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
daemon running maxwell as a daemon
env_config_prefix STRING env vars matching prefix are treated as config values
 
mysql options
host STRING mysql host localhost
user STRING mysql username
password STRING mysql password (no password)
port INT mysql port 3306
jdbc_options STRING mysql jdbc connection options DEFAULT_JDBC_OPTS
ssl SSL_OPT SSL behavior for mysql cx DISABLED
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
recapture_schema BOOLEAN recapture the latest schema. Not available in config.properties. false
 
replication_host STRING server to replicate from. See split server roles 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
replication_ssl SSL_OPT SSL behavior for replication cx cx DISABLED
 
schema_host STRING server to capture schema from. See split server roles 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
schema_ssl SSL_OPT SSL behavior for schema-capture server DISABLED
 
producer options
producer PRODUCER_TYPE type of producer to use stdout
custom_producer.factory CLASS_NAME fully qualified custom producer factory class, see example
producer_ack_timeout PRODUCER_ACK_TIMEOUT time in milliseconds before async producers consider a message lost
producer_partition_by PARTITION_BY input to kafka/kinesis partition function database
producer_partition_columns STRING if partitioning by 'column', a comma separated list of columns
producer_partition_by_fallback PARTITION_BY_FALLBACK required when producer_partition_by=column. Used when the column is missing
ignore_producer_error BOOLEAN Maxwell will be terminated on kafka/kinesis errors when false. Otherwise, those producer errors are only logged. true
 
"file" producer options
output_file STRING output file for file producer
 
"kafka" producer options
kafka.bootstrap.servers STRING kafka brokers, given as HOST:PORT[,HOST:PORT]
kafka_topic STRING kafka topic to write to. maxwell
kafka_version KAFKA_VERSION run maxwell with specified kafka producer version. Not available in config.properties. 0.11.0.1
kafka_partition_hash [ default | murmur3 ] hash function to use when choosing kafka partition default
kafka_key_format [ array | hash ] how maxwell outputs kafka keys, either a hash or an array of hashes hash
ddl_kafka_topic STRING if output_ddl is true, kafka topic to write DDL changes to kafka_topic
 
"kinesis" producer options
kinesis_stream STRING kinesis stream name
 
"sqs" producer options
sqs_queue_uri STRING SQS Queue URI
 
"pubsub" producer options
pubsub_topic STRING Google Cloud pub-sub topic
pubsub_platform_id STRING Google Cloud platform id associated with topic
ddl_pubsub_topic STRING Google Cloud pub-sub topic to send DDL events to
 
"rabbitmq" producer options
rabbitmq_user STRING Username of Rabbitmq connection guest
rabbitmq_pass STRING Password of Rabbitmq connection guest
rabbitmq_host STRING Host of Rabbitmq machine
rabbitmq_port INT Port of Rabbitmq machine
rabbitmq_virtual_host STRING Virtual Host of Rabbitmq
rabbitmq_exchange STRING Name of exchange for rabbitmq publisher
rabbitmq_exchange_type STRING Exchange type for rabbitmq
rabbitmq_exchange_durable BOOLEAN Exchange durability. false
rabbitmq_exchange_autodelete BOOLEAN If set, the exchange is deleted when all queues have finished using it. false
rabbitmq_routing_key_template STRING A string template for the routing key, %db% and %table% will be substituted. %db%.%table%.
rabbitmq_message_persistent BOOLEAN Eanble message persistence. false
rabbitmq_declare_exchange BOOLEAN Should declare the exchange for rabbitmq publisher true
 
"redis" producer options
redis_host STRING Host of Redis server localhost
redis_port INT Port of Redis server 6379
redis_auth STRING Authentication key for a password-protected Redis server
redis_database INT Database of Redis server 0
redis_pub_channel STRING Redis Pub/Sub channel maxwell
redis_list_key STRING Redis LPUSH List Key maxwell
redis_type [ pubsub | lpush ] Selects either Redis Pub/Sub or LPUSH. pubsub
 
formatting
output_binlog_position BOOLEAN records include binlog position false
output_gtid_position BOOLEAN records include gtid position, if available false
output_commit_info BOOLEAN records include commit and xid true
output_xoffset BOOLEAN records include virtual tx-row offset false
output_nulls BOOLEAN records include fields with NULL values true
output_server_id BOOLEAN records include server_id false
output_thread_id BOOLEAN records include thread_id false
output_row_query BOOLEAN records include INSERT/UPDATE/DELETE statement. Mysql option "binlog_rows_query_log_events" must be enabled false
output_ddl BOOLEAN output DDL (table-alter, table-create, etc) events false
 
