act/pkg/runner/job_executor.go
Sam Foo ceeb6c160c
Add support for service containers (#1949)
* Support services (#42)

Removed createSimpleContainerName and AutoRemove flag

Co-authored-by: Lunny Xiao <xiaolunwen@gmail.com>
Co-authored-by: Jason Song <i@wolfogre.com>
Reviewed-on: https://gitea.com/gitea/act/pulls/42
Reviewed-by: Jason Song <i@wolfogre.com>
Co-authored-by: Zettat123 <zettat123@gmail.com>
Co-committed-by: Zettat123 <zettat123@gmail.com>

* Support services options (#45)

Reviewed-on: https://gitea.com/gitea/act/pulls/45
Reviewed-by: Lunny Xiao <xiaolunwen@gmail.com>
Co-authored-by: Zettat123 <zettat123@gmail.com>
Co-committed-by: Zettat123 <zettat123@gmail.com>

* Support intepolation for `env` of `services` (#47)

Reviewed-on: https://gitea.com/gitea/act/pulls/47
Reviewed-by: Lunny Xiao <xiaolunwen@gmail.com>
Co-authored-by: Zettat123 <zettat123@gmail.com>
Co-committed-by: Zettat123 <zettat123@gmail.com>

* Support services `credentials` (#51)

If a service's image is from a container registry requires authentication, `act_runner` will need `credentials` to pull the image, see [documentation](https://docs.github.com/en/actions/using-workflows/workflow-syntax-for-github-actions#jobsjob_idservicesservice_idcredentials).
Currently, `act_runner` incorrectly uses the `credentials` of `containers` to pull services' images and the `credentials` of services won't be used, see the related code: 0c1f2edb99/pkg/runner/run_context.go (L228-L269)

Co-authored-by: Jason Song <i@wolfogre.com>
Reviewed-on: https://gitea.com/gitea/act/pulls/51
Reviewed-by: Jason Song <i@wolfogre.com>
Reviewed-by: Lunny Xiao <xiaolunwen@gmail.com>
Co-authored-by: Zettat123 <zettat123@gmail.com>
Co-committed-by: Zettat123 <zettat123@gmail.com>

* Add ContainerMaxLifetime and ContainerNetworkMode options

from: b9c20dcaa4

* Fix container network issue (#56)

Follow: https://gitea.com/gitea/act_runner/pulls/184
Close https://gitea.com/gitea/act_runner/issues/177

- `act` create new networks only if the value of `NeedCreateNetwork` is true, and remove these networks at last. `NeedCreateNetwork` is passed by `act_runner`. 'NeedCreateNetwork' is true only if  `container.network` in the configuration file of the `act_runner` is empty.
- In the `docker create` phase, specify the network to which containers will connect. Because, if not specify , container will connect to `bridge` network which is created automatically by Docker.
  - If the network is user defined network ( the value of `container.network` is empty or `<custom-network>`.  Because, the network created by `act` is also user defined network.), will also specify alias by `--network-alias`. The alias of service is `<service-id>`. So we can be access service container by `<service-id>:<port>` in the steps of job.
- Won't try to `docker network connect ` network after `docker start` any more.
  - Because on the one hand,  `docker network connect` applies only to user defined networks, if try to `docker network connect host <container-name>` will return error.
  - On the other hand, we just specify network in the stage of `docker create`, the same effect can be achieved.
- Won't try to remove containers and networks berfore  the stage of `docker start`, because the name of these containers and netwoks won't be repeat.

Co-authored-by: Jason Song <i@wolfogre.com>
Reviewed-on: https://gitea.com/gitea/act/pulls/56
Reviewed-by: Jason Song <i@wolfogre.com>
Co-authored-by: sillyguodong <gedong_1994@163.com>
Co-committed-by: sillyguodong <gedong_1994@163.com>

* Check volumes (#60)

This PR adds a `ValidVolumes` config. Users can specify the volumes (including bind mounts) that can be mounted to containers by this config.

Options related to volumes:
- [jobs.<job_id>.container.volumes](https://docs.github.com/en/actions/using-workflows/workflow-syntax-for-github-actions#jobsjob_idcontainervolumes)
- [jobs.<job_id>.services.<service_id>.volumes](https://docs.github.com/en/actions/using-workflows/workflow-syntax-for-github-actions#jobsjob_idservicesservice_idvolumes)

In addition, volumes specified by `options` will also be checked.

