
......@@ -2680,4 +3527,105 @@ static void print_msg(THD *thd, const char *table_name, const char *op_name,
thd->killed=1;
}
/* Load shared area with rows per key statistics */
void
ha_gemini::get_index_stats(THD *thd)
{
dsmStatus_t rc = 0;
ha_rows *rec_per_key = share->rec_per_key;
for(uint i = 0; i < table->keys && !rc; i++)
{
for (uint j = 0; j < table->key_info[i].key_parts && !rc;j++)
{
LONG64 rows_per_key;
rc = dsmIndexStatsGet((dsmContext_t *)thd->gemini.context,
tableNumber, pindexNumbers[i],(int)j,
&rows_per_key);
if (rc)
{
gemini_msg((dsmContext_t *)thd->gemini.context,
"Index Statistics faild for table %d index %d, error %d",
tableNumber, pindexNumbers[i], rc);
}
*rec_per_key = (ha_rows)rows_per_key;
rec_per_key++;
}
}
return;
}
/****************************************************************************
Handling the shared GEM_SHARE structure that is needed to provide
a global in memory storage location of the rec_per_key stats used
by the optimizer.
****************************************************************************/
static byte* gem_get_key(GEM_SHARE *share,uint *length,
my_bool not_used __attribute__((unused)))
{
*length=share->table_name_length;
return (byte*) share->table_name;
}
static GEM_SHARE *get_share(const char *table_name, TABLE *table)
{
GEM_SHARE *share;
pthread_mutex_lock(&gem_mutex);
uint length=(uint) strlen(table_name);
if (!(share=(GEM_SHARE*) hash_search(&gem_open_tables, (byte*) table_name,
length)))
{
ha_rows *rec_per_key;
char *tmp_name;
if ((share=(GEM_SHARE *)
my_multi_malloc(MYF(MY_WME | MY_ZEROFILL),
&share, sizeof(*share),
&rec_per_key, table->key_parts * sizeof(ha_rows),
&tmp_name, length+1,
NullS)))
{
share->rec_per_key = rec_per_key;
share->table_name = tmp_name;
share->table_name_length=length;
strcpy(share->table_name,table_name);
if (hash_insert(&gem_open_tables, (byte*) share))
{
pthread_mutex_unlock(&gem_mutex);
my_free((gptr) share,0);
return 0;
}
thr_lock_init(&share->lock);
pthread_mutex_init(&share->mutex,NULL);
}
}
pthread_mutex_unlock(&gem_mutex);
return share;
}
static int free_share(GEM_SHARE *share, bool mutex_is_locked)
{
pthread_mutex_lock(&gem_mutex);
if (mutex_is_locked)
pthread_mutex_unlock(&share->mutex);
if (!--share->use_count)
{
hash_delete(&gem_open_tables, (byte*) share);
thr_lock_delete(&share->lock);
pthread_mutex_destroy(&share->mutex);
my_free((gptr) share, MYF(0));
}
pthread_mutex_unlock(&gem_mutex);
return 0;
}
static void gemini_lock_table_overflow_error(dsmContext_t *pcontext)
{
gemini_msg(pcontext, "The total number of locks exceeds the lock table size");
gemini_msg(pcontext, "Either increase gemini_lock_table_size or use a");
gemini_msg(pcontext, "different transaction isolation level");
}
#endif /* HAVE_GEMINI_DB */
......@@ -19,17 +19,26 @@