filtering
filter STRING filter rules, eg exclude: db.*, include: *.tbl, include: *./bar(bar)?/, exclude: foo.bar.col=val
 
encryption
encrypt [ none | data | all ] encrypt mode: none = no encryption. "data": encrypt the data field only. all: encrypt entire maxwell message none
secret_key STRING specify the encryption key to be used null
 
monitoring / metrics
metrics_prefix STRING the prefix maxwell will apply to all metrics MaxwellMetrics
metrics_type [slf4j | jmx | http | datadog] how maxwell metrics will be reported
metrics_jvm BOOLEAN enable jvm metrics: memory usage, GC stats, etc. false
metrics_slf4j_interval SECONDS the frequency metrics are emitted to the log, in seconds, when slf4j reporting is configured 60
http_port INT the port the server will bind to when http reporting is configured 8080
http_path_prefix STRING http path prefix for the server /
http_bind_address STRING the address the server will bind to when http reporting is configured all addresses
http_diagnostic BOOLEAN enable http diagnostic endpoint false
http_diagnostic_timeout MILLISECONDS the http diagnostic response timeout 10000
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.

SSL_OPTION: [ DISABLED | PREFERRED | REQUIRED | VERIFY_CA | VERIFY_IDENTITY ]

PRODUCER_TYPE: [ stdout | file | kafka | kinesis | pubsub | sqs | rabbitmq | redis ]

DEFAULT_JDBC_OPTS: zeroDateTimeBehavior=convertToNull&connectTimeout=5000

PARTITION_BY: [ database | table | primary_key | column ]

PARTITION_BY_FALLBACK: [ database | table | primary_key ]

KAFKA_VERSION: [ 0.8.2.2 | 0.9.0.1 | 0.10.0.1 | 0.10.2.1 | 0.11.0.1 ]

PRODUCER_ACK_TIMEOUT: In certain failure modes, async producers (kafka, kinesis, pubsub, sqs) may simply disappear a message, never notifying maxwell of success or failure. This timeout can be set as a heuristic; after this many milliseconds, maxwell will consider an outstanding message lost and fail it.

Configuration methods

Maxwell is configurable via the command-line, a properties file, or the environment. The configuration priority is:

command line options > scoped env vars > properties file > default values

config.properties

If Maxwell finds the file config.properties in $PWD it will use it. Any command line options (except init_position, replay, kafka_version and daemon) 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.

via environment

If env_config_prefix given via command line or in config.properties, Maxwell will configure itself with all environment variables that match the prefix. The environment variable names are case insensitive. For example, if maxwell is started with --env_config_prefix=FOO_ and the environment contains FOO_USER=auser, this would be equivalent to passing --user=auser.

Deployment scenarios

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".

RDS configuration

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

Split server roles

Maxwell uses MySQL for 3 different functions:

  1. A host to store the captured schema in (--host).
  2. A host to replicate from (--replication_host).
  3. A host to capture the schema from (--schema_host).

Often, all three hosts are the same. host and replication_host should be different if mysql is chained off a slave. schema_host should only be used when using the maxscale replication proxy.

Note that bootstrapping is currently not available when host and replication_host are separate, due to some implementation details.

Multiple Maxwell Instances

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.

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.

Filtering

Example 1

Maxwell can be configured to filter out updates from specific tables. This is controlled by the --filter command line flag. Here's how that flag looks:

--filter = "exclude: foodb.*, include: foodb.tbl, include: foodb./table_\d+/"

This example tells Maxwell to suppress all updates that happen on foodb, except for updates to tbl and any table in foodb matching the regexp /table_\d+/.

Example 2

Filter options are evaluated in the order specified, so in this example we suppress everything except updates in the db1 database.

--filter = "exclude: *.*, include: db1.*"

Column Filters

Maxwell can also include/exclude based on column values:

--filter = "exclude: db.tbl.col = reject"

will reject any row in db.tbl that contains col and where the stringified value of "col" is "reject".

Column Filters / Missing Columns

Column filters are ignored if the specified column is not present, so:

--filter = "exclude: *.*.col_a = *"

will exclude updates to any table that contains col_a, but include every other table.

Blacklisting tables

In rare cases, you may wish to tell Maxwell to completely ignore a database or table, including schema changes. In general, don't use this. If you must use this:

--filter = "blacklist: bad_db.*"

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.

Supressing columns

If you wish to suppress columns from Maxwell's output (for instance, a password field), you can use exclude_columns to filter out columns by name.