Fix findEndpointForJob

Signed-off-by: Gabriel Adrian Samfira <gsamfira@cloudbasesolutions.com>
This commit is contained in:
Gabriel Adrian Samfira 2025-05-14 21:09:02 +00:00
parent 56be5eb698
commit f66b651b59
13 changed files with 287 additions and 113 deletions

View file

@ -107,8 +107,15 @@ func (a *APIController) handleWorkflowJobEvent(ctx context.Context, w http.Respo
signature := r.Header.Get("X-Hub-Signature-256") signature := r.Header.Get("X-Hub-Signature-256")
hookType := r.Header.Get("X-Github-Hook-Installation-Target-Type") hookType := r.Header.Get("X-Github-Hook-Installation-Target-Type")
giteaTargetType := r.Header.Get("X-Gitea-Hook-Installation-Target-Type")
if err := a.r.DispatchWorkflowJob(hookType, signature, body); err != nil { forgeType := runnerParams.GithubEndpointType
if giteaTargetType != "" {
forgeType = runnerParams.GiteaEndpointType
hookType = giteaTargetType
}
if err := a.r.DispatchWorkflowJob(hookType, signature, forgeType, body); err != nil {
switch { switch {
case errors.Is(err, gErrors.ErrNotFound): case errors.Is(err, gErrors.ErrNotFound):
metrics.WebhooksReceived.WithLabelValues( metrics.WebhooksReceived.WithLabelValues(

12
cache/cache_test.go vendored
View file

@ -55,8 +55,8 @@ func (c *CacheTestSuite) TestSetCacheWorks() {
c.Require().Len(githubToolsCache.entities, 0) c.Require().Len(githubToolsCache.entities, 0)
SetGithubToolsCache(c.entity, tools) SetGithubToolsCache(c.entity, tools)
c.Require().Len(githubToolsCache.entities, 1) c.Require().Len(githubToolsCache.entities, 1)
cachedTools, ok := GetGithubToolsCache(c.entity.ID) cachedTools, err := GetGithubToolsCache(c.entity.ID)
c.Require().True(ok) c.Require().NoError(err)
c.Require().Len(cachedTools, 1) c.Require().Len(cachedTools, 1)
c.Require().Equal(tools[0].GetDownloadURL(), cachedTools[0].GetDownloadURL()) c.Require().Equal(tools[0].GetDownloadURL(), cachedTools[0].GetDownloadURL())
} }
@ -76,16 +76,16 @@ func (c *CacheTestSuite) TestTimedOutToolsCache() {
entity.updatedAt = entity.updatedAt.Add(-2 * time.Hour) entity.updatedAt = entity.updatedAt.Add(-2 * time.Hour)
githubToolsCache.entities[c.entity.ID] = entity githubToolsCache.entities[c.entity.ID] = entity
cachedTools, ok := GetGithubToolsCache(c.entity.ID) cachedTools, err := GetGithubToolsCache(c.entity.ID)
c.Require().False(ok) c.Require().NoError(err)
c.Require().Nil(cachedTools) c.Require().Nil(cachedTools)
} }
func (c *CacheTestSuite) TestGetInexistentCache() { func (c *CacheTestSuite) TestGetInexistentCache() {
c.Require().NotNil(githubToolsCache) c.Require().NotNil(githubToolsCache)
c.Require().Len(githubToolsCache.entities, 0) c.Require().Len(githubToolsCache.entities, 0)
cachedTools, ok := GetGithubToolsCache(c.entity.ID) cachedTools, err := GetGithubToolsCache(c.entity.ID)
c.Require().False(ok) c.Require().NoError(err)
c.Require().Nil(cachedTools) c.Require().Nil(cachedTools)
} }

