← Back to team overview

drizzle-discuss team mailing list archive

Re: Improving the Engine API

 

Hi Jay,

On Dec 7, 2009, at 8:54 PM, Jay Pipes wrote:

Paul McCullagh wrote:
Hi Jay,
Yes, how ALTER TABLE, TRUNCATE, CHECK, etc is handled does depend (almost entirely) on the engine. I actually think that, for the moment anyway, the statement type will be sufficient. Maybe I have misunderstood what you mean by pluggable statements, but lets assume some engine adds a custom statement: "ENCRYPT TABLE". My engine will not know what type of lock to acquire for this statement, but on the other hand, it will also not know how to execute the statement. So it must reject the statement outright. Which means an engine needs a list of statements it supports, and everything else must be rejected. So the engine needs to know the statement type. I think it is important that we distinguish between statements and operations. In general, MySQL implements the ALTER TABLE using 3 operations: create-table, rename-table and delete-table.

Actually, it uses at least 4 :)

create-new-table-definition
copy-table-data
rename-table
delete-old-table

But...the above *operations* is only how *MySQL/Drizzle kernel* does an ALTER TABLE. This is *not* how InnoDB does an ALTER TABLE. It actually constructs a stored procedure (InnoDB internal stored procedure, not a MySQL stored procedure) and executes that stored procedure...

Hmmm, interesting.

OK, I think we agree on this, but to sum up, here are the best arguments for providing both methods (create/copy/rename/drop method & the engine handles the entire operation itself):

- The create/copy/rename/drop method must be used to alter the table engine. - It saves engine developers time because they only have to implement the basic operations to support ALTER TABLE. - Engines can concentrate on optimizing certain operations (e.g. add index) without having to implement the entire ALTER TABLE.

No matter what method the engine chooses. The server should still just replicate the GPB statement data (as you suggest in https://lists.launchpad.net/drizzle-discuss/msg05305.html) .

This basically means that the statement is replicated, and not the operations.

The cool thing is: if my engine can execute these 3 operations, then it can perform all ALTER TABLE statements, although the implementation is not optimal.

Right, because in that case the kernel is deciding what the *operations* should be instead of the engine, which knows its own shortcuts. This is why an earlier email of mine suggested that for ALTER TABLE, instead of constructing a GPB message of operations/ instructions to pass to the engine, we instead just pass a before and an after image of the table definition and leave it up to the engine to do its thing:

https://lists.launchpad.net/drizzle-discuss/msg05305.html

Yes, excellent, it make great sense to use exactly the same GPB that is prepared for the replication!

In this way the engine has easy access to all the information it needs, no parsing required (bad memories of the FK implementation return)!

> To implement this better, the engine needs to support a
custom alter-table operation. But the engine's alter-table operation may, for example, only support ADD INDEX. So basically, the engine examines the statement in startStatement(), add decides on one of the following:
- Use the default execution method for the statement
- Use the engine custom statement method
- Reject the statement as not implemented

OK, sure.

However, what should the API look like? I strongly prefer passing GPB protobuffer messages in the APIs, over pointers to internal structures. GPB is naturally able to version itself, providing a layer of insulation for adding future (pluggable) statement types...

Absolutely. GPB is one of the best features of Drizzle! :)

What if there was the following API? Would this API provide your engine with all the information that you would need?

Assume message::Statement is the same/similar to the message::Statement defined in /drizzled/message/transaction.proto:

http://bazaar.launchpad.net/~drizzle-developers/drizzle/development/annotate/head%3A/drizzled/message/transaction.proto

class plugin::StorageEngine ...
{
public:
/**
* returned from startStatement().  This tells the Drizzle
* kernel what the engine's intent is for executing this particular
* statement or type of statement
*/
enum StatementExecutionIntent
{
DEFAULT= 0, /* "Normal" or default execution. Defer to the kernel...
 INTERNAL= 1, /* Engine will handle all locking and execution */
 UNKNOWN_STATEMENT= 99 /* Engine has no clue what the statement is */
};

/**
* Indicate that the Session is starting a new SQL Statement, and
* provide the engine with information about the statement.  This
* call would come shortly after the parser identifies the type of
* statement to be executed.
*
* @param[in] The Session which is executing this statement
* @param[in] GPB Message of type message::Statement
*/
enum StatementExecutionIntent
StorageEngine::startStatement(Session &session,
                             message::Statement &statement);

Yes.