#pragma interface /* gcc class implementation */
#endif
#include "gem_global.h"
#include "dstd.h"
#include "dsmpub.h"
/* class for the the gemini handler */
enum enum_key_string_options{KEY_CREATE,KEY_DELETE,KEY_CHECK};
#define READ_UNCOMMITED 0
#define READ_COMMITED 1
#define REPEATABLE_READ 2
#define SERIALIZEABLE 3
typedef struct st_gemini_share {
ha_rows *rec_per_key;
THR_LOCK lock;
pthread_mutex_t mutex;
char *table_name;
uint table_name_length,use_count;
} GEM_SHARE;
typedef struct gemBlobDesc
{
dsmBlobId_t blobId;
dsmBuffer_t *pBlob;
} gemBlobDesc_t;
class ha_gemini: public handler
{
......@@ -38,7 +47,7 @@ class ha_gemini: public handler
uint int_option_flag;
int tableNumber;
dsmIndex_t *pindexNumbers; // dsm object numbers for the indexes on this table
unsigned long lastRowid;
dsmRecid_t lastRowid;
uint last_dup_key;
bool fixed_length_row, key_read, using_ignore;
byte *rec_buff;
......@@ -46,10 +55,12 @@ class ha_gemini: public handler
dsmKey_t *pbracketLimit;
dsmKey_t *pfoundKey;
dsmMask_t tableStatus; // Crashed/repair status
gemBlobDesc_t *pBlobDescs;
int index_open(char *tableName);
int pack_row(byte **prow, int *ppackedLength, const byte *record);
void unpack_row(char *record, char *prow);
int pack_row(byte **prow, int *ppackedLength, const byte *record,
bool update);
int unpack_row(char *record, char *prow);
int findRow(THD *thd, dsmMask_t findMode, byte *buf);
int fetch_row(void *gemini_context, const byte *buf);
int handleIndexEntries(const byte * record, dsmRecid_t recid,
......@@ -70,24 +81,28 @@ class ha_gemini: public handler
void unpack_key(char *record, dsmKey_t *key, uint index);
int key_cmp(uint keynr, const byte * old_row,
const byte * new_row);
const byte * new_row, bool updateStats);
int saveKeyStats(THD *thd);
void get_index_stats(THD *thd);
short cursorId; /* cursorId of active index cursor if any */
dsmMask_t lockMode; /* Shared or exclusive */
/* FIXFIX Don't know why we need this because I don't know what
store_lock method does but we core dump without this */
THR_LOCK alock;
THR_LOCK_DATA lock;
GEM_SHARE *share;
public:
ha_gemini(TABLE *table): handler(table), file(0),
int_option_flag(HA_READ_NEXT | HA_READ_PREV |
HA_REC_NOT_IN_SEQ |
HA_KEYPOS_TO_RNDPOS | HA_READ_ORDER | HA_LASTKEY_ORDER |
HA_LONGLONG_KEYS | HA_NULL_KEY | HA_HAVE_KEY_READ_ONLY |
HA_NO_BLOBS | HA_NO_TEMP_TABLES |
/* HA_BLOB_KEY | */ /*HA_NOT_EXACT_COUNT | */
HA_BLOB_KEY |
HA_NO_TEMP_TABLES | HA_NO_FULLTEXT_KEY |
/*HA_NOT_EXACT_COUNT | */
/*HA_KEY_READ_WRONG_STR |*/ HA_DROP_BEFORE_CREATE),
pbracketBase(0),pbracketLimit(0),pfoundKey(0),
cursorId(0)
......@@ -100,7 +115,7 @@ class ha_gemini: public handler
uint max_record_length() const { return MAXRECSZ; }
uint max_keys() const { return MAX_KEY-1; }
uint max_key_parts() const { return MAX_REF_PARTS; }