40
cache/tools_cache.go vendored
View file

@ -1,6 +1,7 @@
package cache package cache
import ( import (
"fmt"
"sync" "sync"
"time" "time"
@ -20,17 +21,25 @@ func init() {
type GithubEntityTools struct { type GithubEntityTools struct {
updatedAt time.Time updatedAt time.Time
expiresAt time.Time expiresAt time.Time
err error
entity params.ForgeEntity entity params.ForgeEntity
tools []commonParams.RunnerApplicationDownload tools []commonParams.RunnerApplicationDownload
} }
func (g GithubEntityTools) Error() string {
if g.err != nil {
return g.err.Error()
}
return ""
}
type GithubToolsCache struct { type GithubToolsCache struct {
mux sync.Mutex mux sync.Mutex
// entity IDs are UUID4s. It is highly unlikely they will collide (🤞). // entity IDs are UUID4s. It is highly unlikely they will collide (🤞).
entities map[string]GithubEntityTools entities map[string]GithubEntityTools
} }
func (g *GithubToolsCache) Get(entityID string) ([]commonParams.RunnerApplicationDownload, bool) { func (g *GithubToolsCache) Get(entityID string) ([]commonParams.RunnerApplicationDownload, error) {
g.mux.Lock() g.mux.Lock()
defer g.mux.Unlock() defer g.mux.Unlock()
@ -39,12 +48,12 @@ func (g *GithubToolsCache) Get(entityID string) ([]commonParams.RunnerApplicatio
if time.Now().UTC().After(cache.expiresAt.Add(-5 * time.Minute)) { if time.Now().UTC().After(cache.expiresAt.Add(-5 * time.Minute)) {
// Stale cache, remove it. // Stale cache, remove it.
delete(g.entities, entityID) delete(g.entities, entityID)
return nil, false return nil, fmt.Errorf("cache expired for entity %s", entityID)
} }
} }
return cache.tools, true return cache.tools, cache.err
} }
return nil, false return nil, fmt.Errorf("no cache found for entity %s", entityID)
} }
func (g *GithubToolsCache) Set(entity params.ForgeEntity, tools []commonParams.RunnerApplicationDownload) { func (g *GithubToolsCache) Set(entity params.ForgeEntity, tools []commonParams.RunnerApplicationDownload) {
@ -55,6 +64,7 @@ func (g *GithubToolsCache) Set(entity params.ForgeEntity, tools []commonParams.R
updatedAt: time.Now(), updatedAt: time.Now(),
entity: entity, entity: entity,
tools: tools, tools: tools,
err: nil,
} }
if entity.Credentials.ForgeType == params.GithubEndpointType { if entity.Credentials.ForgeType == params.GithubEndpointType {
@ -64,10 +74,30 @@ func (g *GithubToolsCache) Set(entity params.ForgeEntity, tools []commonParams.R
g.entities[entity.ID] = forgeTools g.entities[entity.ID] = forgeTools
} }
func (g *GithubToolsCache) SetToolsError(entity params.ForgeEntity, err error) {
g.mux.Lock()
defer g.mux.Unlock()
// If the entity is not in the cache, add it with the error.
cache, ok := g.entities[entity.ID]
if !ok {
g.entities[entity.ID] = GithubEntityTools{
updatedAt: time.Now(),
entity: entity,
err: err,
}
return
}
// Update the error for the existing entity.
cache.err = err
g.entities[entity.ID] = cache
}
func SetGithubToolsCache(entity params.ForgeEntity, tools []commonParams.RunnerApplicationDownload) { func SetGithubToolsCache(entity params.ForgeEntity, tools []commonParams.RunnerApplicationDownload) {
githubToolsCache.Set(entity, tools) githubToolsCache.Set(entity, tools)
} }
func GetGithubToolsCache(entityID string) ([]commonParams.RunnerApplicationDownload, bool) { func GetGithubToolsCache(entityID string) ([]commonParams.RunnerApplicationDownload, error) {
return githubToolsCache.Get(entityID) return githubToolsCache.Get(entityID)
} }

View file

@ -1099,13 +1099,18 @@ func (g ForgeEntity) GetCreatedAt() time.Time {
} }
func (g ForgeEntity) ForgeURL() string { func (g ForgeEntity) ForgeURL() string {
switch g.EntityType { switch g.Credentials.ForgeType {
case ForgeEntityTypeRepository: case GiteaEndpointType:
return fmt.Sprintf("%s/%s/%s", g.Credentials.BaseURL, g.Owner, g.Name) return g.Credentials.Endpoint.APIBaseURL
case ForgeEntityTypeOrganization: default:
return fmt.Sprintf("%s/%s", g.Credentials.BaseURL, g.Owner) switch g.EntityType {
case ForgeEntityTypeEnterprise: case ForgeEntityTypeRepository:
return fmt.Sprintf("%s/enterprises/%s", g.Credentials.BaseURL, g.Owner) return fmt.Sprintf("%s/%s/%s", g.Credentials.BaseURL, g.Owner, g.Name)
case ForgeEntityTypeOrganization:
return fmt.Sprintf("%s/%s", g.Credentials.BaseURL, g.Owner)
case ForgeEntityTypeEnterprise:
return fmt.Sprintf("%s/enterprises/%s", g.Credentials.BaseURL, g.Owner)
}
} }
return "" return ""
} }

