1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
|
ETDB
====
To enable edge computing in its full capacity, a crucial requirement is to manage the state of edge applications, preferably in database that provides the full features of SQL including joins and transactions. The key challenge here is to provide a replicated database for the edge that can scale to thousands of geo-distributed nodes. Existing solutions either provide semantics that are too weak (PostgreSQL replicates asynchronously) or too strong and hence expensive to realize in a geo-distributed network with its WAN latencies and complex failure modes (MariaDb, Spanner, provide full transactionality). Inspired by entry consistency in shared memory systems, wherein only the lock holder for an object obtains sequential consistency for the object, we define the notion of an entry transactional database, which is a novel partitioned database in which only the “owner” of a partition obtains full ACID transactionality. In this work, we define the semantics of an entry transactional database, describe the hard challenges faced in building it and present a novel middleware called mdbc that combines existing SQL databases with an underlying geo-distributed entry consistent store to provide entry transactional semantics. Further, we present crucial use cases such as a federated regional controller for the network control plane and a state management service for edge mobility enabled by entry transactionality.
## Building ETDB
ETDB is built with Maven. This directory contains two pom.xml files.
The first (*pom.xml*) will build a jar file to be used by applications wishing to use the
ETDB JDBC driver.
The second (*pom-h2server.xml*) is used to built the special code that needs to be loaded
into an H2 server, when running ETDB against a copy of H2 running as a server.
### Building the JBoss ETDB Module
There is a shell script (located in `src/main/shell/mk_jboss_module`) which, when run in
the mdbc source directory, will create a tar file `target/mdbc-jboss-module.tar` which can
be used as a JBoss module. This tar file should be installed by un-taring it in the
$JBOSS_DIR/modules directory on the JBoss server.
## Using ETDB
This package provides a JDBC driver that can be used to mirror the contents of a database
to and from Cassandra. The mirroring occurs as a side effect of execute() statements against
a JDBC connection, and triggers placed in the database to catch database modifications.
The initial implementation is written to support H2, MySQL, and MariaDB databases.
This JDBC driver will intercept all table creations, SELECTs, INSERTs, DELETEs, and UPDATEs
made to the underlying database, and make sure they are copied to Cassandra.
In addition, for every table XX that is created, another table DIRTY\_XX will be created to
communicate the existence of dirty rows to other Cassandra replicas (with the Cassandra2
Mixin, the table is called DIRTY\_\_\_\_ and there is only one table).
Dirty rows will be copied, as needed back into the database from Cassandra before any SELECT.
### To use directly with JDBC
1. Add this jar, and all dependent jars to your CLASSPATH.
2. Rewrite your JDBC URLs from jdbc:_yourdb_:... to jdbc:mdbc:....
3. If you supply properties to the DriverManager.getConnection(String, Properties) call,
use the properties defined below to control behavior of the proxy.
4. Load the driver using the following call:
Class.forName("com.att.research.mdbc.ProxyDriver");
The following properties can be passed to the JDBC DriverManager.getConnection(String, Properties)
call to influence how ETDB works.
| Property Name | Property Value | Default Value |
|--------------------|--------------------------------------------------------------------------------|---------------|
| MDBC\_DB\_MIXIN | The mixin name to use to select the database mixin to use for this connection. | mysql |
| MDBC\_MUSIC\_MIXIN | The mixin name to use to select the MUSIC mixin to use for this connection. | cassandra2 |
| myid | The ID of this replica in the collection of replicas sharing the same tables. | 0 |
| replicas | A comma-separated list of replica names for the collection of replicas sharing the same tables. | the value of myid |
| music\_keyspace | The keyspace name to use in Cassandra for all tables created by this instance of ETDB. | mdbc |
| music\_address | The IP address to use to connect to Cassandra. | localhost |
| music\_rfactor | The replication factor to use for the new keyspace that is created. | 2 |
| disabled | If set to true the mirroring is completely disabled; this is the equivalent of using the database driver directly. | false |
The values of the mixin properties may be:
| Property Name | Property Value | Purpose |
|--------------------|----------------|---------------|
| MDBC\_DB\_MIXIN | h2 | This mixin provides access to either an in-memory, or a local (file-based) version of the H2 database. |
| MDBC\_DB\_MIXIN | h2server | This mixin provides access to a copy of the H2 database running as a server. Because the server needs special Java classes in order to handle certain TRIGGER actions, the server must be et up in a special way (see below). |
| MDBC\_DB\_MIXIN | mysql | This mixin provides access to MySQL or MariaDB (10.2+) running on a remote server. |
| MDBC\_MUSIC\_MIXIN | cassandra | A Cassandra based persistence layer (without any of the table locking that MUSIC normally provides). |
| MDBC\_MUSIC\_MIXIN | cassandra2 | Similar to the _cassandra_ mixin, but stores all dirty row information in one table, rather than one table per real table. |
### To Define a JBoss DataSource
The following code snippet can be used as a guide when setting up a JBoss DataSource.
