Mysql Server configuration

At a minimum, Maxwell needs row-level-replication turned on into order to operate:

[mysqld]
server-id=1
log-bin=master
binlog_format=row

GTID support

As of 1.8.0, Maxwell contains support for GTID-based replication. Enable it with the --gtid_mode configuration param.

Here's how you might configure your mysql server for GTID mode:

$ vi my.cnf

[mysqld]
server-id=1
log-bin=master
binlog_format=row
gtid-mode=ON
log-slave-updates=ON
enforce-gtid-consistency=true

When in GTID-mode, Maxwell will transparently pick up a new replication position after a master change. Note that you will still have to re-point maxwell to the new master.

GTID support in Maxwell is considered alpha-quality at the moment; notably, Maxwell is unable to transparently upgrade from a traditional-replication scenario to a GTID-replication scenario; currently, when you enable gtid mode Maxwell will recapture the schema and GTID-position from "wherever the master is at".

Properties file

If maxwell finds the file config.properties in $PWD it will use it. Any command line options (except init_position, replay, and kafka_version) may be given as "key=value" pairs.

Additionally, any configuration file options prefixed with 'kafka.' will be passed into the kafka producer library, after having 'kafka.' stripped off the front of the key. So, for example if config.properties contains

kafka.batch.size=16384

then Maxwell will send batch.size=16384 to the kafka producer library.

RDS configuration

To run Maxwell against RDS, (either Aurora or Mysql) you will need to do the following:

Filters

The options include_dbs, exclude_dbs, include_tables, and exclude_tables control whether Maxwell will send an update for a given row to its producer. All the options take a single value PATTERN, which may either be a literal table/database name, given as option=name, or a regular expression, given as option=/regex/. The options are evaluated as follows:

  1. only accept databases in include_dbs if non-empty
  2. reject databases in exclude_dbs
  3. only accept tables in include_tables if non-empty
  4. reject tables in exclude_tables

So an example like --include_dbs=/foo.*/ --exclude_tables=bar will include footy.zab and exclude footy.bar

The option blacklist_tables and blacklist_dbs controls whether Maxwell will send updates for a table to its producer AND whether it captures schema changes for that table or database. Note that once Maxwell has been running with a table or database marked as blacklisted, you must continue to run Maxwell with that table or database blacklisted or else Maxwell will halt. If you want to stop blacklisting a table or database, you will have to drop the maxwell schema first.

Schema storage host vs replica host

Maxwell needs two sets of mysql permissions to operate properly: a mysql database in which to store schema snapshots, and a mysql host to replicate from. The recommended configuration is that these two functions are provided by a single mysql host. In this case, just specify host, user, etc.

Some configurations, however, may need to write data to a different server than it replicates from. In this case, the host described by host, user, ..., will be used to write schema information to, and Maxwell will replicate events from the host described by replication_host, replication_user, ... Note that bootstrapping is not available in this configuration.

It's also possible to specify a different host to capture schema from than to replicate from, via schema_host, ... This is useful when using Maxscale as a replication proxy.

running multiple instances of maxwell against the same master

Maxwell can operate with multiple instances running against a single master, in different configurations. This can be useful if you wish to have producers running in different configurations, for example producing different groups of tables to different topics. Each instance of Maxwell must be configured with a unique client_id, in order to store unique binlog positions.

multiple instances on a 5.5 server

With MySQL 5.5 and below, each replicator (be it mysql, maxwell, whatever) must also be configured with a unique replica_server_id. This is a 32-bit integer that corresponds to mysql's server_id parameter. The value you configure should be unique across all mysql and maxwell instances.

reference

At the minimum, you will need to specify 'host', 'user', 'password', 'producer'. The kafka producer requires 'kafka.bootstrap.servers', the kinesis producer requires 'kinesis_stream'.

