Files
act_runner/pkg/runner/runner.go
Pascal Zimmermann c0f19d9a26 fix: Max parallel Support for Matrix Jobs and Remote Action Tests (#150)
## Summary

This PR fixes the `max-parallel` strategy configuration for matrix jobs and resolves all failing tests in `step_action_remote_test.go`. The implementation ensures that matrix jobs respect the `max-parallel` setting, preventing resource exhaustion when running GitHub Actions workflows.

## Problem Statement

### Issue 1: max-parallel Not Working Correctly
Matrix jobs were running in parallel regardless of the `max-parallel` setting in the strategy configuration. This caused:
- Resource contention on limited runners
- Unpredictable job execution behavior
- Inability to control concurrency for resource-intensive workflows

### Issue 2: Failing Remote Action Tests
All tests in `step_action_remote_test.go` were failing due to:
- Missing `ActionCacheDir` configuration
- Incorrect mock expectations using fixed strings instead of hash-based paths
- Incompatibility with the hash-based action cache implementation

## Changes

### 1. max-parallel Implementation (`pkg/runner/runner.go`)

#### Robust Initialization
Added fallback logic to ensure `MaxParallel` is always properly initialized:
```go
if job.Strategy.MaxParallel == 0 {
    job.Strategy.MaxParallel = job.Strategy.GetMaxParallel()
}
```

#### Eliminated Unnecessary Nesting
Fixed inefficient nested parallelization when only one pipeline element exists:
```go
if len(pipeline) == 1 {
    // Execute directly without additional wrapper
    log.Debugf("Single pipeline element, executing directly")
    return pipeline[0](ctx)
}
```

#### Enhanced Logging
Added comprehensive debug and info logging:
- Shows which `maxParallel` value is being used
- Logs adjustments based on matrix size
- Reports final parallelization decisions

### 2. Worker Logging (`pkg/common/executor.go`)

Enhanced `NewParallelExecutor` with detailed worker activity logging:
```go
log.Infof("NewParallelExecutor: Creating %d workers for %d executors", parallel, len(executors))

for i := 0; i < parallel; i++ {
    go func(workerID int, work <-chan Executor, errs chan<- error) {
        log.Debugf("Worker %d started", workerID)
        taskCount := 0
        for executor := range work {
            taskCount++
            log.Debugf("Worker %d executing task %d", workerID, taskCount)
            errs <- executor(ctx)
        }
        log.Debugf("Worker %d finished (%d tasks executed)", workerID, taskCount)
    }(i, work, errs)
}
```

**Benefits:**
- Easy verification of correct parallelization
- Better debugging capabilities
- Clear visibility into worker activity

### 3. Documentation (`pkg/model/workflow.go`)

Added clarifying comment to ensure strategy values are always set:
```go
// Always set these values, even if there's an error later
j.Strategy.FailFast = j.Strategy.GetFailFast()
j.Strategy.MaxParallel = j.Strategy.GetMaxParallel()
```

### 4. Test Fixes (`pkg/runner/step_action_remote_test.go`)

#### Added Missing Configuration
All tests now include `ActionCacheDir`:
```go
Config: &Config{
    GitHubInstance: "github.com",
    ActionCacheDir: "/tmp/test-cache",
}
```

#### Hash-Based Suffix Matchers
Updated mocks to use hash-based paths instead of fixed strings:
```go
// Before
sarm.On("readAction", sar.Step, suffixMatcher("org-repo-path@ref"), ...)

// After
sarm.On("readAction", sar.Step, suffixMatcher(sar.Step.UsesHash()), ...)
```

#### Flexible Exec Matcher for Post Tests
Implemented flexible path matching for hash-based action directories:
```go
execMatcher := mock.MatchedBy(func(args []string) bool {
    if len(args) != 2 {
        return false
    }
    return args[0] == "node" && strings.Contains(args[1], "post.js")
})
```

#### Token Test Improvements
- Uses unique cache directory to force cloning
- Validates URL redirection to github.com
- Accepts realistic token behavior

### 5. New Tests

#### Unit Tests (`pkg/runner/max_parallel_test.go`)
Tests various `max-parallel` configurations:
- `max-parallel: 1` → Sequential execution
- `max-parallel: 2` → Max 2 parallel jobs
- `max-parallel: 4` (default) → Max 4 parallel jobs
- `max-parallel: 10` → Max 10 parallel jobs

#### Concurrency Test (`pkg/common/executor_max_parallel_test.go`)
Verifies that `max-parallel: 2` actually limits concurrent execution using atomic counters.

