2
# Drizzle Client & Protocol Library
4
# Copyright (C) 2008 Eric Day (eday@oddments.org)
7
# Use and distribution licensed under the BSD license. See
8
# the COPYING file in this directory for full text.
11
STATUS: This is currently a proposed draft as of November 29, 2008
16
The Drizzle protocol works over TCP, UDP, and Unix Domain Sockets
17
(UDS, also known as IPC sockets), although there are limitations when
18
using UDP (this is discussed below). In the case of TCP and UDS,
19
a connection is made, a command is sent, and a response loop is
20
started. Socket communication ends when either side closes the
21
connection or a QUIT command is issued.
23
TCP and UDS communications will be full duplex. This means that as
24
the client is sending a command, it is possible for the server to
25
report an error before the sending of data completes. This allows
26
the server to do preliminary checks (table exists, authentication,
27
...) before a request is completely sent so the client may abort. This
28
will primarily be used for large requests (INSERTing large BLOBs).
30
TCP and UDS communications will also allow for pipe-lining of requests
31
and concurrent command execution. This means a client does not need
32
to wait for a command to finish before a new command is sent. It is
33
even possible a later command issued will complete and have a result
34
before an earlier command. Result packets may be interleaved so a
35
client issuing concurrent commands must be able to parse results
38
UDP sockets are supported to allow small, fast updates for
39
applications such as statistical gathering. Since UDP does not
40
guarantee delivery, this method should not be used for applications
41
that require reliable transport. When using UDP, the authentication
42
packet (if needed) and command packet are bundled into a single UDP
43
packet and sent. This puts a limitation on the size of the request
44
being made, and this limit can be different between network hosts. The
45
absolute limit is 65,507 bytes (28 bytes used for IPv4 and UDP
46
headers), but again, this can depend on the network hosts. Responses
47
are optional when issuing UDP commands, and this preference is
48
specified in the handshake packet.
50
All sizes given throughout this document are in bytes. Byte order
51
for all multi-byte binary objects such as lengths and mutli-byte
52
bit-fields are packed little-endian.
55
Packet Sequence Overview
56
------------------------
58
The sequence of packets for a simple connection and command that
59
responds with an OK packet:
64
The sequence of packets for a simple connection and query command
69
S: Fields (optional, multiple packets)
70
S: Rows (multiple packets)
73
When authentication is required for a command, the server will ask
77
S: Authentication Required
78
C: Authentication Credentials
84
The server will use the most recent credential information when
85
processing subsequent commands.
87
If a client wishes to multiplex commands on a single connection,
88
it can do so using the command identifiers. Here is an example of
89
how the packets could be ordered, but this will largely depend on
90
the servers ability to process the commands concurrently and the
91
processing time for each command.
93
C: Command (Command ID=1)
94
C: Command (Command ID=2)
96
S: Field (Command ID=2)
98
S: Fields (Command ID=2)
99
S: Rows (Command ID=2)
102
As you can see, the commands may be executed with results generated
103
in any order, and the packet containing the results may be interleaved.
109
Some lengths used within the protocol packets are length encoded. This
110
means the size of the length field will vary between 1 and 9 bytes,
111
and is determined by the value of the first byte.
113
0-252 - Actual length
114
253 - NULL value (only applicable in row results)
115
254 - Following 8 bytes contain actual length
116
255 - Depends on context, usually signifies end
122
Packets consist of two layers. The first is meant to be small,
123
simple, and have just enough information for fast router and proxy
124
processing. It consists of a fixed-size part, along with a variable
125
sized client id (explained later), a series of chunked data, followed
126
by a checksum at the end. The chunked transfer encoding allows for
127
not having to pre-compute the packet data length before sending,
128
and support packets of any size. It also allows for a large packet
129
to be aborted gracefully (without having to close the connection)
130
in the event of an error.
132
|----------------------------- 32 Bits -----------------------------|
134
|----------------|----------------|---------------------------------|
135
0 | Magic | Protocol | Command ID |
136
|----------------|----------------|---------------------------------|
137
32 | Command / Result Code | Client ID Length |
138
|---------------------------------|---------------------------------|
139
64 | Client ID (optional, variable length) |
140
|---------------------------------|---------------------------------|
141
64+ | Chunk Length and Value Pairs (optional, variable length) |
142
|---------------------------------|---------------------------------|
143
64+ | Chunk Length = 0 |
144
|---------------------------------|---------------------------------|
146
|-------------------------------------------------------------------|
148
The first part of a packet is:
150
1-byte Magic number, the value should be 0x44.
152
1-byte Protocol version, currently 1.
154
2-byte Command ID. This is a unique number among all other queries
155
currently being executed on the connection. The client is
156
responsible for choosing a unique number while generating a
157
command packet, and all response packets associated with that
158
command must have the same command ID. Once a command has been
159
completed, the client may reuse the ID.