One thing to note here is that if 2 (or more) engines are involved in the statement, then all get the startStatement() call.

So when ALTER TABLE is used to change the engine, both engines get the startStatement() call. Of course, both engines should recognize the fact that they cannot handle the statement internally (because before and after images have different engines) and return DEFAULT.

/**
* Let the engine execute the statement itself.
*
* @note
*
* Instead of returning a simple bool, we could have a StatementResult
* enum or something like that...
*/
bool passthruStatement(Session &session,
                      message::Statement &statement);

more below...

It also decides (and maybe acquires) the locks it will need, and does any other statement initialization. So here is an example of the call sequence, if the engine uses the default method for alter table:
startStatement("ALTER TABLE", "t1") --> return: use default method
createTable("tmp")
// Copy all rows from "t1" to "tmp" using standard cursor operations
renameTable("t1", "tmp2") // t1 --> tmp2
renameTable("tmp", "t1") // tmp --> t1
dropTable("tmp2")
endStatement()
So Drizzle does not know anything about table locks. The engine knows it is in an ALTER TABLE statement, and knows (for example) that table t1 must be read locked when the row copy begins (i.e. when the cursor is opened), and that an exclusive lock is required to rename the table.

OK, to dig a little further into the API I am proposing, instead of the string-based API you are doing above, it would look something like this:

// in drizzle_parse(), a message::Statement would be constructed
// which contains something like this:

message::Statement *statement= new message::Statement();
statement->set_type(message::Statement::ALTER_TABLE);

message::AlterTableStatement *alter=
 statement->mutable_alter_table_statement();

message::Table *before_table= alter->mutable_before();
message::Table *after_table= alter->mutable_after();

// parser then fills in the before and after image of the
// message::Table definition...
//
// At this point, we notify the engine that we wish to start
// a new SQL statement for the ALTER TABLE t1 ...

plugin::StorageEngine *engine= getStorageEngineForTable("t1");
Session &session= getCurrentSession();

enum StorageEngine::StatementExecutionIntent intent=
 engine->startStatement(session, *statement);

// If the engine is happy to let Drizzle deal with the statement,
// follow a "normal" kernel path for this statement...
//
// If the engine wants to handle this statement, let it...
//
if (likely(intent == StorageEngine::DEFAULT))
{
 // do the "normal", unoptimized set of operations
}
elseif (intent == StorageEngine::INTERNAL)
{
 // Let the engine do it internally
 if (engine->passthruStatement(session, *statement) == false)
 {
    // Ask engine for errors to report to the user...
 }
}

Does the above look workable to you?

Yes, absolutely. That would work for me :)

You would have the Statement type as listed in the transaction.proto:

message Statement
{
 enum Type
 {
   ROLLBACK = 0; /* A ROLLBACK indicator */
   INSERT = 1; /* An INSERT statement */
   DELETE = 2; /* A DELETE statement */
   UPDATE = 3; /* An UPDATE statement */
   TRUNCATE_TABLE = 4; /* A TRUNCATE TABLE statement */
   CREATE_SCHEMA = 5; /* A CREATE SCHEMA statement */
   ALTER_SCHEMA = 6; /* An ALTER SCHEMA statement */
   DROP_SCHEMA = 7; /* A DROP SCHEMA statement */
   CREATE_TABLE = 8; /* A CREATE TABLE statement */
   ALTER_TABLE = 9; /* An ALTER TABLE statement */
   DROP_TABLE = 10; /* A DROP TABLE statement */
   SET_VARIABLE = 98; /* A SET statement */
   RAW_SQL = 99; /* A raw SQL statement */
 }
...
}


There is no SELECT in the list, but maybe this is correct. I am just thinking allowed...

We have 3 types of statements:
DML update statements:    INSERT, UPDATE, DELETE
DML read-only statements: SELECT.
DDL statements:           CREATE TABLE, ALTER TABLE, etc.

And assuming we have 2 sets of calls:
- beginTransaction, commitTransaction/rollbackTransaction
- startStatement, endStatement

We could say, all types of statements require a beginTransaction() and a startStatement() (and the corresponding endStatement() and commitTransaction/rollbackTransaction()).

But I don't think this is absolutely correct:

* DML update statements require both beginTransaction() and a startStatement(). * DML read-only statements only require a beginTransaction() call because a SELECT does not need a statement level transaction (because they cannot be rolled back). * And DDL statements only require a startStatement() because it is up to the engine to decide if this can be done within a transaction or not.

