Awesome
Brucke - Inter-cluster bridge of kafka topics
Brucke is a Inter-cluster bridge of kafka topics powered by Brod
Brucke bridges messages from upstream topic to downstream topic with configurable re-partitionning strategy and message filter/transform plugins
Configuration
A brucke config file is a YAML file.
Config file path should be set in config_file variable of brucke app config, or via BRUCKE_CONFIG_FILE OS env variable.
Cluster names and client names must comply to erlang atom syntax.
kafka_clusters:
kafka_cluster_1:
- "localhost:9092"
kafka_cluster_2:
- "kafka-1:9092"
- "kafka-2:9092"
brod_clients:
- client: brod_client_1
cluster: kafka_cluster_1
config: [] # optional
- client: brod_client_2 # example for SSL connection
cluster: kafka_cluster_1
config:
ssl: true
- client: brod_client_3 # example for SASL_SSL and custom certificates
cluster: kafka_cluster_1
config:
ssl:
# start with "priv/" or provide full path
cacertfile: priv/ssl/ca.crt
certfile: priv/ssl/client.crt
keyfile: priv/ssl/client.key
sasl:
# or 'plain' or 'scram_sha_256'
mechanism: scram_sha_512
username: brucke
password: secret
routes:
- upstream_client: brod_client_1
downstream_client: brod_client_1
upstream_topics:
- "topic_1"
downstream_topic: "topic_2"
repartitioning_strategy: strict_p2p
default_begin_offset: earliest # optional
filter_module: brucke_filter # optional
filter_init_arg: "" # optional
upstream_cg_id: example-consumer-group-1 # optional, build from cluster name otherwise
Client Configs
sslEither set totrueor be specific about the cert/key files to usecacertfile,certfileandkeyfile. In case the file paths start withpriv/bruckewill look in applicationprivdir (i.e.code:priv_dir(brucke)).saslConfigure SASL authmechanism,usernameandpasswordfor SASL PLAIN authentication.query_api_versionsSet tofalsewhen connecting to kafka 0.9.x or earlier
See more configs in brod:start_client/4 API doc.
Routing Configs
Upstream Consumer Group ID
upstream_cg_id identifies the consumer group which route workers (maybe distributed across different Erlang nodes) should join in, and commit offsets to.
Group IDs don't have to be unique, two or more routes may share the same group ID.
However, the same topic may not appear in two routes share the same group ID.
Default Begin Offset
The default_begin_offset route config is to tell upstream consumer from where to start streaming messsages.
Supported values are earliest, latest (default) or a kafka offset (integer) value.
This config is to tell brucke route worker from which offset it should start streaming messages for the first run.
In case of restart, brucke should continue from committed offset.
NOTE: Offsets committed to kafka may expire, in that case, brucke will fallback to this default being offset. NOTE: Currently this option is used for ALL partitions in upstream topic(s).
In case there is a need to discard committed offsets, pick a new group ID.
Offset Commit Policy
The Offset_commit_policy specify how upstream consumer manages the offset per topic-partition.
Two values are available: commit_to_kafka_v2 or consumer_managed.
when consumer_managed is used, topic-partition offsets will be stored in dets (set). see "PATH to offsets DETS file".
default: commit_to_kafka_v2
Repartitioning strategy
NOTE: For compacted topics, strict_p2p is the only choice.
- key_hash: hash the message key to downstream partition number
- strict_p2p: strictly map the upstream partition number to downstream partition number, worker will refuse to start if upstream and downstream topic has different number of partitions
- random: randomly distribute upstream messages to downstream partitions
Message Transformation
Depending on message format version in kafka broker, possible message formats from kafka are
- Version 0: key-value without timestamp (kafka prior to
0.10) - Version 1: key-value with timestamp (since kafka
0.10) - Version 2: key-value with headers and timestamp (since kafka
0.11)
When downstream kafka supports lower version than upstream, unsupported fields are dropped. Otherwise messages are upgraded using default values as below:
- Local host's OS time is used as default timestamp if upstream message message has no
createtimestamp - Empty list
[]is used as defaultheadersfield for downstream message if upstream message has noheaders
Customized Message Filtering and or Transformation
Implement brucke_filter behaviour to have messages filtered and or transformed before produced to downstream topic.
If brucke is packed standalone (i.e. not used as a dependency in a parent project), and the beam files for
brucke_filter modules are installed elsewhere, configure filter_ebin_dirs app env (sys.config),
or set system OS env variables BRUCKE_FILTER_EBIN_PATHS.
Other Routing Options
compression:no_compression(defulat),gziporsnappymax_partitions_per_group_member: default = 12, Number of partitions one group member should work on.required_acks:all(default),leaderornone
Offsets DETS file
You can config brucke where to start consuming by providing the none empty 'Offsets DETS file' when brucke starts.
It should be set table and the record should be in following spec:
{ {GroupID :: binary() , Topic :: binary(), Partition :: non_neg_integer() }, Offset :: -1 | non_neg_integer() }.
offsets_dets_path specify the PATH to the offsets dets file which is used by 'Offset Commit Policy'.
