Skip to main content

Read/Write Kafka

You can read data from Apache Kafka (as well as Confluent Cloud, or Redpanda) in Proton with External Stream. Combining with Materialized View and Target Stream, you can also write data to Apache Kafka with External Stream.

CREATE EXTERNAL STREAM

Currently Timeplus external stream only supports Kafka API as the only type.

To create an external stream in Proton:

CREATE EXTERNAL STREAM [IF NOT EXISTS] stream_name (<col_name1> <col_type>)
SETTINGS type='kafka', brokers='ip:9092',topic='..',security_protocol='..',username='..',password='..',sasl_mechanisms='..'

The supported values for security_protocol are:

  • PLAINTEXT: when this option is omitted, this is also the default value.
  • SASL_SSL: when this value is set, username and password should be specified.

The supported values for sasl_mechanisms are the followings. You can list multiple ones, such as " GSSAPI,PLAIN"

  • PLAIN: when you set security_protocol to SASL_SSL, this is the default value for sasl_mechanisms.
  • SCRAM-SHA-256
  • SCRAM-SHA-512
  • GSSAPI

Connect to local Kafka or Redpanda

Example:

CREATE EXTERNAL STREAM ext_github_events(raw string)
SETTINGS type='kafka',
brokers='localhost:9092',
topic='github_events'

Connect to Confluent Cloud

Example:

CREATE EXTERNAL STREAM ext_github_events(raw string)
SETTINGS type='kafka',
brokers='pkc-1234.us-west-2.aws.confluent.cloud:9092',
topic='github_events',
security_protocol='SASL_SSL',
username='..',
password='..'

Define columns

Single column to read from Kafka

If the message in Kafka topic is in plain text format or JSON, you can create an external stream with only a raw column in string type.

Example:

CREATE EXTERNAL STREAM ext_github_events
(raw string)
SETTINGS type='kafka',
brokers='localhost:9092',
topic='github_events'

Then use query time JSON extraction functions or shortcut to access the values, e.g. raw:id.

Multiple columns to read from Kafka

If the keys in the JSON message never change, you can also create the external stream with multiple columns (only available to Proton v1.3.24+).

You can either:

  • make sure all keys in the JSON are defined as columns, with proper data types. Otherwise, if there are more key/value pairs in the JSON message than what're defined in the external stream, the query won't show any result.
  • or only define some keys as columns and append this to your query: SETTINGS input_format_skip_unknown_fields=true

Example:

CREATE EXTERNAL STREAM ext_github_events
(actor string,
created_at string,
id string,
payload string,
repo string,
type string
)
SETTINGS type='kafka',
brokers='localhost:9092',
topic='github_events',
data_format='JSONEachRow';

If there are nested complex JSON in the message, you can define the column as a string type.

Multiple columns to write to Kafka

To write data via Kafka API (only available to Proton v1.3.18+), you can choose different data formats:

JSONEachRow

You can use data_format='JSONEachRow' to inform Proton to write each event as a JSON document. The columns of the external stream will be converted to keys in the JSON documents. For example:

CREATE EXTERNAL STREAM target(
_tp_time datetime64(3),
url string,
method string,
ip string)
SETTINGS type='kafka',
brokers='redpanda:9092',
topic='masked-fe-event',
data_format='JSONEachRow';

The messages will be generated in the specific topic as

{
"_tp_time":"2023-10-29 05:36:21.957"
"url":"https://www.nationalweb-enabled.io/methodologies/killer/web-readiness"
"method":"POST"
"ip":"c4ecf59a9ec27b50af9cc3bb8289e16c"
}
info

Please note, since 1.3.25, by default multiple JSON documents will be inserted to the same Kafka message. One JSON document each row/line. Such default behavior aims to get the maximum writing performance to Kafka/Redpanda. But you need to make sure the downstream applications are able to properly split the JSON documents per Kafka message.

If you need a valid JSON per each Kafka message, instead of a JSONL, please set one_message_per_row=true e.g.

CREATE EXTERNAL STREAM target(_tp_time datetime64(3), url string, ip string) 
SETTINGS type='kafka', brokers='redpanda:9092', topic='masked-fe-event',
data_format='JSONEachRow',one_message_per_row=true

The default value of one_message_per_row, if not specified, is false.