View file

@ -16,7 +16,7 @@ import (
"github.com/cloudbase/garm/params" "github.com/cloudbase/garm/params"
) )
var systemdUnitTemplate = `[Unit] var githubSystemdUnitTemplate = `[Unit]
Description=GitHub Actions Runner ({{.ServiceName}}) Description=GitHub Actions Runner ({{.ServiceName}})
After=network.target After=network.target
@ -32,11 +32,24 @@ TimeoutStopSec=5min
WantedBy=multi-user.target WantedBy=multi-user.target
` `
func validateInstanceState(ctx context.Context) (params.Instance, error) { var giteaSystemdUnitTemplate = `[Unit]
if !auth.InstanceHasJITConfig(ctx) { Description=Act Runner ({{.ServiceName}})
return params.Instance{}, fmt.Errorf("instance not configured for JIT: %w", runnerErrors.ErrNotFound) After=network.target
}
[Service]
ExecStart=/home/{{.RunAsUser}}/act-runner/act_runner daemon --once
User={{.RunAsUser}}
WorkingDirectory=/home/{{.RunAsUser}}/act-runner
KillMode=process
KillSignal=SIGTERM
TimeoutStopSec=5min
Restart=always
[Install]
WantedBy=multi-user.target
`
func validateInstanceState(ctx context.Context) (params.Instance, error) {
status := auth.InstanceRunnerStatus(ctx) status := auth.InstanceRunnerStatus(ctx)
if status != params.RunnerPending && status != params.RunnerInstalling { if status != params.RunnerPending && status != params.RunnerInstalling {
return params.Instance{}, runnerErrors.ErrUnauthorized return params.Instance{}, runnerErrors.ErrUnauthorized
@ -49,6 +62,56 @@ func validateInstanceState(ctx context.Context) (params.Instance, error) {
return instance, nil return instance, nil
} }
func (r *Runner) getForgeEntityFromInstance(ctx context.Context, instance params.Instance) (params.ForgeEntity, error) {
var entityGetter params.EntityGetter
var err error
switch {
case instance.PoolID != "":
entityGetter, err = r.store.GetPoolByID(r.ctx, instance.PoolID)
case instance.ScaleSetID != 0:
entityGetter, err = r.store.GetScaleSetByID(r.ctx, instance.ScaleSetID)
default:
return params.ForgeEntity{}, errors.New("instance not associated with a pool or scale set")
}
if err != nil {
slog.With(slog.Any("error", err)).ErrorContext(
ctx, "failed to get entity getter",
"instance", instance.Name)
return params.ForgeEntity{}, errors.Wrap(err, "fetching entity getter")
}
poolEntity, err := entityGetter.GetEntity()
if err != nil {
slog.With(slog.Any("error", err)).ErrorContext(
ctx, "failed to get entity",
"instance", instance.Name)
return params.ForgeEntity{}, errors.Wrap(err, "fetching entity")
}
entity, err := r.store.GetForgeEntity(r.ctx, poolEntity.EntityType, poolEntity.ID)
if err != nil {
slog.With(slog.Any("error", err)).ErrorContext(
ctx, "failed to get entity",
"instance", instance.Name)
return params.ForgeEntity{}, errors.Wrap(err, "fetching entity")
}
return entity, nil
}
func (r *Runner) getServiceNameForEntity(entity params.ForgeEntity) (string, error) {
switch entity.EntityType {
case params.ForgeEntityTypeEnterprise:
return fmt.Sprintf("actions.runner.%s.%s", entity.Owner, entity.Name), nil
case params.ForgeEntityTypeOrganization:
return fmt.Sprintf("actions.runner.%s.%s", entity.Owner, entity.Name), nil
case params.ForgeEntityTypeRepository:
return fmt.Sprintf("actions.runner.%s-%s.%s", entity.Owner, entity.Name, entity.Name), nil
default:
return "", errors.New("unknown entity type")
}
}
func (r *Runner) GetRunnerServiceName(ctx context.Context) (string, error) { func (r *Runner) GetRunnerServiceName(ctx context.Context) (string, error) {
instance, err := validateInstanceState(ctx) instance, err := validateInstanceState(ctx)
if err != nil { if err != nil {
@ -56,64 +119,51 @@ func (r *Runner) GetRunnerServiceName(ctx context.Context) (string, error) {
ctx, "failed to get instance params") ctx, "failed to get instance params")
return "", runnerErrors.ErrUnauthorized return "", runnerErrors.ErrUnauthorized
} }
var entity params.ForgeEntity entity, err := r.getForgeEntityFromInstance(ctx, instance)
if err != nil {
switch { slog.ErrorContext(r.ctx, "failed to get entity", "error", err)
case instance.PoolID != "": return "", errors.Wrap(err, "fetching entity")