uint max_key_length() const { return MAXKEYSZ; }
uint max_key_length() const { return MAXKEYSZ / 2; }
bool fast_key_read() { return 1;}
bool has_transactions() { return 1;}
......@@ -129,8 +144,12 @@ class ha_gemini: public handler
void info(uint);
int extra(enum ha_extra_function operation);
int reset(void);
int analyze(THD* thd, HA_CHECK_OPT* check_opt);
int check(THD* thd, HA_CHECK_OPT* check_opt);
int repair(THD* thd, HA_CHECK_OPT* check_opt);
int restore(THD* thd, HA_CHECK_OPT* check_opt);
int backup(THD* thd, HA_CHECK_OPT* check_opt);
int optimize(THD* thd, HA_CHECK_OPT* check_opt);
int external_lock(THD *thd, int lock_type);
virtual longlong get_auto_increment();
void position(byte *record);
......@@ -139,7 +158,7 @@ class ha_gemini: public handler
enum ha_rkey_function start_search_flag,
const byte *end_key,uint end_key_len,
enum ha_rkey_function end_search_flag);
void update_create_info(HA_CREATE_INFO *create_info);
int create(const char *name, register TABLE *form,
HA_CREATE_INFO *create_info);
int delete_table(const char *name);
......@@ -167,6 +186,7 @@ extern long gemini_locktablesize;
extern long gemini_lock_wait_timeout;
extern long gemini_spin_retries;
extern long gemini_connection_limit;
extern char *gemini_basedir;
extern TYPELIB gemini_recovery_typelib;
extern ulong gemini_recovery_options;
......@@ -175,12 +195,13 @@ bool gemini_end(void);
bool gemini_flush_logs(void);
int gemini_commit(THD *thd);
int gemini_rollback(THD *thd);
int gemini_recovery_logging(THD *thd, bool on);
void gemini_disconnect(THD *thd);
int gemini_rollback_to_savepoint(THD *thd);
int gemini_parse_table_name(const char *fullname, char *dbname, char *tabname);
int gemini_is_vst(const char *pname);
int gemini_set_option_long(int optid, long optval);
const int gemini_blocksize = 8192;
const int gemini_recbits = 7;
const int gemini_blocksize = BLKSIZE;
const int gemini_recbits = DEFAULT_RECBITS;
......@@ -694,6 +694,15 @@ void handler::print_error(int error, myf errflag)
case HA_ERR_RECORD_FILE_FULL:
textno=ER_RECORD_FILE_FULL;
break;
case HA_ERR_LOCK_WAIT_TIMEOUT:
textno=ER_LOCK_WAIT_TIMEOUT;
break;
case HA_ERR_LOCK_TABLE_FULL:
textno=ER_LOCK_TABLE_FULL;
break;
case HA_ERR_READ_ONLY_TRANSACTION:
textno=ER_READ_ONLY_TRANSACTION;
break;
default:
{
my_error(ER_GET_ERRNO,errflag,error);
......@@ -757,6 +766,25 @@ int ha_commit_rename(THD *thd)
return error;
}
/* Tell the handler to turn on or off logging to the handler's
recovery log
*/
int ha_recovery_logging(THD *thd, bool on)
{
int error=0;
DBUG_ENTER("ha_recovery_logging");
#ifdef USING_TRANSACTIONS
if (opt_using_transactions)
{
#ifdef HAVE_GEMINI_DB
error = gemini_recovery_logging(thd, on);
}
#endif
#endif
DBUG_RETURN(error);
}
int handler::index_next_same(byte *buf, const byte *key, uint keylen)
{
int error;
......
