rpk topic consume

Consume records from topics.

Consuming records reads from any amount of input topics, formats each record according to --format, and prints them to STDOUT. The output formatter understands a wide variety of formats.

The default output format --format json is a special format that outputs each record as JSON.

Formatting

Formatting is based on percent escapes and modifiers. Slashes can be used for common escapes:

Escape Description

\t

Tabs

\n

Newlines

\r

Carriage returns

\\

Slashes

\xNN

Hex encoded characters

The percent encodings are represented like this:

Percent encoding Description

%t

Topic

%T

Topic length

%k

Key

%K

Key length

%v

Value

%V

Value length

%h

Begin the header specification

%H

Number of headers

%p

Partition

%o

Offset

%e

Leader epoch

%d

Timestamp (formatting described below)

%x

Producer ID

%y

Producer epoch

%[

Partition log start offset

%|

Partition last stable offset

%]

Partition high watermark

%%

Record attributes (formatting described below)

%a

Percent sign

%{

Left brace

%}

Right brace

%i

Number of records formatted

Modifiers

Text and numbers can be formatted in many different ways, and the default format can be changed within brace modifiers. %v prints a value, while %v{hex} prints the value hex encoded. %T prints the length of a topic in ASCII, while %T{big8} prints the length of the topic as an eight byte big endian.

All modifiers go within braces following a percent-escape.

Numbers

Formatting number values can have the following modifiers:

Format Description

ascii

Print the number as ASCII (default)

hex64

Sixteen hex characters

hex32

Eight hex characters

hex16

Four hex characters

hex8

Two hex characters

hex4

One hex character

big64

Eight byte big endian number

big32

Four byte big endian number

big16

Two byte big endian number

big8

Alias for byte

little64

Eight byte little endian number

little32

Four byte little endian number

little16

Two byte little endian number

little8

Alias for byte

byte

One byte number

bool

true if the number is non-zero, false if the number is zero

All numbers are truncated as necessary per the modifier. Printing %V{byte} for a length 256 value prints a single null, whereas printing %V{big8} prints the bytes 1 and 0.

When writing number sizes, the size corresponds to the size of the raw values, not the size of encoded values. %T% t{hex} for the topic foo prints 3 666f6f, not 6 666f6f.

Timestamps

By default, the timestamp field is printed as a millisecond number value. In addition to the number modifiers above, timestamps can be printed with either Go formatting:

%d{go[2006-01-02T15:04:05Z07:00]}
go

Or strftime formatting:

%d{strftime[%F]}
go

An arbitrary amount of brackets (or braces, or # symbols) can wrap your date formatting:

%d{strftime=== [%F] ===}
go

This prints [YYYY-MM-DD], while the surrounding three # on each side are used to wrap the formatting.

For more information on Go time formatting, see the Go documentation.

For more information on strftime formatting, run man strftime.

Attributes

Each record (or batch of records) has a set of possible attributes. Internally, these are packed into bit flags. Printing an attribute requires first selecting which attribute you want to print, and then optionally specifying how you want it to be printed:

%a{compression}
%a{compression;number}
%a{compression;big64}
%a{compression;hex8}
bash

Compression is by default printed as text (none, gzip, …​). Compression can be printed as a number with ;number, where number is any number formatting option described above. No compression is 0, gzip is 1, etc.

%a{timestamp-type}
%a{timestamp-type;big64}
bash

The record’s timestamp type prints as:

  • -1 for very old records (before timestamps existed)

  • 0 for client-generated timestamps

  • 1 for broker-generated timestamps

Number formatting can be controlled with ;number. 
%a{transactional-bit}
%a{transactional-bit;bool}
bash

Prints 1 if the record is a part of a transaction or 0 if it is not.

%a{control-bit}
%a{control-bit;bool}
bash

Prints 1 if the record is a commit marker or 0 if it is not.

Text

Text fields without modifiers default to writing the raw bytes. Alternatively, there are the following modifiers:

Modifier Description

%t{hex}

Hex encoding

%k{base64}

Base64 standard encoding

%k{base64raw}

Base64 encoding raw

%v{unpack[<bBhH>iIqQc.$]}

The unpack modifier has a further internal specification, similar to timestamps above.

Unpacking text can allow translating binary input into readable output. If a value is a big-endian uint32, %v prints the raw four bytes, while %v{unpack[>I]} prints the number in as ASCII. If unpacking exhausts the input before something is unpacked fully, an error message is appended to the output.

