CEL expression extensions

The CEL expression is configured to expose parts of the request, and some custom functions to make matching easier.

In addition to the custom function extension listed below, you can craft any valid CEL expression as defined by the cel-spec language definition

String functions

The upstream CEL implementation provides extensions to the CEL specification for manipulating strings.

For example:

'refs/heads/main'.split('/') // result = list ['refs', 'heads', 'main']
['refs', 'heads', 'main'].join('/') // result = string 'refs/heads/main'
'my place'.replace('my ',' ') // result = string 'place'
'this that another'.replace('th ',' ', 2) // result = 'is at another'

The replace overload allows an optional limit on replacements.

Notes on numbers in CEL expressions

One thing to be aware of is how numeric values are treated in CEL expressions, JSON numbers are decoded to CEL double values.

For example:

  "count": 2,
  "measure": 1.7

In the JSON above, both numbers are parsed as CEL double (Go float64) values.

This means that if you want to do integer arithmetic, you’ll need to use explicit conversion functions.

From the CEL specification:

Note that currently there are no automatic arithmetic conversions for the numeric types (int, uint, and double).

You can either explicitly convert the number, or add another double value e.g.

  - ref:
      name: cel
      - name: overlays:
        - key: count_plus_1
          expression: "body.count + 1.0"
        - key: count_plus_2
          expression: "int(body.count) + 2"
        - key: measure_times_3
          expression: "body.measure * 3.0"

These will be serialised back to JSON appropriately:

  "count_plus_1": 2,
  "count_plus_2": 3,
  "measure_times_3": 5.1

Error messages in conversions

The following example will generate an error with the JSON example.

  - ref:
      name: cel
      - name: overlays:
          - key: bad_measure_times_3
            expression: "body.measure * 3"

bad_measure_times_3 will fail with failed to evaluate overlay expression 'body.measure * 3': no such overload because there’s no automatic conversion.

CEL expression examples

Matching on an element in an array.

CEL provides several macros which can operate on JSON objects.

If you have a JSON body like this:

  "labels": [
      "name": "test-a"

You can use this in filters in the following ways:

  • filter: body.labels.exists(x, x.name == 'test-b') is true
  • filter: body.labels.exists(x, x.name == 'test-c') is false
  • filter: body.labels.exists_one(x, x.name.endsWith('-b')) is true
  • filter: body.labels.exists_one(x, x.name.startsWith('test-')) is false
  • filter: body.labels.all(x, x.name.startsWith('test-')) is true
  • filter: body.labels.all(x, x.name.endsWith('-b')) is false

You can also parse additional data from each of the labels:

- key: suffixes
  expression: "body.labels.map(x, x.name.substring(x.name.lastIndexOf('-')+1))"

This yields an array of ["a", "b"] in the suffixes extension key.

- key: filtered
  expression: "body.labels.filter(x, x.name.endsWith('-b'))"

This would add an extensions key filtered with only one of the labels.

    "name": "test-b"

cel-go extensions

All the functionality from the cel-go project’s CEL extension is available in your CEL expressions.

The following extensions are available:

cel-go Bytes

The cel-go project function base64.decode returns a CEL Bytes value.

To compare this to a string, you will need to convert it to a Bytes type:

base64.decode(body.b64value) == b'hello' # compare to Bytes literal
base64.decode(body.b64value) == bytes('hello') # convert to bytes.

Returning Bytes

If you decode a base64 string with the cel-go base64 decoder, the result will be a set of base64 decoded bytes. To ensure the result is encoded as a string you will need to explicitly convert it to a CEL string.

  - cel:
        - key: base64_decoded
          expression: "string(base64.decode(body.b64Value))"

This will correctly appear in the extension as the decoded version.

List of extensions

The body from the http.Request value is decoded to JSON and exposed, and the headers are also available.

Symbol Type Description Example
body map(string, dynamic) This is the decoded JSON body from the incoming http.Request exposed as a map of string keys to any value types.
body.value == 'test'
header map(string, list(string)) This is the request Header.
header['X-Test'][0] == 'test-value'
requestURL string This is the URL for the incoming HTTP request.

NOTE: The header value is a Go http.Header, which is defined as:

type Header map[string][]string

i.e. the header is a mapping of strings, to arrays of strings, see the match function on headers below for an extension that makes looking up headers easier.

List of extension functions

This lists custom functions that can be used from CEL expressions in the CEL interceptor.