If file does not exists, brucke will create empty one and use it.
default: "/tmp/brucke_offsets.DETS"
ratelimit
Ratelimit the number of messages produced to downstream.
ratelimit_interval defines the interval in milliseconds, default: 0 means disabled.
ratelimit_threshold defines the threshold, set to 0 to pause.
You can change these two settings in realtime via restapi.
example:
Limit the msg rate of consumer group: group1 of cluster: clusterA to 10 msgs/s:
POST /plugins/ratelimiter/clusterA/group1
{"interval": "100", "threshold", "1"}
Graphite reporting
If the following app config variables are set, brucke will send metrics to a configured graphite endpoint:
- graphite_root_path: a prefix for metrics, e.g "myservice"
- graphite_host: e.g. "localhost"
- graphite_port: e.g. 2003
Alternatively, you can use corresponding OS env variables:
- BRUCKE_GRAPHITE_ROOT_PATH
- BRUCKE_GRAPHITE_HOST
- BRUCKE_GRAPHITE_PORT
RPM packaging
Generate a release and package it into an rpm package:
make rpm
Brucke package installs release and creates corresponding systemd service. Config files are in /etc/brucke, OS env can be set at /etc/sysconfig/brucke, logs are in /var/log/brucke.
Operating via systemctl:
systemctl start brucke
systemctl enable brucke
systemctl status brucke
Script for config validation
Build:
make escript
Usage:
brucke <path/to/config.yml> [filter-module-ebin-dir ...]
Set BRUCKE_FILTER_MODULE_BEAM_DIRS for extra filter module beam lookup.
For valid configs the validation command should silently exit with code 0.
Http Endpoint for Healthcheck
Default port is 8080, customize via http_port config option or via BRUCKE_HTTP_PORT OS env variable.
GET /ping
Returns pong if the application is up and running.
GET /healthcheck
Responds with status 200 if everything is OK, and 500 if something is not OK. Also returns healthy and unhealthy routes in response body in JSON format.
Example response:
{
"discarded": [
{
"downstream": {
"endpoints": [
{
"host": "localhost",
"port": 9192
}
],
"topic": "brucke-filter-test-downstream"
},
"options": {
"default_begin_offset": "earliest",
"filter_init_arg": [],
"filter_module": "brucke_test_filter",
"repartitioning_strategy": "strict_p2p"
},
"reason": [
"filter module brucke_test_filter is not found\nreason:embedded\n"
],
"upstream": {
"endpoints": [
{
"host": "localhost",
"port": 9092
}
],
"topics": [
"brucke-filter-test-upstream"
]
}
}
],
"healthy": [
{
"downstream": {
"endpoints": [
{
"host": "localhost",
"port": 9192
}
],
"topic": "brucke-basic-test-downstream"
},
"options": {
"consumer_config": {
"begin_offset": "earliest"
},
"filter_init_arg": [],
"filter_module": "brucke_filter",
"max_partitions_per_group_member": 12,
"producer_config": {
"compression": "no_compression"
},
"repartitioning_strategy": "strict_p2p"
},
"upstream": {
"endpoints": [
{
"host": "localhost",
"port": 9092
}
],
"topics": "brucke-basic-test-upstream"
}
},
{
"downstream": {
"endpoints": [
{
"host": "localhost",
"port": 9092
}
],
"topic": "brucke-test-topic-2"
},
"options": {
"consumer_config": {
"begin_offset": "earliest"
},
"filter_init_arg": [],
"filter_module": "brucke_filter",
"max_partitions_per_group_member": 12,
"producer_config": {
"compression": "no_compression"
},
"repartitioning_strategy": "strict_p2p"
},
"upstream": {
"endpoints": [
{
"host": "localhost",
"port": 9092
}
],
"topics": "brucke-test-topic-1"
}
},
{
"downstream": {
"endpoints": [
{
"host": "localhost",
"port": 9092
}
],
"topic": "brucke-test-topic-3"
},
"options": {
"consumer_config": {
"begin_offset": "latest"
},
"filter_init_arg": [],
"filter_module": "brucke_filter",
"max_partitions_per_group_member": 12,
"producer_config": {
"compression": "no_compression"
},
"repartitioning_strategy": "key_hash"
},
"upstream": {
"endpoints": [
{
"host": "localhost",
"port": 9092
}
],
"topics": "brucke-test-topic-2"
}
},
{
"downstream": {
"endpoints": [
{
"host": "localhost",
"port": 9192
}
],
"topic": "brucke-test-topic-1"
},
"options": {
"consumer_config": {
"begin_offset": "latest"
},
"filter_init_arg": [],
"filter_module": "brucke_filter",
"max_partitions_per_group_member": 12,
"producer_config": {
"compression": "no_compression"
},
"repartitioning_strategy": "random"
},
"upstream": {
"endpoints": [
{
"host": "localhost",
"port": 9192
}
],
"topics": "brucke-test-topic-3"
}
}
],
"status": "failing",
"unhealthy": []
}