Currently, the following default volumes (see a72822b3f8/pkg/runner/run_context.go (L116-L166)) will be added to `ValidVolumes`:
- `act-toolcache`
- `<container-name>` and `<container-name>-env`
- `/var/run/docker.sock` (We need to add a new configuration to control whether the docker daemon can be mounted)

Co-authored-by: Jason Song <i@wolfogre.com>
Reviewed-on: https://gitea.com/gitea/act/pulls/60
Reviewed-by: Jason Song <i@wolfogre.com>
Co-authored-by: Zettat123 <zettat123@gmail.com>
Co-committed-by: Zettat123 <zettat123@gmail.com>

* Remove ContainerMaxLifetime; fix lint

* Remove unused ValidVolumes

* Remove ConnectToNetwork

* Add docker stubs

* Close docker clients to prevent file descriptor leaks

* Fix the error when removing network in self-hosted mode (#69)

Fixes https://gitea.com/gitea/act_runner/issues/255

Reviewed-on: https://gitea.com/gitea/act/pulls/69
Co-authored-by: Zettat123 <zettat123@gmail.com>
Co-committed-by: Zettat123 <zettat123@gmail.com>

* Move service container and network cleanup to rc.cleanUpJobContainer

* Add --network flag; default to host if not using service containers or set explicitly

* Correctly close executor to prevent fd leak

* Revert to tail instead of full path

* fix network duplication

* backport networkingConfig for aliaes

* don't hardcode netMode host

* Convert services test to table driven tests

* Add failing tests for services

* Expose service container ports onto the host

* Set container network mode in artifacts server test to host mode

* Log container network mode when creating/starting a container

* fix: Correctly handle ContainerNetworkMode

* fix: missing service container network

* Always remove service containers

Although we usually keep containers running if the workflow errored
(unless `--rm` is given) in order to facilitate debugging and we have
a flag (`--reuse`) to always keep containers running in order to speed
up repeated `act` invocations, I believe that these should only apply
to job containers and not service containers, because changing the
network settings on a service container requires re-creating it anyway.

* Remove networks only if no active endpoints exist

* Ensure job containers are stopped before starting a new job

* fix: go build -tags WITHOUT_DOCKER

---------

Co-authored-by: Zettat123 <zettat123@gmail.com>
Co-authored-by: Lunny Xiao <xiaolunwen@gmail.com>
Co-authored-by: Jason Song <i@wolfogre.com>
Co-authored-by: sillyguodong <gedong_1994@163.com>
Co-authored-by: ChristopherHX <christopher.homberger@web.de>
Co-authored-by: ZauberNerd <zaubernerd@zaubernerd.de>
2023-10-19 09:24:52 +00:00