## Expected Behavior

### Before
-  Jobs ran in parallel regardless of `max-parallel` setting
-  Unnecessary nested parallelization (8 workers for 1 element)
-  No logging to debug parallelization issues
-  All `step_action_remote_test.go` tests failing

### After
-  `max-parallel: 1` → Jobs run strictly sequentially
-  `max-parallel: N` → Maximum N jobs run simultaneously
-  Efficient single-level parallelization for matrix jobs
-  Comprehensive logging for debugging
-  All tests passing (10/10)

## Example Log Output

With `max-parallel: 2` and 6 matrix jobs:
```
[DEBUG] Using job.Strategy.MaxParallel: 2
[INFO] Running job with maxParallel=2 for 6 matrix combinations
[DEBUG] Single pipeline element, executing directly
[INFO] NewParallelExecutor: Creating 2 workers for 6 executors
[DEBUG] Worker 0 started
[DEBUG] Worker 1 started
[DEBUG] Worker 0 executing task 1
[DEBUG] Worker 1 executing task 1
...
[DEBUG] Worker 0 finished (3 tasks executed)
[DEBUG] Worker 1 finished (3 tasks executed)
```

## Test Results

All tests pass successfully:
```
 TestStepActionRemote (3 sub-tests)
 TestStepActionRemotePre
 TestStepActionRemotePreThroughAction
 TestStepActionRemotePreThroughActionToken
 TestStepActionRemotePost (4 sub-tests)
 TestMaxParallelStrategy
 TestMaxParallel2Quick

Total: 12/12 tests passing
```

## Breaking Changes

None. This PR is fully backward compatible. All changes improve existing behavior without altering the API.

## Impact

-  Fixes resource management for CI/CD pipelines
-  Prevents runner exhaustion on limited infrastructure
-  Enables sequential execution for resource-intensive jobs
-  Improves debugging with detailed logging
-  Ensures test suite reliability

## Files Modified

### Core Implementation
- `pkg/runner/runner.go` - max-parallel fix + logging
- `pkg/common/executor.go` - Worker logging
- `pkg/model/workflow.go` - Documentation

### Tests
- `pkg/runner/step_action_remote_test.go` - Fixed all 10 tests
- `pkg/runner/max_parallel_test.go` - **NEW** - Unit tests
- `pkg/common/executor_max_parallel_test.go` - **NEW** - Concurrency test

## Verification

### Manual Testing
```bash
# Build
go build -o dist/act main.go

# Test with max-parallel: 2
./dist/act -W test-max-parallel-2.yml -v

# Expected: Only 2 jobs run simultaneously
```

### Automated Testing
```bash
# Run all tests
go test ./pkg/runner -run TestStepActionRemote -v
go test ./pkg/runner -run TestMaxParallel -v
go test ./pkg/common -run TestMaxParallel -v
```

## Related Issues

Fixes issues where matrix jobs in Gitea ignored the `max-parallel` strategy setting, causing resource contention and unpredictable behavior.

Reviewed-on: https://gitea.com/gitea/act/pulls/150
Reviewed-by: Lunny Xiao <xiaolunwen@gmail.com>
Reviewed-by: silverwind <silverwind@noreply.gitea.com>
Co-authored-by: Pascal Zimmermann <pascal.zimmermann@theiotstudio.com>
Co-committed-by: Pascal Zimmermann <pascal.zimmermann@theiotstudio.com>
2026-02-11 00:43:38 +00:00