Headers

Headers are formatted with percent encoding inside of the modifier:

%h{%k=%v{hex}}

This prints all headers with a space before the key and after the value, an equals sign between the key and value, and with the value hex encoded. Header formatting actually just parses the internal format as a record format, so all of the above rules about %K, %V, text, and numbers apply.

Values

Values for consumed records can be omitted by using the --meta-only flag.

Tombstone records (records with a null value) have their value omitted from the JSON output by default. All other records, including those with an empty-string value (""), will have their values printed.

Offsets

The --offset flag allows for specifying where to begin consuming, and optionally, where to stop consuming. The literal words start and end specify consuming from the start and the end.

Offset Description

start

Consume from the beginning

end

Consume from the end

:end

Consume until the current end

+oo

Consume oo after the current start offset

-oo

Consume oo before the current end offset

oo

Consume after an exact offset

oo:

Alias for oo

:oo

Consume until an exact offset

o1:o2

Consume from exact offset o1 until exact offset o2

@t

Consume starting from a given timestamp

@t:

alias for @t

@:t

Consume until a given timestamp

@t1:t2

Consume from timestamp t1 until timestamp t2

Each timestamp option is evaluated until one succeeds.

Timestamp Description

13 digits

Parsed as a unix millisecond

9 digits

Parsed as a unix second

YYYY-MM-DD

Parsed as a day, UTC

YYYY-MM-DDTHH:MM:SSZ

Parsed as RFC3339, UTC; fractional seconds optional (.MMM)

-dur

Duration; from now (as t1) or from t1 (as t2)

dur

For t2 in @t1:t2, relative duration from t1

end

For t2 in @t1:t2, the current end of the partition

Durations are parsed simply:

3ms    three milliseconds
10s    ten seconds
9m     nine minutes
1h     one hour
1m3ms  one minute and three milliseconds

For example:

-o @2022-02-14:1h   consume 1h of time on Valentine's Day 2022
-o @-48h:-24h       consume from 2 days ago to 1 day ago
-o @-1m:end         consume from 1m ago until now
-o @:-1hr           consume from the start until an hour ago

Examples

A key and value, separated by a space and ending in newline:

-f '%k %v\n'

A key length as four big endian bytes and the key as hex:

-f '%K{big32}%k{hex}'

A little endian uint32 and a string unpacked from a value:

-f '%v{unpack[is$]}'

Usage

rpk topic consume TOPICS... [flags]
bash

Flags

Value Type Description

-b, --balancer

string

Group balancer to use if group consuming (range, roundrobin, sticky, cooperative-sticky) (default "cooperative-sticky").

--fetch-max-bytes

int32

Maximum amount of bytes per fetch request per broker (default 1048576).

--fetch-max-wait

duration

Maximum amount of time to wait when fetching from a broker before the broker replies (default 5s).

-f, --format

string

Output format (see --help for details) (default "json").

-g, --group

string

Group to use for consuming (incompatible with -p).

-h, --help

-

Help for consume.

--meta-only

-

Print all record info except the record value (for -f json).

-n, --num

int

Quit after consuming this number of records (0 is unbounded).

-o, --offset

string

Offset to consume from / to (start, end, 47, +2, -3) (default "start").

-p, --partitions

int32

int32Slice Comma delimited list of specific partitions to consume (default []).

--pretty-print

-

Pretty print each record over multiple lines (for -f json) (default true).

--print-control-records

-

Opt in to printing control records.

--rack

string

Rack to use for consuming, which opts into follower fetching.

--read-committed

-

Opt in to reading only committed offsets.

-r, --regex

-

Parse topics as regex; consume any topic that matches any expression.

--use-schema-registry

strings

[=key,value] If present, rpk will decode the key and the value with the schema registry. Also accepts use-schema-registry=key or use-schema-registry=value.

--config

string

Redpanda or rpk config file; default search paths are /var/lib/redpanda/.config/rpk/rpk.yaml, $PWD/redpanda.yaml, and /etc/redpanda/redpanda.yaml.

-X, --config-opt

stringArray

Override rpk configuration settings. See rpk -X or execute rpk -X help for inline detail or rpk -X list for terser detail.

--profile

string

Profile to use. See rpk profile for more details.

-v, --verbose

-

Enable verbose logging.