CSV

You can use data_format='CSV' to inform Proton to write each event as a JSON document. The columns of the external stream will be converted to keys in the JSON documents. For example:

CREATE EXTERNAL STREAM target(
_tp_time datetime64(3),
url string,
method string,
ip string)
SETTINGS type='kafka',
brokers='redpanda:9092',
topic='masked-fe-event',
data_format='CSV';

The messages will be generated in the specific topic as

"2023-10-29 05:35:54.176","https://www.nationalwhiteboard.info/sticky/recontextualize/robust/incentivize","PUT","3eaf6372e909e033fcfc2d6a3bc04ace"

DROP EXTERNAL STREAM

DROP EXTERNAL STREAM [IF EXISTS] stream_name

Query Kafka Data with SQL

You can run streaming SQL on the external stream, e.g.

SELECT raw:timestamp, raw:car_id, raw:event FROM ext_stream WHERE raw:car_type in (1,2,3);
SELECT window_start, count() FROM tumble(ext_stream,to_datetime(raw:timestamp)) GROUP BY window_start;

Read existing messages

When you run SELECT raw FROM ext_stream , Proton will read the new messages in the topics, not the existing ones. If you need to read all existing messages, you can use the following settings:

SELECT raw FROM ext_stream SETTINGS seek_to='earliest'

Read specified partitions

Starting from Proton 1.3.18, you can also read in specified Kafka partitions. By default, all partitions will be read. But you can also read from a single partition via the shards setting, e.g.

SELECT raw FROM ext_stream SETTINGS shards='0'

Or you can specify a set of partition ID, separated by comma, e.g.

SELECT raw FROM ext_stream SETTINGS shards='0,2'

Write to Kafka with SQL

You can use materialized views to write data to Kafka as an external stream, e.g.

-- read the topic via an external stream
CREATE EXTERNAL STREAM frontend_events(raw string)
SETTINGS type='kafka',
brokers='redpanda:9092',
topic='owlshop-frontend-events';

-- create the other external stream to write data to the other topic
CREATE EXTERNAL STREAM target(
_tp_time datetime64(3),
url string,
method string,
ip string)
SETTINGS type='kafka',
brokers='redpanda:9092',
topic='masked-fe-event',
data_format='JSONEachRow';

-- setup the ETL pipeline via a materialized view
CREATE MATERIALIZED VIEW mv INTO target AS
SELECT now64() AS _tp_time,
raw:requestedUrl AS url,
raw:method AS method,
lower(hex(md5(raw:ipAddress))) AS ip
FROM frontend_events;

Tutorial with Docker Compose

A docker-compose file is created to bundle proton image with Redpanda (as lightweight server with Kafka API), Redpanda Console, and owl-shop as sample live data.

  1. Download the docker-compose.yml and put into a new folder.
  2. Open a terminal and run docker compose up in this folder.
  3. Wait for few minutes to pull all required images and start the containers. Visit http://localhost:8080 to use Redpanda Console to explore the topics and live data.
  4. Use proton-client to run SQL to query such Kafka data: docker exec -it <folder>-proton-1 proton-client You can get the container name via docker ps
  5. Create an external stream to connect to a topic in the Kafka/Redpanda server and run SQL to filter or aggregate data.

Create an external stream

CREATE EXTERNAL STREAM frontend_events(raw string)
SETTINGS type='kafka',
brokers='redpanda:9092',
topic='owlshop-frontend-events'
info

Since Proton 1.3.24, you can also define multiple columns.

CREATE EXTERNAL STREAM frontend_events_json(
version int,
requestedUrl string,
method string,
correlationId string,
ipAddress string,
requestDuration int,
response string,
headers string
)
SETTINGS type='kafka',
brokers='redpanda:9092',
topic='owlshop-frontend-events',
data_format='JSONEachRow';

Then select the columns directly, without JSON parsing, e.g. select method from frontend_events_json For nested data, you can select headers:referrer from frontend_events_json

Explore the data in Kafka

Then you can scan incoming events via

select * from frontend_events

There are about 10 rows in each second. Only one column raw with sample data as following:

