Copyright © 2007 Cisco Systems
|Revision $Revision: 5901 $||$Date: 2009-07-21 09:45:05 +0200 (Tue, 21 Jul 2009) $|
Table of Contents
List of Examples
This is a module which integrates the Berkeley DB into OpenSIPS. It implements the DB API defined in OpenSIPS.
The following modules must be loaded before this module:
No dependencies on other OpenSIPS modules.
The auto-reload will close and reopen a Berkeley DB when the files inode has changed. The operation occurs only duing a query. Other operations such as insert or delete, do not invoke auto_reload.
Default value is 0 (1 - on / 0 - off).
The log_enable boolean controls when to create journal files. The following operations can be journaled: INSERT, UPDATE, DELETE. Other operations such as SELECT, do not. This journaling are required if you need to recover from a corrupt DB file. That is, bdb_recover requires these to rebuild the db file. If you find this log feature useful, you may also be interested in the METADATA_LOGFLAGS bitfield that each table has. It will allow you to control which operations to journal, and the destination (like syslog, stdout, local-file). Refer to bdblib_log() and documentation on METADATA.
Default value is 0 (1 - on / 0 - off).
The journal_roll_interval will close and open a new log file. The roll operation occurs only at the end of writing a log, so it is not guaranteed to to roll 'on time'.
Default value is 0 (off).
Example 1.3. Set
... modparam("db_berkeley", "journal_roll_interval", 3600) ...
Causes db_berkeley module to re-read the contents of specified table (or dbenv). The db_berkeley DB actually loads each table on demand, as opposed to loading all at mod_init time. The bdb_reload operation is implemented as a close followed by a reopen. Note- bdb_reload will fail if a table has not been accessed before (because the close will fail).
Parameters: tablename (or db_path); to reload a particular table provide the tablename as the arguement (eg subscriber); to reload all tables provide the db_path to the db files. The path can be found in opensipsctlrc DB_PATH variable.
First download, compile and install the Berkeley DB. This is outside the scope of this document. Documentation for this procedure is available on the Internet.
Next, prepare to compile OpenSIPS with the db_berkeley module. In the directory /modules/db_berkeley, modify the Makefile to point to your distribution of Berkeley DB. You may also define 'BDB_EXTRA_DEBUG' to compile in extra debug logs. However, it is not a recommended deployment to production servers.
Because the module dependes on an external library, the db_berkeley module is not compiled and installed by default. You can use one of the next options.
edit the "Makefile" and remove "db_berkeley" from "excluded_modules" list. Then follow the standard procedure to install OpenSIPS: "make all; make install".
from command line use: 'make all include_modules="db_berkeley"; make install include_modules="db_berkeley"'.
Installation of OpenSIPS is performed by simply running make install as root user of the main directory. This will install the binaries in /usr/local/sbin/. If this was successful, OpenSIPS control engine files should now be installed as /usr/local/sbin/opensipsdbctl.
Decide where (on the filesystem) you want to install the Berkeley DB files. For instance, '/usr/local/etc/opensips/db_berkeley' directory. Make note of this directory as we need to add this path to the opensipsctlrc file. Note: OpenSIPS will not startup without these DB files.
Edit opensipsctlrc - There are two parameters in this file that should be configured before opensipsctldb script can work properly: DBENGINE and DB_PATH. Edit file: '/usr/local/etc/opensips/opensipsctlrc'
## database type: MYSQL, PGSQL, DB_BERKELEY, or DBTEXT, by default none is loaded # DBENGINE=DB_BERKELEY ## database path used by dbtext or db_berkeley # DB_PATH="/usr/local/etc/opensips/db_berkeley"
(Optional) Pre creation step- Customize your meta-data. The DB files are initially seeded with necessary meta-data. This is a good time to review the meta-data section details, before making modifications to your tables dbschema. By default, the files are installed in '/usr/local/share/opensips/db_berkeley/opensips' By default these tables are created Read/Write and without any journalling as shown. These settings can be modified on a per table basis. Note: If you plan to use bdb_recover, you must change the LOGFLAGS.
METADATA_READONLY 0 METADATA_LOGFLAGS 0
Execute opensipsdbctl - There are three (3) groups of tables you may need depending on your situation.
opensipsdbctl create (required) opensipsdbctl presence (optional) opensipsdbctl extra (optional)
Modify the OpenSIPS configuration file to use db_berkeley module. The database URL for modules must be the path to the directory where the Berkeley DB table-files are located, prefixed by "berkeley://", e.g., "berkeley:///usr/local/etc/opensips/db_berkeley".
A couple other IMPORTANT things to consider are the 'db_mode' and the 'use_domain' modparams. The description of these parameters are found in usrloc documentation.
Note on db_mode- The db_berkeley module will only journal the moment usrloc writes back to the DB. The safest mode is mode 3 , since the db_berkeley journal files will always be up-to-date. The main point is the db_mode vs. recovery by journal file interaction. Writing journal entries is 'best effort'. So if the hard drive becomes full, the attempt to write a journal entry may fail.
Note on use_domain- The db_berkeley module will attempt natural joins when performing a query. This is basically a lexigraphical string compare using the keys provided. In most places in the db_berkeley dbschema (unless you customize), the domainname is identified as a natural key. Consider an example where use_domain = 0. In table subscriber, the db will be keying on 'username|NULL' because the default value will be used when that key column is not provided. This effectivly means that later queries must consistently use the username (w.o domain) in order to find a result to that particular subscriber query. The main point is 'use_domain' can not be changed once the db_berkeley is setup.
All Berkeley DB tables are created via the opensipsdbctl script. This section provides details as to the content and format of the DB file upon creation.