330 lines
14 KiB
Go

package runner
import (
"context"
"encoding/json"
"fmt"
"os"
"runtime"
"sync"
"time"
docker_container "github.com/docker/docker/api/types/container"
log "github.com/sirupsen/logrus"
"github.com/nektos/act/pkg/common"
"github.com/nektos/act/pkg/model"
)
// Runner provides capabilities to run GitHub actions
type Runner interface {
NewPlanExecutor(plan *model.Plan) common.Executor
}
// Config contains the config for a new runner
type Config struct {
Actor string // the user that triggered the event
Workdir string // path to working directory
ActionCacheDir string // path used for caching action contents
ActionOfflineMode bool // when offline, use caching action contents
BindWorkdir bool // bind the workdir to the job container
EventName string // name of event to run
EventPath string // path to JSON file to use for event.json in containers
DefaultBranch string // name of the main branch for this repository
ReuseContainers bool // reuse containers to maintain state
ForcePull bool // force pulling of the image, even if already present
ForceRebuild bool // force rebuilding local docker image action
LogOutput bool // log the output from docker run
JSONLogger bool // use json or text logger
LogPrefixJobID bool // switches from the full job name to the job id
Env map[string]string // env for containers
Inputs map[string]string // manually passed action inputs
Secrets map[string]string // list of secrets
Vars map[string]string // list of vars
Token string // GitHub token
InsecureSecrets bool // switch hiding output when printing to terminal
Platforms map[string]string // list of platforms
Privileged bool // use privileged mode
UsernsMode string // user namespace to use
ContainerArchitecture string // Desired OS/architecture platform for running containers
ContainerDaemonSocket string // Path to Docker daemon socket
ContainerOptions string // Options for the job container
UseGitIgnore bool // controls if paths in .gitignore should not be copied into container, default true
GitHubInstance string // GitHub instance to use, default "github.com"
ContainerCapAdd []string // list of kernel capabilities to add to the containers
ContainerCapDrop []string // list of kernel capabilities to remove from the containers
AutoRemove bool // controls if the container is automatically removed upon workflow completion
ArtifactServerPath string // the path where the artifact server stores uploads
ArtifactServerAddr string // the address the artifact server binds to
ArtifactServerPort string // the port the artifact server binds to
NoSkipCheckout bool // do not skip actions/checkout
RemoteName string // remote name in local git repo config
ReplaceGheActionWithGithubCom []string // Use actions from GitHub Enterprise instance to GitHub
ReplaceGheActionTokenWithGithubCom string // Token of private action repo on GitHub.
Matrix map[string]map[string]bool // Matrix config to run
ContainerNetworkMode docker_container.NetworkMode // the network mode of job containers (the value of --network)
ActionCache ActionCache // Use a custom ActionCache Implementation
PresetGitHubContext *model.GithubContext // the preset github context, overrides some fields like DefaultBranch, Env, Secrets etc.
EventJSON string // the content of JSON file to use for event.json in containers, overrides EventPath
ContainerNamePrefix string // the prefix of container name
ContainerMaxLifetime time.Duration // the max lifetime of job containers
DefaultActionInstance string // the default actions web site
PlatformPicker func(labels []string) string // platform picker, it will take precedence over Platforms if isn't nil
JobLoggerLevel *log.Level // the level of job logger
ValidVolumes []string // only volumes (and bind mounts) in this slice can be mounted on the job container or service containers