......@@ -74,6 +74,7 @@
#define HA_NOT_DELETE_WITH_CACHE (HA_NOT_READ_AFTER_KEY*2)
#define HA_NO_TEMP_TABLES (HA_NOT_DELETE_WITH_CACHE*2)
#define HA_NO_PREFIX_CHAR_KEYS (HA_NO_TEMP_TABLES*2)
#define HA_NO_FULLTEXT_KEY (HA_NO_PREFIX_CHAR_KEYS*2)
/* Parameters for open() (in register form->filestat) */
/* HA_GET_INFO does a implicit HA_ABORT_IF_LOCKED */
......@@ -353,3 +354,4 @@ int ha_autocommit_or_rollback(THD *thd, int error);
void ha_set_spin_retries(uint retries);
bool ha_flush_logs(void);
int ha_commit_rename(THD *thd);
int ha_recovery_logging(THD *thd, bool on);
......@@ -35,6 +35,7 @@ static MYSQL_LOCK *get_lock_data(THD *thd, TABLE **table,uint count,
bool unlock, TABLE **write_locked);
static int lock_external(TABLE **table,uint count);
static int unlock_external(THD *thd, TABLE **table,uint count);
static void print_lock_error(int error);
MYSQL_LOCK *mysql_lock_tables(THD *thd,TABLE **tables,uint count)
......@@ -154,7 +155,7 @@ static int lock_external(TABLE **tables,uint count)
(*tables)->file->external_lock(thd, F_UNLCK);
(*tables)->current_lock=F_UNLCK;
}
my_error(ER_CANT_LOCK,MYF(ME_BELL+ME_OLDWIN+ME_WAITTANG),error);
print_lock_error(error);
DBUG_RETURN(error);
}
else
......@@ -325,7 +326,7 @@ static int unlock_external(THD *thd, TABLE **table,uint count)
}
}
if (error_code)
my_error(ER_CANT_LOCK,MYF(ME_BELL+ME_OLDWIN+ME_WAITTANG),error_code);
print_lock_error(error_code);
DBUG_RETURN(error_code);
}
......@@ -480,3 +481,24 @@ bool wait_for_locked_table_names(THD *thd, TABLE_LIST *table_list)
}
DBUG_RETURN(result);
}
static void print_lock_error(int error)
{
int textno;
DBUG_ENTER("print_lock_error");
switch (error) {
case HA_ERR_LOCK_WAIT_TIMEOUT:
textno=ER_LOCK_WAIT_TIMEOUT;
break;
case HA_ERR_READ_ONLY_TRANSACTION:
textno=ER_READ_ONLY_TRANSACTION;
break;
default:
textno=ER_CANT_LOCK;
break;
}
my_error(textno,MYF(ME_BELL+ME_OLDWIN+ME_WAITTANG),error);
DBUG_VOID_RETURN;
}
......@@ -215,3 +215,6 @@
"Could not create slave thread, check system resources",
"User %-.64s has already more than 'max_user_connections' active connections",
"You may only use constant expressions with SET",
"Lock wait timeout exceeded",
"The total number of locks exceeds the lock table size",
"Update locks cannot be acquired during a READ UNCOMMITTED transaction",
......@@ -209,3 +209,6 @@
"Could not create slave thread, check system resources",
"User %-.64s has already more than 'max_user_connections' active connections",
"You may only use constant expressions with SET",
"Lock wait timeout exceeded",
"The total number of locks exceeds the lock table size",
"Update locks cannot be acquired during a READ UNCOMMITTED transaction",
......@@ -206,3 +206,6 @@
"Could not create slave thread, check system resources",
"User %-.64s has already more than 'max_user_connections' active connections",
"You may only use constant expressions with SET",
"Lock wait timeout exceeded",
"The total number of locks exceeds the lock table size",
"Update locks cannot be acquired during a READ UNCOMMITTED transaction",
......@@ -206,3 +206,6 @@
"Could not create slave thread, check system resources",
"User %-.64s has already more than 'max_user_connections' active connections",
"You may only use constant expressions with SET",
"Lock wait timeout exceeded",
"The total number of locks exceeds the lock table size",
"Update locks cannot be acquired during a READ UNCOMMITTED transaction",
......@@ -210,3 +210,6 @@
"Could not create slave thread, check system resources",
"User %-.64s has already more than 'max_user_connections' active connections",
"You may only use constant expressions with SET",
"Lock wait timeout exceeded",
"The total number of locks exceeds the lock table size",
"Update locks cannot be acquired during a READ UNCOMMITTED transaction",
......@@ -206,3 +206,6 @@
"Could not create slave thread, check system resources",
"User %-.64s has already more than 'max_user_connections' active connections",
"You may only use constant expressions with SET",