This snippet goes in the JBoss *service.xml* file. The connection-property tags may
need to be added/modified for your purposes. See the table above for names and values for
these tags.
```
<datasources>
<datasource jta="true" jndi-name="java:jboss/datasources/ProcessEngine" pool-name="ProcessEngine" enabled="true" use-java-context="true" use-ccm="true">
<connection-url>jdbc:mdbc:/opt/jboss-eap-6.2.4/standalone/camunda-h2-dbs/process-engine;DB_CLOSE_DELAY=-1;MVCC=TRUE;DB_CLOSE_ON_EXIT=FALSE</connection-url>
<connection-property name="music_keyspace">
camunda
</connection-property>
<driver>mdbc</driver>
<security>
<user-name>sa</user-name>
<password>sa</password>
</security>
</datasource>
<drivers>
<driver name="mdbc" module="com.att.research.mdbc">
<driver-class>com.att.research.mdbc.ProxyDriver</driver-class>
</driver>
</drivers>
</datasources>
```
Note: This assumes that you have built and installed the com.att.research.mdbc module within JBoss.
### To Define a Tomcat DataSource Resource
The following code snippet can be used as a guide when setting up a Tomcat DataSource resource.
This snippet goes in the Tomcat *server.xml* file. As with the JBoss DataSource, you will
probably need to make changes to the _connectionProperties_ attribute.
```
<Resource name="jdbc/ProcessEngine"
auth="Container"
type="javax.sql.DataSource"
factory="org.apache.tomcat.jdbc.pool.DataSourceFactory"
uniqueResourceName="process-engine"
driverClassName="com.att.research.mdbc.ProxyDriver"
url="jdbc:mdbc:./camunda-h2-dbs/process-engine;MVCC=TRUE;TRACE_LEVEL_FILE=0;DB_CLOSE_ON_EXIT=FALSE"
connectionProperties="myid=0;replicas=0,1,2;music_keyspace=camunda;music_address=localhost"
username="sa"
password="sa"
maxActive="20"
minIdle="5" />
```
## Databases Supported
Currently, the following databases are supported with ETDB:
* H2: The `H2Mixin` mixin is used when H2 is used with an in-memory (`jdbc:h2:mem:...`)
or local file based (`jdbc:h2:path_to_file`) database.
* H2 (Server): The `H2ServerMixin` mixin is used when H2 is used with an H2 server (`jdbc:h2:tcp:...`).
* MySQL: The `MySQLMixin` mixin is used for MySQL.
* MariaDB: The `MySQLMixin` mixin is used for MariaDB, which is functionally identical to MySQL.
## Testing Mixin Combinations
The files under `src/main/java/com/att/research/mdbc/tests` can be used to test various ETDB
operations with various combinations of Mixins. The tests are controlled via the file
`src/main/resources/tests.json`. More details are available in the javadoc for this package.
## Limitations of ETDB
* The `java.sql.Statement.executeBatch()` method is not supported by ETDB.
It is not prohibited either; if you use this, your results will be unpredictable (and probably wrong).
* When used with a DB server, there is some delay as dirty row information is copied
from a table in the database, to the dirty table in Cassandra. This opens a window
during which all sorts of mischief may occur.
* ETDB *only* copies the results of SELECTs, INSERTs, DELETEs, and UPDATEs. Other database
operations must be performed outside of the purview of ETDB. In particular, CREATE-ing or
DROP-ing tables or databases must be done individually on each database instance.
* Some of the table definitions may need adjusting depending upon the variables of your use
of ETDB. For example, the MySQL mixin assumes (in its definition of the ETDB_TRANSLOG table)
that all table names will be no more than 255 bytes, and that tables rows (expressed in JSON)
will be no longer than 512 bytes. If this is not true, you should adjust, edit, and recompile.
* ETDB is limited to only data types that can be easily translated to a Cassandra equivalent;
e.g. BIGINT, BOOLEAN, BLOB, DOUBLE, INT, TIMESTAMP, VARCHAR
* To find the data types that your database is currently using run the following command:
SELECT DISTINCT DATA_TYPE FROM INFORMATION_SCHEMA.COLUMNS WHERE TABLE_SCHEMA='<your db name here>';
|