161
2-byte Command/result code. For commands, this may be:
163
1 ECHO - The entire packet is simply echoed back to the caller.
164
2 SET - Set protocol options.
165
3 QUERY - Execute query.
166
4 QUERY_RO - Same as QUERY, but hints that this is a read-only
167
query. This is only useful for routers/proxies who may want
168
to redirect the request to a read slave.
172
1 OK - Single packet success response. No data associated
173
with the result besides parameters.
174
2 ERROR - Single packet error response.
175
3 DATA - Start of a multi-packet result set.
176
3 DATA_END - Mark the end of a series of data packets. This is
177
useful so a low level router or proxy can know when a
178
response is complete without inspecting the contents of
181
2-byte Client ID length.
182
X-byte Client ID (length is value of client ID length). The client ID is
183
there for the client and routers/proxies to use. The server
184
treats this as opaque data, and will only preserve it to send
185
in responses. This can be used as a sharding key, to keep
186
state information in a proxy, or any other use.
188
Next, zero or more chunks are given, terminated by a chunk length of
189
0. Each chunk consist of a length and then that amount of data.
192
X-byte Chunk (length is value of chunk length)
194
After the the chunk length of 0 is given, a checksum value is given
195
that was computed for the entire packet.
199
The second layer of the protocol is encapsulated inside of the
200
chunked encoding. This consists of zero or more packet parameters,
201
an end of parameter marker, followed by an optional data set that is
202
given until the end of a packet (or the end of all chunks).
208
Packet parameter names are defined in a global namespace, although
209
not all parameters are relevant for all packet types. Parameters are
210
enumerated, and the name is specified with a 1-byte value representing
211
the enumerated name. Each packet parameter may have a value associated
212
with it, and each parameter defines the size and how that value is
213
given. The list of possible packet parameters are:
215
0 END_OF_PARAMETERS - Marks the end of a parameter list.
217
Parameters used for setting options:
219
1 AUTH - 1-byte value with authentication mechanism
220
to use. Possible values are:
222
1 - MD5 on user and password.
224
2 CHECKSUM - 1-byte value with preferred checksum
225
type. Possible values are:
228
3 COMPRESSION - 1-byte value with preferred compression
229
type. Possible values are:
233
4 FIELD_ENCODING - 1-byte value with preferred field encoding
234
type. Possible values are:
237
5 FIELD_INFO - 1-byte value to determine if field information
238
should be sent. Possible values are:
242
(6-63 Reserved for future options that can be set)
244
Parameters used in responses:
246
64 STATUS - 4-byte bit field.
247
65 NUM_ROWS_AFFECTED - Length-encoded count of rows affected.
248
66 NUM_ROWS_SCANNED - Length-encoded count of rows scanned.
249
67 NUM_WARNINGS - Length-encoded count of warnings encountered.
250
68 INSERT_ID - Last insert ID.
251
69 ERROR_CODE - 4-byte error code.
252
70 ERROR_STRING - Length-encoded string.
253
71 SQL_STATE - Length-encoded string.
254
72 NUM_FIELDS - 4-byte integer.
255
73 FIELD_START - No value, starts a new set of field parameters.
256
74 FIELD_TYPE - 2-byte enumerated type.
257
75 FIELD_LENGTH - Length-encoded value.
258
76 FIELD_FLAGS - 4-byte bit-field.
259
77 DB_NAME - Length-encoded string.
260
78 TABLE_NAME - Length-encoded string.
261
79 ORIG_TABLE_NAME - Length-encoded string.
262
80 FIELD_NAME - Length-encoded string.
263
81 ORIG_FIELD_NAME - Length-encoded string.
264
82 DEFAULT_VALUE - Length-encoded string.
266
(83-255 Reserved for future responses parameters)
268
"Length-encoded string" means a length-encoded value, followed by a
269
string of that length.
275
Inside of the chunked data, command packets consist of zero or more
276
parameters depending on which options are being set, followed by
277
a end of parameter marker, and then all data until the end of the
278
chunks are considered arguments for the command. For a QUERY, this
279
will be the actual query to run.
285
The server responds with an OK or ERROR if no row data is given. A
286
list of parameters may follow, and the marked with an end of parameter
293
A data packet consists of a series of parameters, followed by the end
294
of parameter, and then a series of length-encoded values holding field
295
values. The NUM_FIELDS parameter must be given before any values, as
296
this indicates when a start of a new row happens. The field values may
297
either be in string format or native data type, depending on the value
300
There may be multiple rows inside of a single DATA result packet. In
301
the case of large result sets, the result should be split into multiple
302
DATA packets since other concurrent commands on the connection will
303
block if a single large packet is sent. By breaking resulting rows
304
into multiple DATA packets, other commands are then allowed to send
305
interleaved response packets.