11
* There is no embedded server. The Drizzle Server is not loadable as a shared library.
12
* Drizzle is optimized for massively concurrent environments. If we have the choice of improving performance for 1024 simultaneous connections to the detriment of performance with only 64 connections, we will take that choice.
11
* There is no embedded server. The Drizzle Server is not loadable as a shared
13
* Drizzle is optimized for massively concurrent environments. If we have the
14
choice of improving performance for 1024 simultaneous connections to the
15
detriment of performance with only 64 connections, we will take that choice.
13
16
* It is designed for modern POSIX systems
14
17
* Microsoft Windows is not a supported platform (neither is HP-UX or IRIX).
15
18
* Drizzle doesn't use timezones. Everything is UTC.
20
23
* No scripts/mysql_install_db or similar. Drizzle aims for a "just works" installation, without administrative overhead.
21
24
* No system database that needs upgrading between versions.
22
* Drizzle can listen on the Drizzle port (4427) and/or MySQL port (3306) and speak the respective protocols.
25
* Drizzle can listen on the Drizzle port (4427) and/or MySQL port (3306)
26
and speak the respective protocols.
27
31
Drizzle is designed around the concept of being a microkernel. There should
28
32
be a small core of the server with most functionality being provided through
29
33
small, efficient and hard to misuse plugin interfaces. The goal is a small,
30
light-weight kernel that is easy to maintain, understand and extend.
34
light weight kernel that is easy to maintain, understand and extend.
32
36
Drizzle is written in C++ and makes use of the Standard Template Library (STL)
33
37
and Boost. Only where performance or correctness proves to be inadequate will
40
44
Pluggable network protocols allow Drizzle to speak one (or more) of several
41
45
protocols. Currently we support the MySQL protocol (compatible with existing
42
MySQL client libraries) and the Drizzle protocol, which is still under
46
MySQL client libraries) and the Drizzle protocol which is still under
45
49
The Drizzle protocol embodies several important differences from MySQL:
47
* Client sends first packet (rather than the server)
51
* Client sends first packet instead of server
48
52
* Built in sharding
49
* Multi statement support (without requiring a semicolon to separate them)
50
* Room for expansion to include NoSQL-type commands inline with SQL commands.
53
* Multi statement support (without using a semicolon to separate them)
54
* Room for expansion to include NoSQL type commands inline with SQL.
52
56
There is also a console plugin -- instead of providing access over a network
53
57
socket, this plugin allows access from the current tty.
60
64
* User Defined Functions (UDFs) now follow the same API as a given
61
65
server instead of a different C API. This means that UDFs are on the
62
exact same level as built-in functions
63
* Some parts of the storage Engine API have been extensively reworked, especially
64
around transactions and DDL
66
exact same level as built-in functions.
67
* Storage Engine API has had some parts extensively reworked, especially
68
around transactions and DDL.
65
69
* Logging is now pluggable
66
70
* Authentication is pluggable
67
71
* Replication is pluggable
68
72
* INFORMATION_SCHEMA plugins have been replaced by the function_engine, which
69
is a lot more space and time efficient
73
is a lot more space and time efficient.
70
74
* Network protocols are pluggable
71
* Scheduler is pluggable (multi_thread, pool_of_threads, etc)
72
* Plugin points for manipulating rows before/after operations: these can be used for
73
replication and the PBMS Blob Streaming plugin
75
* Scheduler is pluggable (multi_thread, pool_of_threads etc)
76
* Plugin points for manipulating rows before/after operations: used for
77
replication and the PBMS Blob Streaming plugin.
78
82
Drizzle does not currently have any plugins that implement stored procedures. We
79
viewed the implementation in MySQL to be non-optimal. They bloat the parser
80
and only support one language (SQL2003 stored procedures), which was not
83
viewed the implementation in MySQL to be non-optimal, bloating the parser
84
and only supporting one language (SQL2003 stored procedures), which was not
83
87
Fundamentally, stored procedures usually are not the correct architectural
84
88
decision for applications that need to scale. Pushing more computation down
85
89
into the database (which is the trickiest layer to scale) isn't a good idea.
87
We do recognize the value of using stored procedures to reduce the time row locks are held, but think we can achieve the same advantage by improved batching of commands over the wire. This removes adding and administering stored procedures from the list of things that can go wrong in administering the database.
91
We do recognize that the ability to reduce the time row locks are held
92
by using stored procedures is valuable, but think we can achieve the same
93
advantage by improved batching of commands over the wire instead of adding and
94
administering stored procedures to the list of things that can go wrong in
95
administering the database.