InsecureSkipTLS bool // whether to skip verifying TLS certificate of the Gitea instance
MaxParallel int // max parallel jobs to run across all workflows (0 = no limit, uses CPU count)
}
// GetToken: Adapt to Gitea
func (c Config) GetToken() string {
token := c.Secrets["GITHUB_TOKEN"]
if c.Secrets["GITEA_TOKEN"] != "" {
token = c.Secrets["GITEA_TOKEN"]
}
return token
}
type caller struct {
runContext *RunContext
updateResultLock sync.Mutex // For Gitea
reusedWorkflowJobResults map[string]string // For Gitea
}
type runnerImpl struct {
config *Config
eventJSON string
caller *caller // the job calling this runner (caller of a reusable workflow)
}
// New Creates a new Runner
func New(runnerConfig *Config) (Runner, error) {
runner := &runnerImpl{
config: runnerConfig,
}
return runner.configure()
}
func (runner *runnerImpl) configure() (Runner, error) {
runner.eventJSON = "{}"
if runner.config.EventJSON != "" {
runner.eventJSON = runner.config.EventJSON
} else if runner.config.EventPath != "" {
log.Debugf("Reading event.json from %s", runner.config.EventPath)
eventJSONBytes, err := os.ReadFile(runner.config.EventPath)
if err != nil {
return nil, err
}
runner.eventJSON = string(eventJSONBytes)
} else if len(runner.config.Inputs) != 0 {
eventMap := map[string]map[string]string{
"inputs": runner.config.Inputs,
}
eventJSON, err := json.Marshal(eventMap)
if err != nil {
return nil, err
}
runner.eventJSON = string(eventJSON)
}
return runner, nil
}
// NewPlanExecutor ...
func (runner *runnerImpl) NewPlanExecutor(plan *model.Plan) common.Executor {
maxJobNameLen := 0
stagePipeline := make([]common.Executor, 0)
log.Debugf("Plan Stages: %v", plan.Stages)
for i := range plan.Stages {
stage := plan.Stages[i]
stagePipeline = append(stagePipeline, func(ctx context.Context) error {
pipeline := make([]common.Executor, 0)
for _, run := range stage.Runs {
log.Debugf("Stages Runs: %v", stage.Runs)
stageExecutor := make([]common.Executor, 0)
job := run.Job()
log.Debugf("Job.Name: %v", job.Name)
log.Debugf("Job.RawNeeds: %v", job.RawNeeds)
log.Debugf("Job.RawRunsOn: %v", job.RawRunsOn)
log.Debugf("Job.Env: %v", job.Env)
log.Debugf("Job.If: %v", job.If)
for step := range job.Steps {
if nil != job.Steps[step] {
log.Debugf("Job.Steps: %v", job.Steps[step].String())
}
}
log.Debugf("Job.TimeoutMinutes: %v", job.TimeoutMinutes)
log.Debugf("Job.Services: %v", job.Services)
log.Debugf("Job.Strategy: %v", job.Strategy)
log.Debugf("Job.RawContainer: %v", job.RawContainer)
log.Debugf("Job.Defaults.Run.Shell: %v", job.Defaults.Run.Shell)
log.Debugf("Job.Defaults.Run.WorkingDirectory: %v", job.Defaults.Run.WorkingDirectory)
log.Debugf("Job.Outputs: %v", job.Outputs)
log.Debugf("Job.Uses: %v", job.Uses)
log.Debugf("Job.With: %v", job.With)
// log.Debugf("Job.RawSecrets: %v", job.RawSecrets)
log.Debugf("Job.Result: %v", job.Result)
if job.Strategy != nil {
log.Debugf("Job.Strategy.FailFast: %v", job.Strategy.FailFast)
log.Debugf("Job.Strategy.MaxParallel: %v", job.Strategy.MaxParallel)
log.Debugf("Job.Strategy.FailFastString: %v", job.Strategy.FailFastString)
log.Debugf("Job.Strategy.MaxParallelString: %v", job.Strategy.MaxParallelString)
log.Debugf("Job.Strategy.RawMatrix: %v", job.Strategy.RawMatrix)
strategyRc := runner.newRunContext(ctx, run, nil)
if err := strategyRc.NewExpressionEvaluator(ctx).EvaluateYamlNode(ctx, &job.Strategy.RawMatrix); err != nil {
log.Errorf("Error while evaluating matrix: %v", err)
}
}
var matrixes []map[string]interface{}
if m, err := job.GetMatrixes(); err != nil {
log.Errorf("Error while get job's matrix: %v", err)
} else {
log.Debugf("Job Matrices: %v", m)
log.Debugf("Runner Matrices: %v", runner.config.Matrix)
matrixes = selectMatrixes(m, runner.config.Matrix)
}
log.Debugf("Final matrix after applying user inclusions '%v'", matrixes)
maxParallel := 4
if job.Strategy != nil {