"Lock wait timeout exceeded",
"The total number of locks exceeds the lock table size",
"Update locks cannot be acquired during a READ UNCOMMITTED transaction",
......@@ -209,3 +209,6 @@
"Could not create slave thread, check system resources",
"User %-.64s has already more than 'max_user_connections' active connections",
"You may only use constant expressions with SET",
"Lock wait timeout exceeded",
"The total number of locks exceeds the lock table size",
"Update locks cannot be acquired during a READ UNCOMMITTED transaction",
......@@ -206,3 +206,6 @@
"Could not create slave thread, check system resources",
"User %-.64s has already more than 'max_user_connections' active connections",
"You may only use constant expressions with SET",
"Lock wait timeout exceeded",
"The total number of locks exceeds the lock table size",
"Update locks cannot be acquired during a READ UNCOMMITTED transaction",
......@@ -208,3 +208,6 @@
"Could not create slave thread, check system resources",
"User %-.64s has already more than 'max_user_connections' active connections",
"You may only use constant expressions with SET",
"Lock wait timeout exceeded",
"The total number of locks exceeds the lock table size",
"Update locks cannot be acquired during a READ UNCOMMITTED transaction",
......@@ -206,3 +206,6 @@
"Could not create slave thread, check system resources",
"User %-.64s has already more than 'max_user_connections' active connections",
"You may only use constant expressions with SET",
"Lock wait timeout exceeded",
"The total number of locks exceeds the lock table size",
"Update locks cannot be acquired during a READ UNCOMMITTED transaction",
......@@ -208,3 +208,6 @@
"Could not create slave thread, check system resources",
"User %-.64s has already more than 'max_user_connections' active connections",
"You may only use constant expressions with SET",
"Lock wait timeout exceeded",
"The total number of locks exceeds the lock table size",
"Update locks cannot be acquired during a READ UNCOMMITTED transaction",
......@@ -206,3 +206,6 @@
"Could not create slave thread, check system resources",
"User %-.64s has already more than 'max_user_connections' active connections",
"You may only use constant expressions with SET",
"Lock wait timeout exceeded",
"The total number of locks exceeds the lock table size",
"Update locks cannot be acquired during a READ UNCOMMITTED transaction",
......@@ -208,3 +208,6 @@
"Could not create slave thread, check system resources",
"User %-.64s has already more than 'max_user_connections' active connections",
"You may only use constant expressions with SET",
"Lock wait timeout exceeded",
"The total number of locks exceeds the lock table size",
"Update locks cannot be acquired during a READ UNCOMMITTED transaction",
......@@ -208,3 +208,6 @@
"Could not create slave thread, check system resources",
"User %-.64s has already more than 'max_user_connections' active connections",
"You may only use constant expressions with SET",
"Lock wait timeout exceeded",
"The total number of locks exceeds the lock table size",
"Update locks cannot be acquired during a READ UNCOMMITTED transaction",
......@@ -210,3 +210,6 @@
"Could not create slave thread, check system resources",
"User %-.64s has already more than 'max_user_connections' active connections",
"You may only use constant expressions with SET",
"Lock wait timeout exceeded",
"The total number of locks exceeds the lock table size",
"Update locks cannot be acquired during a READ UNCOMMITTED transaction",
......@@ -206,3 +206,6 @@
"Could not create slave thread, check system resources",
"User %-.64s has already more than 'max_user_connections' active connections",
"You may only use constant expressions with SET",
"Lock wait timeout exceeded",
"The total number of locks exceeds the lock table size",
"Update locks cannot be acquired during a READ UNCOMMITTED transaction",
......@@ -210,3 +210,6 @@
"Could not create slave thread, check system resources",