{
"version": 0,
"requestedUrl": "http://www.internationalinteractive.name/end-to-end",
"method": "PUT",
"correlationId": "0c7e970a-f65d-429a-9acf-6a136ce0a6ae",
"ipAddress": "186.58.241.7",
"requestDuration": 678,
"response": { "size": 2232, "statusCode": 200 },
"headers": {
"accept": "*/*",
"accept-encoding": "gzip",
"cache-control": "max-age=0",
"origin": "http://www.humanenvisioneer.com/engage/transparent/evolve/target",
"referrer": "http://www.centralharness.org/bandwidth/paradigms/target/whiteboard",
"user-agent": "Opera/10.41 (Macintosh; U; Intel Mac OS X 10_9_8; en-US) Presto/2.10.292 Version/13.00"
}
}

Cancel the query by pressing Ctrl+C.

Get live count

select count() from frontend_events

This query will show latest count every 2 seconds, without rescanning older data. This is a good example of incremental computation in Proton.

Filter events by JSON attributes

select _tp_time, raw:ipAddress, raw:requestedUrl from frontend_events where raw:method='POST'

Once you start the query, any new event with method value as POST will be selected. raw:key is a shortcut to extract string value from the JSON document. It also supports nested structure, such as raw:headers.accept

Aggregate data every second

select window_start, raw:method, count() from tumble(frontend_events,now(),1s)
group by window_start, raw:method

Every second, it will show the aggregation result for the number of events per HTTP method.

Show a live ASCII bar chart

Combining the interesting bar function from ClickHouse, you can use the following streaming SQL to visualize the top 5 HTTP methods per your clickstream.

select raw:method, count() as cnt, bar(cnt, 0, 40,5) as bar from frontend_events
group by raw:method order by cnt desc limit 5 by emit_version()
┌─raw:method─┬─cnt─┬─bar───┐
│ DELETE │ 35 │ ████▍ │
│ POST │ 29 │ ███▋ │
│ GET │ 27 │ ███▍ │
│ HEAD │ 25 │ ███ │
│ PUT │ 22 │ ██▋ │
└────────────┴─────┴───────┘

Note:

  • This is a global aggregation, emitting results every 2 seconds (configurable).
  • emit_version() function to show an auto-increasing number for each emit of streaming query result
  • limit 5 by emit_version() to get the first 5 rows with the same emit_version(). This is a special syntax in Proton. The regular limit 5 will cancel the entire SQL once 5 results are returned. But in this streaming SQL, we'd like to show 5 rows for each emit interval.

Create a materialized view to save notable events in Proton

With External Stream, you can query data in Kafka without saving the data in Proton. You can create a materialized view to selectively save some events, so that even the data in Kafka is removed, they remain available in Timeplus.

For example, the following SQL will create a materialized view to save those broken links with parsed attributes from JSON, such as URL, method, referrer.

create materialized view mv_broken_links as
select raw:requestedUrl as url,raw:method as method, raw:ipAddress as ip,
raw:response.statusCode as statusCode, domain(raw:headers.referrer) as referrer
from frontend_events where raw:response.statusCode<>'200';

Later on you can directly query on the materialized view:

-- streaming query
select * from mv_broken_links;

-- historical query
select method, count() as cnt, bar(cnt,0,40,5) as bar from table(mv_broken_links)
group by method order by cnt desc;
┌─method─┬─cnt─┬─bar─┐
│ GET │ 25 │ ███ │
│ DELETE │ 20 │ ██▌ │
│ HEAD │ 17 │ ██ │
│ POST │ 17 │ ██ │
│ PUT │ 17 │ ██ │
│ PATCH │ 17 │ ██ │
└────────┴─────┴─────┘

Streaming JOIN

In the owlshop-customers topic, there are a list of customers with the following metadata

  • id
  • firstName
  • lastName
  • gender
  • email

In the owlshop-addresses topic, it contains the detailed address for each customer

  • customer.id
  • street, state, city, zip
  • firstName, lastName

You can create a streaming JOIN to validate the data in these 2 topics matches to each other.