199 lines
5.5 KiB
Go

package runner
import (
"context"
"fmt"
"time"
"github.com/nektos/act/pkg/common"
"github.com/nektos/act/pkg/model"
)
type jobInfo interface {
matrix() map[string]interface{}
steps() []*model.Step
startContainer() common.Executor
stopContainer() common.Executor
closeContainer() common.Executor
interpolateOutputs() common.Executor
result(result string)
}
//nolint:contextcheck,gocyclo
func newJobExecutor(info jobInfo, sf stepFactory, rc *RunContext) common.Executor {
steps := make([]common.Executor, 0)
preSteps := make([]common.Executor, 0)
var postExecutor common.Executor
steps = append(steps, func(ctx context.Context) error {
logger := common.Logger(ctx)
if len(info.matrix()) > 0 {
logger.Infof("\U0001F9EA Matrix: %v", info.matrix())
}
return nil
})
infoSteps := info.steps()
if len(infoSteps) == 0 {
return common.NewDebugExecutor("No steps found")
}
preSteps = append(preSteps, func(ctx context.Context) error {
// Have to be skipped for some Tests
if rc.Run == nil {
return nil
}
rc.ExprEval = rc.NewExpressionEvaluator(ctx)
// evaluate environment variables since they can contain
// GitHub's special environment variables.
for k, v := range rc.GetEnv() {
rc.Env[k] = rc.ExprEval.Interpolate(ctx, v)
}
return nil
})
for i, stepModel := range infoSteps {
stepModel := stepModel
if stepModel == nil {
return func(ctx context.Context) error {
return fmt.Errorf("invalid Step %v: missing run or uses key", i)
}
}
if stepModel.ID == "" {
stepModel.ID = fmt.Sprintf("%d", i)
}
step, err := sf.newStep(stepModel, rc)
if err != nil {
return common.NewErrorExecutor(err)
}
preSteps = append(preSteps, useStepLogger(rc, stepModel, stepStagePre, step.pre()))
stepExec := step.main()
steps = append(steps, useStepLogger(rc, stepModel, stepStageMain, func(ctx context.Context) error {
logger := common.Logger(ctx)
err := stepExec(ctx)
if err != nil {
logger.Errorf("%v", err)
common.SetJobError(ctx, err)
} else if ctx.Err() != nil {
logger.Errorf("%v", ctx.Err())
common.SetJobError(ctx, ctx.Err())
}
return nil
}))
postExec := useStepLogger(rc, stepModel, stepStagePost, step.post())
if postExecutor != nil {
// run the post executor in reverse order
postExecutor = postExec.Finally(postExecutor)
} else {
postExecutor = postExec
}
}
postExecutor = postExecutor.Finally(func(ctx context.Context) error {
jobError := common.JobError(ctx)
var err error
if rc.Config.AutoRemove || jobError == nil {
// always allow 1 min for stopping and removing the runner, even if we were cancelled
ctx, cancel := context.WithTimeout(common.WithLogger(context.Background(), common.Logger(ctx)), time.Minute)
defer cancel()
logger := common.Logger(ctx)
logger.Infof("Cleaning up container for job %s", rc.JobName)
if err = info.stopContainer()(ctx); err != nil {
logger.Errorf("Error while stop job container: %v", err)
}
}
setJobResult(ctx, info, rc, jobError == nil)
setJobOutputs(ctx, rc)
return err
})
pipeline := make([]common.Executor, 0)
pipeline = append(pipeline, preSteps...)
pipeline = append(pipeline, steps...)
return common.NewPipelineExecutor(info.startContainer(), common.NewPipelineExecutor(pipeline...).
Finally(func(ctx context.Context) error { //nolint:contextcheck
var cancel context.CancelFunc
if ctx.Err() == context.Canceled {
// in case of an aborted run, we still should execute the
// post steps to allow cleanup.
ctx, cancel = context.WithTimeout(common.WithLogger(context.Background(), common.Logger(ctx)), 5*time.Minute)
defer cancel()
}
return postExecutor(ctx)
}).
Finally(info.interpolateOutputs()).
Finally(info.closeContainer()))
}
func setJobResult(ctx context.Context, info jobInfo, rc *RunContext, success bool) {
logger := common.Logger(ctx)
jobResult := "success"
// we have only one result for a whole matrix build, so we need
// to keep an existing result state if we run a matrix
if len(info.matrix()) > 0 && rc.Run.Job().Result != "" {
jobResult = rc.Run.Job().Result
}
if !success {
jobResult = "failure"
}
info.result(jobResult)
if rc.caller != nil {
// set reusable workflow job result
rc.caller.runContext.result(jobResult)
}
jobResultMessage := "succeeded"
if jobResult != "success" {
jobResultMessage = "failed"
}
logger.WithField("jobResult", jobResult).Infof("\U0001F3C1 Job %s", jobResultMessage)
}
func setJobOutputs(ctx context.Context, rc *RunContext) {
if rc.caller != nil {
// map outputs for reusable workflows
callerOutputs := make(map[string]string)
ee := rc.NewExpressionEvaluator(ctx)
for k, v := range rc.Run.Workflow.WorkflowCallConfig().Outputs {
callerOutputs[k] = ee.Interpolate(ctx, ee.Interpolate(ctx, v.Value))
}
rc.caller.runContext.Run.Job().Outputs = callerOutputs
}
}
func useStepLogger(rc *RunContext, stepModel *model.Step, stage stepStage, executor common.Executor) common.Executor {
return func(ctx context.Context) error {
ctx = withStepLogger(ctx, stepModel.ID, rc.ExprEval.Interpolate(ctx, stepModel.String()), stage.String())
rawLogger := common.Logger(ctx).WithField("raw_output", true)
logWriter := common.NewLineWriter(rc.commandHandler(ctx), func(s string) bool {
if rc.Config.LogOutput {
rawLogger.Infof("%s", s)
} else {
rawLogger.Debugf("%s", s)
}
return true
})
oldout, olderr := rc.JobContainer.ReplaceLogWriter(logWriter, logWriter)
defer rc.JobContainer.ReplaceLogWriter(oldout, olderr)
return executor(ctx)
}
}