"User %-.64s has already more than 'max_user_connections' active connections",
"You may only use constant expressions with SET",
"Lock wait timeout exceeded",
"The total number of locks exceeds the lock table size",
"Update locks cannot be acquired during a READ UNCOMMITTED transaction",
......@@ -209,3 +209,6 @@
" SLAVE, ",
"User %-.64s has already more than 'max_user_connections' active connections",
"You may only use constant expressions with SET",
"Lock wait timeout exceeded",
"The total number of locks exceeds the lock table size",
"Update locks cannot be acquired during a READ UNCOMMITTED transaction",
......@@ -214,3 +214,6 @@
"Could not create slave thread, check system resources",
"User %-.64s has already more than 'max_user_connections' active connections",
"You may only use constant expressions with SET",
"Lock wait timeout exceeded",
"The total number of locks exceeds the lock table size",
"Update locks cannot be acquired during a READ UNCOMMITTED transaction",
......@@ -207,3 +207,6 @@
"Could not create slave thread, check system resources",
"User %-.64s has already more than 'max_user_connections' active connections",
"You may only use constant expressions with SET",
"Lock wait timeout exceeded",
"The total number of locks exceeds the lock table size",
"Update locks cannot be acquired during a READ UNCOMMITTED transaction",
......@@ -206,3 +206,6 @@
"Kunde inte starta en tråd för replikering",
"Användare '%-.64s' har redan 'max_user_connections' aktiva inloggningar",
"Du kan endast använda konstant-uttryck med SET",
"Lock wait timeout exceeded",
"The total number of locks exceeds the lock table size",
"Update locks cannot be acquired during a READ UNCOMMITTED transaction",
......@@ -1388,11 +1388,6 @@ TABLE *open_ltable(THD *thd, TABLE_LIST *table_list, thr_lock_type lock_type)
bool refresh;
DBUG_ENTER("open_ltable");
#ifdef __WIN__
/* Win32 can't drop a file that is open */
if (lock_type == TL_WRITE_ALLOW_READ)
lock_type= TL_WRITE;
#endif
thd->proc_info="Opening table";
while (!(table=open_table(thd,table_list->db ? table_list->db : thd->db,
table_list->real_name,table_list->name,
......@@ -1400,6 +1395,19 @@ TABLE *open_ltable(THD *thd, TABLE_LIST *table_list, thr_lock_type lock_type)
if (table)
{
int error;
#ifdef __WIN__
/* Win32 can't drop a file that is open */
if (lock_type == TL_WRITE_ALLOW_READ
#ifdef HAVE_GEMINI_DB
&& table->db_type != DB_TYPE_GEMINI
#endif /* HAVE_GEMINI_DB */
)
{
lock_type= TL_WRITE;
}
#endif /* __WIN__ */
table_list->table=table;
table->grant= table_list->grant;
if (thd->locked_tables)
......
......@@ -423,6 +423,13 @@ int mysql_create_table(THD *thd,const char *db, const char *table_name,
column->field_name);
DBUG_RETURN(-1);
}
if (key->type == Key::FULLTEXT &&
(file->option_flag() & HA_NO_FULLTEXT_KEY))
{
my_printf_error(ER_WRONG_KEY_COLUMN, ER(ER_WRONG_KEY_COLUMN), MYF(0),
column->field_name);
DBUG_RETURN(-1);
}
if (f_is_blob(sql_field->pack_flag))
{
if (!(file->option_flag() & HA_BLOB_KEY))
......@@ -1678,6 +1685,16 @@ copy_data_between_tables(TABLE *from,TABLE *to,
goto err;
};
/* Turn off recovery logging since rollback of an
alter table is to delete the new table so there
is no need to log the changes to it. */
error = ha_recovery_logging(thd,false);
if(error)
{
error = 1;
goto err;
}
init_read_record(&info, thd, from, (SQL_SELECT *) 0, 1,1);
if (handle_duplicates == DUP_IGNORE ||
handle_duplicates == DUP_REPLACE)
......@@ -1723,6 +1740,7 @@ copy_data_between_tables(TABLE *from,TABLE *to,
if (to->file->activate_all_index(thd))
error=1;
tmp_error = ha_recovery_logging(thd,true);
/*
Ensure that the new table is saved properly to disk so that we
can do a rename
......@@ -1734,6 +1752,7 @@ copy_data_between_tables(TABLE *from,TABLE *to,
if (to->file->external_lock(thd,F_UNLCK))
error=1;
err:
tmp_error = ha_recovery_logging(thd,true);
free_io_cache(from);
*copied= found_count;
*deleted=delete_count;
......
Markdown is supported
0%
or
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment