Files
gitea/services/actions/run.go
T
a15f032026 fix(actions): evaluate job-level if: before concurrency check (#39437)
Gitea doesn't evaluate a job's `if:` before checking the job's
concurrency group, which causes a job that should have been skipped to
incorrectly cancel other jobs in the same concurrency group.

This PR makes Gitea decide `if:` for every job before it becomes
waiting, including jobs without `needs` at insertion, on approval and on
rerun. A skipped job therefore no longer takes part in job concurrency
or holds a max-parallel slot, and a reusable caller whose `if:` is false
is no longer expanded on approval or rerun. An invalid `if:` skips the
job with an error summary.

After this PR, Gitea decides all jobs' `if:` expressions and sends `if:
always()` to the runner, so the runner no longer needs to evaluate a
job's `if:` again ([gitea/runner
`run_context.go`](https://gitea.com/gitea/runner/src/commit/81add274599355ec1838b6ebe45804890d40bab9/act/runner/run_context.go#L1195)).

---------

Co-authored-by: silverwind <me@silverwind.io>
2026-09-26 09:01:21 +02:00

300 lines
11 KiB
Go

// Copyright 2025 The Gitea Authors. All rights reserved.
// SPDX-License-Identifier: MIT
package actions
import (
"context"
"fmt"
act_model "gitea.dev/actionslib/pkg/model"
actions_model "gitea.dev/models/actions"
"gitea.dev/models/db"
"gitea.dev/modules/actions/jobparser"
"gitea.dev/modules/log"
"gitea.dev/modules/util"
"go.yaml.in/yaml/v4"
)
// PrepareRunAndInsert prepares a run and inserts it into the database
// It parses the workflow content, evaluates concurrency if needed, and inserts the run and its jobs into the database.
// The title will be cut off at 255 characters if it's longer than 255 characters.
func PrepareRunAndInsert(ctx context.Context, content []byte, run *actions_model.ActionRun, inputsWithDefaults map[string]any) error {
if run.WorkflowRepoID == 0 {
return fmt.Errorf("WorkflowRepoID must be set before insert (repo %d, workflow %q)", run.RepoID, run.WorkflowID)
}
if err := run.LoadAttributes(ctx); err != nil {
return fmt.Errorf("LoadAttributes: %w", err)
}
vars, err := actions_model.GetVariablesOfRun(ctx, run)
if err != nil {
return fmt.Errorf("GetVariablesOfRun: %w", err)
}
wfRawConcurrency, err := jobparser.ReadWorkflowRawConcurrency(content)
if err != nil {
return fmt.Errorf("ReadWorkflowRawConcurrency: %w", err)
}
if err = InsertRun(ctx, run, content, vars, inputsWithDefaults, wfRawConcurrency); err != nil {
return fmt.Errorf("InsertRun: %w", err)
}
// Load the newly inserted jobs with all fields from database (the job models in InsertRun are partial, so load again)
allJobs, err := db.Find[actions_model.ActionRunJob](ctx, actions_model.FindRunJobOptions{RunID: run.ID})
if err != nil {
return fmt.Errorf("FindRunJob: %w", err)
}
CreateCommitStatusForRunJobs(ctx, run, allJobs...)
NotifyWorkflowJobsAndRunsStatusUpdate(ctx, allJobs)
return nil
}
// InsertRun inserts a run
// The title will be cut off at 255 characters if it's longer than 255 characters.
func InsertRun(ctx context.Context, run *actions_model.ActionRun, content []byte, vars map[string]string, inputs map[string]any, wfRawConcurrency *act_model.RawConcurrency) error {
var cancelledConcurrencyJobs []*actions_model.ActionRunJob
var needPostCommitEmit bool
if err := db.WithTx(ctx, func(ctx context.Context) error {
index, err := db.GetNextResourceIndex(ctx, "action_run_index", run.RepoID)
if err != nil {
return err
}
run.Index = index
run.Title = util.EllipsisDisplayString(run.Title, 255)
run.Status = actions_model.StatusWaiting
if wfRawConcurrency != nil {
rawConcurrency, err := yaml.Marshal(wfRawConcurrency)
if err != nil {
return fmt.Errorf("marshal raw concurrency: %w", err)
}
run.RawConcurrency = string(rawConcurrency)
}
// Insert before parsing jobs or evaluating workflow-level concurrency
// so that run.ID is populated. Expressions referencing github.run_id —