option argument description default
general options
config STRING location of config.properties file $PWD/config.properties
log_level [debug | info | warn | error] log level INFO
 
mysql options
host STRING mysql host localhost
user STRING mysql username
password STRING mysql password (no password)
port INT mysql port 3306
schema_database STRING database to store schema and position in maxwell
client_id STRING unique text identifier for maxwell instance maxwell
replica_server_id LONG unique numeric identifier for this maxwell instance 6379 (see notes)
master_recovery BOOLEAN enable experimental master recovery code false
gtid_mode BOOLEAN enable GTID-based replication false
ignore_producer_error BOOLEAN Maxwell will be terminated on kafka/kinesis errors when false. Otherwise, those producer errors are only logged. true
 
replication_host STRING mysql host to replicate from. Only specify if different from host (see notes) schema-store host
replication_password STRING password on replication server (none)
replication_port INT port on replication server 3306
replication_user STRING user on replication server
 
schema_host STRING mysql host to capture schema from. Useful for using Maxscale as a binlog-proxy schema-store host
schema_password STRING password on schema-capture server (none)
schema_port INT port on schema-capture server 3306
schema_user STRING user on schema-capture server
 
producer options
producer [stdout | kafka | file | profiler] type of producer to use stdout
output_file STRING output file for file producer
 
kafka.bootstrap.servers STRING kafka brokers, given as HOST:PORT[,HOST:PORT]
kafka_topic STRING kafka topic to write to. static string or variable replacement maxwell
producer_partition_by [database | table | primary_key | column] input to kafka partition function database
producer_partition_columns STRING if partitioning by 'column', a comma separated list of columns
producer_partition_by_fallback [database | table | primary_key] required when producer_partition_by=column. Used when the column is missing
kafka_partition_hash [default | murmur3] hash function to use when hoosing kafka partition default
ddl_kafka_topic STRING if output_ddl is true, kafka topic to write DDL changes to kafka_topic
kafka_version [0.8 | 0.9 | 0.10 | 0.10.1 | 0.10.2] run maxwell with kafka producer 0.8.2, 0.9.0, 0.10.0.1, 0.10.1.0 or 0.10.2.1. Not available in config.properties. 0.9.0
 
kinesis_stream STRING kinesis stream name
 
formatting
output_binlog_position BOOLEAN should produced records include binlog position false
output_commit_info BOOLEAN should produced records include commit and xid true
output_nulls BOOLEAN produced records include fields with NULL values true
output_server_id BOOLEAN produced records include server_id false
output_thread_id BOOLEAN produced records include thread_id false
output_ddl BOOLEAN output DDL (table-alter, table-create, etc) events false
 
filtering
include_dbs PATTERN only send updates from these databases
exclude_dbs PATTERN ignore updates from these databases
include_tables PATTERN only send updates from tables named like PATTERN
exclude_tables PATTERN ignore updates from tables named like PATTERN
blacklist_dbs PATTERN ignore updates AND schema changes from databases (see warnings below)
blacklist_tables PATTERN ignore updates AND schema changes from tables named like PATTERN (see warnings below)
 
monitoring
metrics_prefix STRING the prefix maxwell will apply to all metrics MaxwellMetrics
metrics_type [slf4j | jmx | http | datadog] how maxwell metrics will be reported, at least one of slf4j | jmx | http | datadog
metrics_slf4j_interval INT the frequency metrics are emitted to the log, in seconds, when slf4j reporting is configured 60
metrics_http_port INT the port the server will bind to when http reporting is configured 8080
metrics_datadog_type [udp | http] when metrics_type includes datadog this is the way metrics will be reported, can only be one of [udp | http] udp
metrics_datadog_tags STRING datadog tags that should be supplied, e.g. tag1:value1,tag2:value2
metrics_datadog_interval INT the frequency metrics are pushed to datadog, in seconds 60
metrics_datadog_apikey STRING the datadog api key to use when metrics_datadog_type = http
metrics_datadog_host STRING the host to publish metrics to when metrics_datadog_type = udp localhost
metrics_datadog_port INT the port to publish metrics to when metrics_datadog_type = udp 8125
 
misc
bootstrapper [async | sync | none] bootstrapper type. See bootstrapping docs. async
init_position FILE:POSITION:HEARTBEAT ignore the information in maxwell.positions and start at the given binlog position. Not available in config.properties.
replay BOOLEAN enable maxwell's read-only "replay" mode: don't store a binlog position or schema changes. Not available in config.properties.