GraalVM Eval
CertifiedExecute inline Python with GraalVM
GraalVM Eval
Execute inline Python with GraalVM
Runs inline Python inside the task JVM via GraalVM. Access runContext, logger, and rendered variables from the bindings; declare names in outputs to return them. File access through Python (open, os, pathlib, tempfile) is limited to the task working directory (the bundled standard library is read-only for scripts), except for native modules (see below); host class access is restricted to java.* and io.kestra.core.models.*, minus classes that reach files, processes, reflection, or the network. To upload a file, read its bytes and pass them to runContext.storage().putFile(java.io.ByteArrayInputStream(data), 'name'). Optional modules preload Python files from content or kestra:// URIs onto the module path. Supports C-extension-backed stdlib modules such as ssl, sqlite3, and lzma, which requires enabling native access at the GraalVM engine level. Standard OS process APIs (os.system, subprocess) stay blocked, but this grant cannot be scoped down further: a script that reaches native code directly (e.g. ctypes) can still execute arbitrary OS commands, read or write any file the worker can access, or manipulate process memory. Native modules also open files themselves, outside the working directory restriction: sqlite3.connect() can read or create a database file anywhere the worker can. Unlike Kestra's Script/Shell tasks, which typically run in an isolated container or process via a TaskRunner, this code runs inline in the worker JVM process itself, so it has direct access to the worker process's own memory, file descriptors, and any secrets or other task state resident in that JVM. Only run scripts from users already trusted with the flow's credentials and infrastructure, as with any script task.
type: io.kestra.plugin.graalvm.python.EvalExamples
Parse a downloaded JSON and update one of its fields.
id: parse_json_data
namespace: company.team
tasks:
- id: download
type: io.kestra.plugin.core.http.Download
uri: http://xkcd.com/info.0.json
- id: graal
type: io.kestra.plugin.graalvm.python.Eval
outputs:
- data
script: |
data = {{ read(outputs.download.uri )}}
data["next_month"] = int(data["month"]) + 1
Execute a Python script using the GraalVM scripting engine.
id: evalPython
namespace: company.team
tasks:
- id: evalPython
type: io.kestra.plugin.graalvm.python.Eval
outputs:
- out
- map
script: |
import java
import java.io.ByteArrayInputStream as ByteArrayInputStream
# types other than one coming from the Java SDK must be defined this way
Counter = java.type("io.kestra.core.models.executions.metrics.Counter")
logger.info('Task started')
runContext.metric(Counter.of('total', 666, 'name', 'bla'))
map = {'test': 'here'}
content = ByteArrayInputStream('Hello World'.encode('utf-8'))
out = runContext.storage().putFile(content, 'out.txt')
{"map": map, "out": out}Define a Python module, then execute a script that imports this module using the GraalVM scripting engine.
id: evalPython
namespace: company.team
tasks:
- id: evalPython
type: io.kestra.plugin.graalvm.python.Eval
modules:
hello.py: |
def hello(name):
return("Hello " + name)
script: |
import hello
logger.info(hello.hello("Kestra"))Properties
script *string
Script body to execute
Template-rendered source code run by GraalVM in the selected language; flow variables are resolved before execution
assets
Assets this task consumes as inputs or produces as outputs, for lineage tracking and the asset graph (Enterprise Edition). A flow declaring this property on a task is rejected in the open-source edition.
io.kestra.core.models.assets.AssetsDeclaration
IGNOREFAILWARNAsset failure behavior
Behavior applied to the task state when a declared asset fails to render, emit, or be persisted (e.g. a lock conflict): FAIL escalates it to FAILED, WARN (default) warns it if it would otherwise succeed, IGNORE leaves the state untouched.
Whether to auto-register assets referenced dynamically at runtime that are not statically declared in inputs or outputs.
The assets consumed as inputs.
io.kestra.core.models.assets.AssetIdentifier
1The assets produced as outputs.
io.kestra.plugin.ee.assets.Dataset
1150{}1150io.kestra.plugin.ee.assets.File
1150{}1150io.kestra.plugin.ee.assets.Table
1150{}1150io.kestra.plugin.ee.assets.VM
1150{}1150io.kestra.core.models.assets.External
1150{}1150io.kestra.core.models.assets.Custom
11501Custom asset type
{}1150modules object
Inline Python modules to preload
Map of filename to module content or a kestra:// URI; files are written to a temporary module path before execution
options object
Advanced GraalVM context options
Maps directly to GraalVM's Context.Builder#options: one entry per option key/value, for example python.WarnOptions or js.ecmascript-version.
Keys that would weaken the sandbox already enforced by this task are rejected at execution time: python.PosixModuleBackend, any key related to host access or class loading, and any key prefixed with engine. or sandbox..
outputs array
Names of script outputs
Rendered list of member names to pull from the evaluated value or bindings and expose as task outputs
Outputs
outputs object
The captured outputs as declared on the outputs task property
result object
Result of the evaluated expression
Metrics
records counter
countTracks a user defined numeric value emitted from the Python script, such as the number of processed records or computed results.