For example if beginTransaction() is called before startStatement() then engines that do not handle DDL in transactions should return an error. In addition, if a engine does atomic DDL, then it can use the startStatement() to begin a transaction.

With these calls the engine will have most of the information it needs.

There is some additional information which should be provided when a cursor is used:

For example, PBXT needs to know:

- which columns will be accessed (an optimization so that not all need to be loaded),
- whether rows retrieved will be updated or deleted,
- if the rows need to be locked (as in SELECT FOR UPDATE).

Toru, what's your opinion?

-jay

And this is how the engine would handle "ADD INDEX", or "ENCRYPT TABLE":
startStatement("ENCRYPT TABLE", "t1") --> return: use custom method
doTableOperation("ENCRYPT TABLE", "t1")
endStatement()
The engine can write table operations to its transaction log, and in this way it could ensure that the entire ALTER TABLE statement is atomic.
On Dec 7, 2009, at 4:10 PM, Jay Pipes wrote:
Paul McCullagh wrote:
Hi Toru,
On Dec 7, 2009, at 3:31 AM, Toru Maesaka wrote:
Great to hear another use-case where knowing a statement type in
advance is useful :)
Yes, generally I need to know the following:
- If I have a update type statement (i.e. whether the statement modifies rows). - Whether I need a table lock (examples: ALTER TABLE, TRUNCATE, CHECK).

But, Paul, doesn't this depend on the engine itself?  I mean, some
engines can do (some types of) ALTER TABLE without taking a table lock. So, is this request really for whether the kernel thinks a table- level
lock is necessary, or is it really just for a descriptor of the
statement type?

And, if it really does just boil down to the statement type, then how do we deal with the reality that Brian speaks about -- that statement type will be pluggable, and how do we deal with future statement types for
pluggable engines?

Is a reasonable solution to pass to engines a sort of "statement
traits"?  So, instead of passing ALTER_TABLE, CREATE_TABLE, UPDATE,
DELETE, etc, we instead pass a std::bitset<> (or uint64_t for C folks)
containing traits of the statement such as:

MODIFIES_DATA
MODIFIES_DEFINITION
etc, etc

And then to deal with transaction locking concerns, just add a method to Cursor:

void Cursor::setTransactionIsolationLevel(enum enum_tx_isolation);

Cheers!

Jay

- If we have a SELECT FOR UPDATE.
I was talking to Toru about this, and another possibility is that we have statements declare a needed "lock type" that any plugin could then query. I outlined the solution for Toru, but I don't know if he has written the patch yet :)

I've taken notes from our discussion the other day. I'm planning on
working on it when I finish testing through my current progress of
BlitzDB.
Great! :)
For now, I'm happy with Jay's advise of using
current_session().

Cheers,
Toru

On Sat, Dec 5, 2009 at 5:59 AM, Brian Aker <brian@xxxxxxxxxxx> wrote:
Hi!

On Dec 4, 2009, at 3:12 AM, Paul McCullagh wrote:

If we have a startStatement() call, then it could be used in place of beginAlter(), assuming we can determine the statement type, and the tables involved.

The problem with relying on statement type is that at some point statement type will be pluggable... which means you would constantly need to update your engine for new statements.

Yuck!

I was talking to Toru about this, and another possibility is that we have statements declare a needed "lock type" that any plugin could then query. I outlined the solution for Toru, but I don't know if he has written the patch yet :)


Then, when a handle is returned to the pool it is deleted, instead of adding it back to the pool.

BTW very soon engines will own their Cursor objects and will be free to reuse them.

The locking thread waits until all handles are returned and deleted before it can proceed. The lock on the pool then prevents a new table handle from being created while the locking thread is busy. Either way, it would be good if Drizzle closes all handlers/ cursors before a table is deleted or renamed.

I would say that long term this will be optional, based on what the engine requires.

OK, this make things a lot simpler! Indeed, if we don't need to support LOCK TABLE then external_lock() can be removed altogether.

Tried removing the external_lock() right now and seeing if any issues pop up?

Cheers,
     -Brian
--
Paul McCullagh
PrimeBase Technologies
www.primebase.org
www.blobstreaming.org
pbxt.blogspot.com


--
Paul McCullagh
PrimeBase Technologies
www.primebase.org
www.blobstreaming.org
pbxt.blogspot.com




--
Paul McCullagh
PrimeBase Technologies
www.primebase.org
www.blobstreaming.org
pbxt.blogspot.com






Follow ups

References