Since the Berkeley DB stores key value pairs, the database is seeded with a few meta-data rows . The keys to these rows must begin with 'METADATA'. Here is an example of table meta-data, taken from the table 'version'.
Note on reserved character- The '|' pipe character is used as a record delimiter within the Berkeley DB implementation and must not be present in any DB field.
In the above example, the row METADATA_COLUMNS defines the column names and type, and the row METADATA_KEY defines which column(s) form the key. Here the value of 0 indicates that column 0 is the key(ie table_name). With respect to column types, the db_berkeley modules only has the following types: string, str, int, double, and datetime. The default type is string, and is used when one of the others is not specified. The columns of the meta-data are delimited by whitespace.
The actual column data is stored as a string value, and delimited by the '|' pipe character. Since the code tokenizes on this delimiter, it is important that this character not appear in any valid data field. The following is the output of the 'db_berkeley.sh dump version' command. It shows contents of table 'version' in plain text.
Example 1.5. contents of version table
VERSION=3 format=print type=hash h_nelem=21 db_pagesize=4096 HEADER=END METADATA_READONLY 1 address| address|3 aliases| aliases|1004 dbaliases| dbaliases|1 domain| domain|1 speed_dial| speed_dial|2 subscriber| subscriber|6 uri| uri|1 METADATA_COLUMNS table_name(str) table_version(int) METADATA_KEY 0 acc| acc|4 grp| grp|2 location| location|1004 missed_calls| missed_calls|3 re_grp| re_grp|1 silo| silo|5 trusted| trusted|4 usr_preferences| usr_preferences|2 DATA=END
The METADATA_COLUMNS row contains the column names and types. Each is space delimited. Here is an example of the data taken from table subscriber :
Example 1.6. METADATA_COLUMNS
METADATA_COLUMNS username(str) domain(str) password(str) ha1(str) ha1b(str) first_name(str) last_name(str) email_address(str) datetime_created(datetime) timezone(str) rpid(str)
Related (hardcoded) limitations:
maximum of 32 columns per table.
maximum tablename size is 64.
maximum data length is 2048
Currently supporting these five types: str, datetime, int, double, string.
The METADATA_KEYS row indicates the indexes of the key columns, with respect to the order specified in METADATA_COLUMNS. Here is an example taken from table subscriber that brings up a good point:
The point is that both the username and domain name are require as the key to this record. Thus, usrloc modparam use_domain = 1 must be set for this to work.
The METADATA_READONLY row contains a boolean 0 or 1. By default, its value is 0. On startup the DB will open initially as read-write (loads metadata) and then if this is set=1, it will close and reopen as read only (ro). I found this useful because readonly has impacts on the internal db locking etc.
The METADATA_LOGFLAGS row contains a bitfield that customizes the journaling on a per table basis. If not present the default value is taken as 0. Here are the masks so far (taken from bdb_lib.h):
Example 1.8. METADATA_LOGFLAGS
#define JLOG_NONE 0 #define JLOG_INSERT 1 #define JLOG_DELETE 2 #define JLOG_UPDATE 4 #define JLOG_STDOUT 8 #define JLOG_SYSLOG 16
This means that if you want to journal INSERTS to local file and syslog the value should be set to 1+16=17. Or if you do not want to journal at all, set this to 0.
Use the opensipsdbctl script for maintaining OpenSIPS Berkeley DB tables. This script assumes you have DBENGINE and DB_PATH setup correctly in opensipsctlrc. Note Unsupported commands are- backup, restore, migrate, copy, serweb.
Example 1.9. opensipsdbctl
usage: opensipsdbctl create opensipsdbctl presence opensipsdbctl extra opensipsdbctl drop opensipsdbctl reinit opensipsdbctl bdb list (lists the underlying db files in DB_PATH) opensipsdbctl bdb cat db (prints the contents of db file to STDOUT in plain-text) opensipsdbctl bdb swap db (installs db.new by db -> db.old; db.new -> db) opensipsdbctl bdb append db datafile (appends data to a new instance of db; output DB_PATH/db.new) opensipsdbctl bdb newappend db datafile (appends data to a new instance of db; output DB_PATH/db.new)
The db_berkeley module uses the Concurrent Data Store (CDS) architecture. As such, no transaction or journaling is provided by the DB natively. The application bdb_recover is specifically written to recover data from journal files that OpenSIPS creates. The bdb_recover application requires an additional text file that contains the table schema.
The schema is loaded with the '-s' option and is required for all operations. Provide the path to the db_berkeley plain-text schema files. By default, these install to '/usr/local/share/opensips/db_berkeley/opensips/'.
The '-h' home option is the DB_PATH path. Unlike the Berkeley utilities, this application does not look for the DB_PATH environment variable, so you have to specify it. If not specified, it will assume the current working directory. The last argument is the operation. There are fundamentally only two operations- create and recover.
The following illustrates the four operations available to the administrator.
Example 1.10. bdb_recover usage
usage: ./bdb_recover -s schemadir [-h home] [-c tablename] This will create a brand new DB file with metadata. usage: ./bdb_recover -s schemadir [-h home] [-C all] This will create all the core tables, each with metadata. usage: ./bdb_recover -s schemadir [-h home] [-r journal-file] This will rebuild a DB and populate it with operation from journal-file. The table name is embedded in the journal-file name by convention. usage: ./bdb_recover -s schemadir [-h home] [-R lastN] This will iterate over all core tables enumerated. If journal files exist in 'home', a new DB file will be created and populated with the data found in the last N files. The files are 'replayed' in chronological order (oldest to newest). This allows the administrator to rebuild the db with a subset of all possible operations if needed. For example, you may only be interested in the last hours data in table location.
Important note- A corrupted DB file must be moved out of the way before bdb_recover is executed.