CREATE EXTERNAL STREAM customers(raw string)
SETTINGS type='kafka',
brokers='redpanda:9092',
topic='owlshop-customers';

CREATE EXTERNAL STREAM addresses(raw string)
SETTINGS type='kafka',
brokers='redpanda:9092',
topic='owlshop-addresses';

WITH parsed_customer AS (SELECT raw:id as id, raw:firstName||' '||raw:lastName as name,
raw:gender as gender FROM customers SETTINGS seek_to='earliest'),
parsed_addr AS (SELECT raw:customer.id as id, raw:street||' '||raw:city as addr,
raw:firstName||' '||raw:lastName as name FROM addresses SETTINGS seek_to='earliest')
SELECT * FROM parsed_customer JOIN parsed_addr USING(id);

Note:

  • Two CTE are defined to parse the JSON attribute as columns
  • SETTINGS seek_to='earliest' is the special settings to fetch earliest data from the Kafka topic
  • USING(id) is same as ON left.id=right.id
  • Check JOIN for more options to join dynamic and static data
info

By default, proton-client is started in single line and single query mode. To run multiple query statements together, start with the -n parameters, i.e. docker exec -it proton-container-name proton-client -n

Streaming ETL

If you don't want your data analysts to see the raw IP addresses for each requests, you can setup a streaming ETL process to mask the IP address, to protect such PII data (Personal Identifiable Information).

info

Require Proton 1.3.18 or above.

-- read the topic via an external stream
CREATE EXTERNAL STREAM frontend_events(raw string)
SETTINGS type='kafka',
brokers='redpanda:9092',
topic='owlshop-frontend-events';

-- create the other external stream to write data to the other topic
CREATE EXTERNAL STREAM target(
_tp_time datetime64(3),
url string,
method string,
ip string)
SETTINGS type='kafka',
brokers='redpanda:9092',
topic='masked-fe-event',
data_format='JSONEachRow';

-- setup the ETL pipeline via a materialized view
CREATE MATERIALIZED VIEW mv INTO target AS
SELECT now64() AS _tp_time,
raw:requestedUrl AS url,
raw:method AS method,
lower(hex(md5(raw:ipAddress))) AS ip
FROM frontend_events;

Properties for Kafka client

For more advanced use cases, you can specify customized properties while creating the external streams. Those properties will be passed to the underlying Kafka client, which is librdkafka.

For example:

CREATE EXTERNAL STREAM ext_github_events(raw string)
SETTINGS type='kafka',
brokers='localhost:9092',
topic='github_events',
properties='message.max.bytes=1000000;message.timeout.ms=6000'

Please note, not all properties in librdkafka are supported. The following ones are accepted in Proton today. Please check the configuration guide of librdkafka for details.

keyrangedefaultdescription
enable.idempotencetrue, falsetrueWhen set to true, the producer will ensure that messages are successfully produced exactly once and in the original produce order.
message.timeout.ms0 .. 21474836470Local message timeout.
queue.buffering.max.messages0 .. 2147483647Maximum number of messages allowed on the producer queue.
queue.buffering.max.kbytes1 .. 2147483647Maximum total message size sum allowed on the producer queue.
queue.buffering.max.ms0 .. 900000Delay in milliseconds to wait for messages in the producer queue to accumulate before constructing message batches (MessageSets) to transmit to brokers.
message.max.bytes1000 .. 1000000000Maximum Kafka protocol request message size.
message.send.max.retries0 .. 2147483647How many times to retry sending a failing Message.
retries0 .. 2147483647Alias for message.send.max.retries: How many times to retry sending a failing Message.
retry.backoff.ms1 .. 300000The backoff time in milliseconds before retrying a protocol reques
retry.backoff.max.ms1 .. 300000The max backoff time in milliseconds before retrying a protocol request,
batch.num.messages1 .. 1000000Maximum number of messages batched in one MessageSet.
batch.size1 .. 2147483647Maximum size (in bytes) of all messages batched in one MessageSet, including protocol framing overhead.
compression.codecnone, gzip, snappy, lz4, zstd, inheritCompression codec to use for compressing message sets. inherit = inherit global compression.codec configuration.
compression.typenone, gzip, snappy, lz4, zstdAlias for compression.codec: compression codec to use for compressing message sets.
compression.level-1 .. 12Compression level parameter for algorithm selected by configuration property compression.codec.
topic.metadata.refresh.interval.ms-1 .. 3600000Period of time in milliseconds at which topic and broker metadata is refreshed in order to proactively discover any new brokers, topics, partitions or partition leader changes.