// in run-name, job names, runs-on, or a workflow-level concurrency
// group like `${{ github.head_ref || github.run_id }}` — would otherwise
// interpolate to an empty string.
if err := db.Insert(ctx, run); err != nil {
return err
}
runAttempt := &actions_model.ActionRunAttempt{
RepoID: run.RepoID,
RunID: run.ID,
Attempt: 1,
TriggerUserID: run.TriggerUserID,
Status: actions_model.StatusWaiting,
}
if wfRawConcurrency != nil {
if err := EvaluateRunConcurrencyFillModel(ctx, run, runAttempt, wfRawConcurrency, vars, inputs); err != nil {
return fmt.Errorf("EvaluateRunConcurrencyFillModel: %w", err)
}
// check run (workflow-level) concurrency
var jobsToCancel []*actions_model.ActionRunJob
runAttempt.Status, jobsToCancel, err = PrepareToStartRunWithConcurrency(ctx, runAttempt)
if err != nil {
return err
}
cancelledConcurrencyJobs = append(cancelledConcurrencyJobs, jobsToCancel...)
}
if err := db.Insert(ctx, runAttempt); err != nil {
return err
}
run.LatestAttemptID = runAttempt.ID
giteaCtx := GenerateGiteaContext(ctx, run, runAttempt, nil)
jobs, err := jobparser.Parse(content, jobparser.WithVars(vars), jobparser.WithGitContext(giteaCtx.ToGitHubContext()), jobparser.WithInputs(inputs))
if err != nil {
return fmt.Errorf("parse workflow: %w", err)
}
titleChanged := len(jobs) > 0 && jobs[0].RunName != ""
if titleChanged {
run.Title = util.EllipsisDisplayString(jobs[0].RunName, 255)
}
cols := []string{"latest_attempt_id"}
if titleChanged {
cols = append(cols, "title")
}
if err := actions_model.UpdateRun(ctx, run, cols...); err != nil {
return err
}
runJobs := make([]*actions_model.ActionRunJob, 0, len(jobs))
var hasWaitingJobs bool
slots := maxParallelSlots{}
for _, v := range jobs {
runJob, jobsToCancel, jobNeedsPostCommitEmit, err := insertRunJob(ctx, run, runAttempt, v, vars, inputs, slots)
if err != nil {
return err
}
cancelledConcurrencyJobs = append(cancelledConcurrencyJobs, jobsToCancel...)
needPostCommitEmit = needPostCommitEmit || jobNeedsPostCommitEmit
// A reusable caller is never dispatched to a runner, so it must not drive the task-version bump.
hasWaitingJobs = hasWaitingJobs || (runJob.Status == actions_model.StatusWaiting && !runJob.IsReusableCaller)
runJobs = append(runJobs, runJob)
}
runAttempt.Status = actions_model.AggregateJobStatus(runJobs)
if err := actions_model.UpdateRunAttempt(ctx, runAttempt, "status"); err != nil {
return err
}
// if there is a job in the waiting status, increase tasks version.
if hasWaitingJobs {
if err := actions_model.IncreaseTaskVersion(ctx, run.OwnerID, run.RepoID); err != nil {
return err
}
}
return nil
}); err != nil {
return err
}
NotifyWorkflowJobsAndRunsStatusUpdate(ctx, cancelledConcurrencyJobs)
EmitJobsIfReadyByJobs(cancelledConcurrencyJobs)
// Post-commit kick: let the job emitter resolve jobs if needed
if needPostCommitEmit {
if err := EmitJobsIfReadyByRun(run.ID); err != nil {
log.Error("emit run %d after InsertRun: %v", run.ID, err)
}
}
return nil
}
// insertRunJob returns the inserted job, the jobs its concurrency cancelled, and whether a post-commit emitter pass is needed.
func insertRunJob(ctx context.Context, run *actions_model.ActionRun, runAttempt *actions_model.ActionRunAttempt, workflowJob *jobparser.SingleWorkflow, vars map[string]string, inputs map[string]any, slots maxParallelSlots) (*actions_model.ActionRunJob, []*actions_model.ActionRunJob, bool, error) {
id, job := workflowJob.Job()
needs := job.Needs()
isMatrixDeferred := jobparser.HasDeferredMatrix(job)
if err := workflowJob.SetJob(id, job.EraseNeeds()); err != nil {
return nil, nil, false, err
}
payload, _ := workflowJob.Marshal()
isReusableWorkflowCaller := job.Uses != ""
shouldBlockJob := runAttempt.Status == actions_model.StatusBlocked || len(needs) > 0 || run.NeedApproval
attemptJobID, err := actions_model.GetNextAttemptJobID(ctx, run.ID)
if err != nil {