// Ensure GetMaxParallel() is called if MaxParallel is still 0
if job.Strategy.MaxParallel == 0 {
job.Strategy.MaxParallel = job.Strategy.GetMaxParallel()
}
maxParallel = job.Strategy.MaxParallel
log.Debugf("Using job.Strategy.MaxParallel: %d", maxParallel)
}
if len(matrixes) < maxParallel {
log.Debugf("Adjusting maxParallel from %d to %d (number of matrix combinations)", maxParallel, len(matrixes))
maxParallel = len(matrixes)
}
log.Infof("Running job with maxParallel=%d for %d matrix combinations", maxParallel, len(matrixes))
for i, matrix := range matrixes {
matrix := matrix
rc := runner.newRunContext(ctx, run, matrix)
rc.JobName = rc.Name
if len(matrixes) > 1 {
rc.Name = fmt.Sprintf("%s-%d", rc.Name, i+1)
}
if len(rc.String()) > maxJobNameLen {
maxJobNameLen = len(rc.String())
}
if rc.caller != nil { // For Gitea
rc.caller.setReusedWorkflowJobResult(rc.JobName, "pending")
}
stageExecutor = append(stageExecutor, func(ctx context.Context) error {
jobName := fmt.Sprintf("%-*s", maxJobNameLen, rc.String())
executor, err := rc.Executor()
if err != nil {
return err
}
return executor(common.WithJobErrorContainer(WithJobLogger(ctx, rc.Run.JobID, jobName, rc.Config, &rc.Masks, matrix)))
})
}
pipeline = append(pipeline, common.NewParallelExecutor(maxParallel, stageExecutor...))
}
// For pipeline execution:
// - If only 1 element: run it directly (no need for additional parallelization)
// - If multiple elements: run them in parallel up to maxParallel or ncpu
if len(pipeline) == 0 {
return nil
}
if len(pipeline) == 1 {
// Single run/job: execute directly without additional parallelization wrapper
// This ensures max-parallel is the only limiting factor
log.Debugf("Single pipeline element, executing directly")
return pipeline[0](ctx)
}
// Multiple runs/jobs: execute in parallel up to maxParallel (if set) or ncpu
parallelism := runtime.NumCPU()
// If MaxParallel is set in config, use it
if runner.config.MaxParallel > 0 {
parallelism = runner.config.MaxParallel
log.Debugf("Using configured max-parallel: %d", parallelism)
} else {
log.Debugf("Using CPU count for parallelism: %d", parallelism)
}
// Don't exceed the number of pipeline elements
if parallelism > len(pipeline) {
parallelism = len(pipeline)
}
log.Infof("Executing %d pipeline elements with parallelism %d", len(pipeline), parallelism)
return common.NewParallelExecutor(parallelism, pipeline...)(ctx)
})
}
return common.NewPipelineExecutor(stagePipeline...).Then(handleFailure(plan))
}
func handleFailure(plan *model.Plan) common.Executor {
return func(ctx context.Context) error {
for _, stage := range plan.Stages {
for _, run := range stage.Runs {
if run.Job().Result == "failure" {
return fmt.Errorf("Job '%s' failed", run.String())
}
}
}
return nil
}
}
func selectMatrixes(originalMatrixes []map[string]interface{}, targetMatrixValues map[string]map[string]bool) []map[string]interface{} {
matrixes := make([]map[string]interface{}, 0)
for _, original := range originalMatrixes {
flag := true
for key, val := range original {
if allowedVals, ok := targetMatrixValues[key]; ok {
valToString := fmt.Sprintf("%v", val)
if _, ok := allowedVals[valToString]; !ok {
flag = false
}
}
}
if flag {
matrixes = append(matrixes, original)
}
}
return matrixes
}
func (runner *runnerImpl) newRunContext(ctx context.Context, run *model.Run, matrix map[string]interface{}) *RunContext {
rc := &RunContext{
Config: runner.config,
Run: run,
EventJSON: runner.eventJSON,
StepResults: make(map[string]*model.StepResult),
Matrix: matrix,
caller: runner.caller,
}
rc.ExprEval = rc.NewExpressionEvaluator(ctx)
rc.Name = rc.ExprEval.Interpolate(ctx, run.String())
return rc
}
// For Gitea
func (c *caller) setReusedWorkflowJobResult(jobName string, result string) {
c.updateResultLock.Lock()
defer c.updateResultLock.Unlock()
c.reusedWorkflowJobResults[jobName] = result
}