Symbol Type Description Example
match header.match(string, string) -> bool Uses the canonical header matching from Go's http.Request to match the header against the value.
header.match('x-test', 'test-value')
canonical header.canonical(string) -> string Uses the canonical header matching from Go's http.Request to get the provided header name.
<string>.truncate(uint) -> string
Truncates a string to no more than the specified length.
<string>.split(string) -> list(string)
Splits a string on the provided separator value.
<list(string)>.join(string) -> string
Joins a list of strings on the provided separator value.
['body', 'refs', 'main'].join('/')
decodeb64 **deprecated: please use base64.decode**
<string>.decodeb64() -> string
Decodes a base64 encoded string.
<string>.compareSecret(string, string, string) -> bool
Constant-time comparison of strings against secrets, this will fetch the secret using the combination of namespace/name and compare the token key to the string using a cryptographic constant-time comparison..

The event-listener service account must have access to the secret. The parameters to the function are 1. the key within the secret, 2. the secret name, and 3. the namespace for the secret (optional, defaults to the namespace of the EventListener).

header.canonical('X-Secret-Token').compareSecret('', 'secret-name', 'namespace')
<string>.compareSecret(string, string) -> bool
This is almost identical to the version above, but only requires two arguments, the namespace is assumed to be the namespace for the event-listener.
header.canonical('X-Secret-Token').compareSecret('key', 'secret-name')
<string>.parseJSON() -> map<string, dyn>
This parses a string that contains a JSON body into a map which which can be subsequently used in other expressions.
'{"testing":"value"}'.parseJSON().testing == "value"
<string>.parseYAML() -> map<string, dyn>
This parses a string that contains a YAML body into a map which which can be subsequently used in other expressions.
'key1: value1\nkey2: value2\n'.parseYAML().key1 == "value"
<string>.parseURL() -> map<string, dyn>
This parses a string that contains a URL into a map with keys for the elements of the URL.
The resulting map will contain the following keys for this URL "https://user:pass@example.com/test/path?s=testing#first"
query{"s": "testing"}
queryStrings{"s": ["testing"]}
auth{"username": "user", "password": "pass"}
Note the difference between query and queryStrings, in query, multiple query params with the same name would be comma separated, for the case where a single string is provided, this will just be the single string value. For queryString the query param values are provided as a list, which can be accessed by indexing.
'https://example.com/test?query=testing'.parseURL().query['query'] == "testing"
<jsonObjectOrList>.marshalJSON() -> <string>
Returns the JSON encoding of 'jsonObjectOrList' as a string.
{"testing":"value"}.marshalJSON() == "{\"testing\": \"value\"}"
<jsonArray>.first() -> <jsonObject>
Returns the first element in the array.
[1, 2, 3, 4, 5].first() == 1
<jsonArray>.last() -> <jsonObject>
Returns the last element in the array.
[1, 2, 3, 4, 5].last() == 5
<string>.translate(string, string) -> <string>
Uses a regular expression to replace characters from the source string with characters from the replacements.
"This is $an Invalid5String ".translate("[^a-z0-9]+", "") == "hisisannvalid5tring"

"This is $an Invalid5String ".translate("[^a-z0-9]+", "ABC") == "ABChisABCisABCanABCnvalid5ABCtring"
<string>.lowerAscii() -> <string>
Returns a new string where all ASCII characters are lower-cased.
"TacoCat".lowerAscii() == "tacocat"

"TacoCÆt Xii".lowerAscii() == "tacocÆt xii"
<string>.upperAscii() -> <string>
Returns a new string where all ASCII characters are upper-cased.
"TacoCat".upperAscii() == "TACOCAT"

"TacoCÆt Xii".upperAscii() == "TACOCÆT XII"

Troubleshooting CEL expressions

You can use the cel-eval tool to evaluate your CEL expressions against a specific HTTP request.

To install the cel-eval tool use the following command:

$ go install github.com/tektoncd/triggers/cmd/cel-eval@latest

Below is an example of using the tool to evaluate a CEL expression:

$ cat testdata/expression.txt
body.test.nested == "value"

$ cat testdata/http.txt
POST /foo HTTP/1.1
Content-Length: 29
Content-Type: application/json
X-Header: tacocat

{"test": {"nested": "value"}}

$ cel-eval -e testdata/expression.txt -r testdata/http.txt