return nil, nil, false, fmt.Errorf("alloc attempt_job_id: %w", err)
}
job.Name = job.DisplayName()
runJob := &actions_model.ActionRunJob{
RunID: run.ID,
RunAttemptID: runAttempt.ID,
RepoID: run.RepoID,
OwnerID: run.OwnerID,
CommitSHA: run.CommitSHA,
IsForkPullRequest: run.IsForkPullRequest,
Name: job.Name,
Attempt: runAttempt.Attempt,
WorkflowPayload: payload,
JobID: id,
AttemptJobID: attemptJobID,
Needs: needs,
RunsOn: job.RunsOn(),
Status: util.Iif(shouldBlockJob, actions_model.StatusBlocked, actions_model.StatusWaiting),
WorkflowSourceRepoID: run.WorkflowRepoID,
WorkflowSourceCommitSHA: run.WorkflowCommitSHA,
ContinueOnError: job.GetContinueOnError(),
IsMatrixDeferred: isMatrixDeferred,
MaxParallel: parseMaxParallel(id, job.Strategy.MaxParallelString),
}
if isMatrixDeferred {
// Expansion overwrites WorkflowPayload; keep the raw payload so a rerun can re-derive the matrix.
runJob.DeferredMatrixPayload = payload
}
// Parse workflow/job permissions (no clamping here)
if perms := ExtractJobPermissionsFromWorkflow(workflowJob, job); perms != nil {
runJob.TokenPermissions = perms
}
if isReusableWorkflowCaller {
runJob.IsReusableCaller = true
runJob.CallUses = job.Uses
}
// a skipped job must neither cancel its group peers nor take a slot
invalidIf, err := decideJobIf(ctx, run, runAttempt, runJob, vars)
if err != nil {
return nil, nil, false, fmt.Errorf("evaluate job if: %w", err)
}
var cancelledConcurrencyJobs []*actions_model.ActionRunJob
// check job concurrency
if job.RawConcurrency != nil {
rawConcurrency, err := yaml.Marshal(job.RawConcurrency)
if err != nil {
return nil, nil, false, fmt.Errorf("marshal raw concurrency: %w", err)
}
runJob.RawConcurrency = string(rawConcurrency)
// the job emitter evaluates it for jobs with `needs`, a skipped job never takes part
if len(needs) == 0 && runJob.Status != actions_model.StatusSkipped {
if err := EvaluateJobConcurrencyFillModel(ctx, run, runAttempt, runJob, vars, inputs); err != nil {
return nil, nil, false, fmt.Errorf("evaluate job concurrency: %w", err)
}
}
// If a job needs other jobs ("needs" is not empty), its status is set to StatusBlocked at the entry of the loop
// No need to check job concurrency for a blocked job (it will be checked by job emitter later)
// A slot-starved job skips the check too: it will not start, so it must not cancel its group peers.
if runJob.Status == actions_model.StatusWaiting && slots.available(runJob) {
var jobsToCancel []*actions_model.ActionRunJob
runJob.Status, jobsToCancel, err = PrepareToStartJobWithConcurrency(ctx, runJob)
if err != nil {
return nil, nil, false, fmt.Errorf("prepare to start job with concurrency: %w", err)
}
cancelledConcurrencyJobs = append(cancelledConcurrencyJobs, jobsToCancel...)
}
}
applyMaxParallel(runJob, slots)
if err := db.Insert(ctx, runJob); err != nil {
return nil, nil, false, err
}
if invalidIf != nil {
if err := upsertJobErrorSummary(ctx, runJob, "if", invalidIf); err != nil {
return nil, nil, false, err
}
}
if isReusableWorkflowCaller && runJob.Status == actions_model.StatusWaiting {
if err := expandInlineReusableCaller(ctx, run, runAttempt, runJob, vars); err != nil {
return nil, nil, false, err
}
}
// the emitter resolves an expanded caller's children and a skipped job's dependents
return runJob, cancelledConcurrencyJobs, runJob.IsExpanded || runJob.Status == actions_model.StatusSkipped, nil
}
func expandInlineReusableCaller(ctx context.Context, run *actions_model.ActionRun, runAttempt *actions_model.ActionRunAttempt, caller *actions_model.ActionRunJob, vars map[string]string) error {
if err := expandReusableWorkflowCaller(ctx, run, runAttempt, caller, vars); err != nil {
return fmt.Errorf("inline trigger caller %d ready: %w", caller.ID, err)
}
if err := actions_model.RefreshReusableCallerStatus(ctx, caller); err != nil {
return fmt.Errorf("refresh caller %d status: %w", caller.ID, err)
}
return nil
}