pool, err := r.store.GetPoolByID(r.ctx, instance.PoolID)
if err != nil {
slog.With(slog.Any("error", err)).ErrorContext(
ctx, "failed to get pool",
"pool_id", instance.PoolID)
return "", errors.Wrap(err, "fetching pool")
}
entity, err = pool.GetEntity()
if err != nil {
slog.With(slog.Any("error", err)).ErrorContext(
ctx, "failed to get pool entity",
"pool_id", instance.PoolID)
return "", errors.Wrap(err, "fetching pool entity")
}
case instance.ScaleSetID != 0:
scaleSet, err := r.store.GetScaleSetByID(r.ctx, instance.ScaleSetID)
if err != nil {
slog.With(slog.Any("error", err)).ErrorContext(
ctx, "failed to get scale set",
"scale_set_id", instance.ScaleSetID)
return "", errors.Wrap(err, "fetching scale set")
}
entity, err = scaleSet.GetEntity()
if err != nil {
slog.With(slog.Any("error", err)).ErrorContext(
ctx, "failed to get scale set entity",
"scale_set_id", instance.ScaleSetID)
return "", errors.Wrap(err, "fetching scale set entity")
}
default:
return "", errors.New("instance not associated with a pool or scale set")
} }
tpl := "actions.runner.%s.%s" serviceName, err := r.getServiceNameForEntity(entity)
var serviceName string if err != nil {
switch entity.EntityType { slog.ErrorContext(r.ctx, "failed to get service name", "error", err)
case params.ForgeEntityTypeEnterprise: return "", errors.Wrap(err, "fetching service name")
serviceName = fmt.Sprintf(tpl, entity.Owner, instance.Name)
case params.ForgeEntityTypeOrganization:
serviceName = fmt.Sprintf(tpl, entity.Owner, instance.Name)
case params.ForgeEntityTypeRepository:
serviceName = fmt.Sprintf(tpl, fmt.Sprintf("%s-%s", entity.Owner, entity.Name), instance.Name)
} }
return serviceName, nil return serviceName, nil
} }
func (r *Runner) GenerateSystemdUnitFile(ctx context.Context, runAsUser string) ([]byte, error) { func (r *Runner) GenerateSystemdUnitFile(ctx context.Context, runAsUser string) ([]byte, error) {
serviceName, err := r.GetRunnerServiceName(ctx) instance, err := validateInstanceState(ctx)
if err != nil { if err != nil {
return nil, errors.Wrap(err, "fetching runner service name") slog.With(slog.Any("error", err)).ErrorContext(
ctx, "failed to get instance params")
return nil, runnerErrors.ErrUnauthorized
}
entity, err := r.getForgeEntityFromInstance(ctx, instance)
if err != nil {
slog.ErrorContext(r.ctx, "failed to get entity", "error", err)
return nil, errors.Wrap(err, "fetching entity")
} }
unitTemplate, err := template.New("").Parse(systemdUnitTemplate) serviceName, err := r.getServiceNameForEntity(entity)
if err != nil { if err != nil {
slog.ErrorContext(r.ctx, "failed to get service name", "error", err)
return nil, errors.Wrap(err, "fetching service name")
}
var unitTemplate *template.Template
switch entity.Credentials.ForgeType {
case params.GithubEndpointType:
unitTemplate, err = template.New("").Parse(githubSystemdUnitTemplate)
case params.GiteaEndpointType:
unitTemplate, err = template.New("").Parse(giteaSystemdUnitTemplate)
default:
slog.ErrorContext(r.ctx, "unknown forge type", "forge_type", entity.Credentials.ForgeType)
return nil, errors.New("unknown forge type")
}
if err != nil {
slog.ErrorContext(r.ctx, "failed to parse template", "error", err)
return nil, errors.Wrap(err, "parsing template") return nil, errors.Wrap(err, "parsing template")
} }
@ -131,12 +181,17 @@ func (r *Runner) GenerateSystemdUnitFile(ctx context.Context, runAsUser string)
var unitFile bytes.Buffer var unitFile bytes.Buffer
if err := unitTemplate.Execute(&unitFile, data); err != nil { if err := unitTemplate.Execute(&unitFile, data); err != nil {
slog.ErrorContext(r.ctx, "failed to execute template", "error", err)
return nil, errors.Wrap(err, "executing template") return nil, errors.Wrap(err, "executing template")
} }
return unitFile.Bytes(), nil return unitFile.Bytes(), nil
} }
func (r *Runner) GetJITConfigFile(ctx context.Context, file string) ([]byte, error) { func (r *Runner) GetJITConfigFile(ctx context.Context, file string) ([]byte, error) {
if !auth.InstanceHasJITConfig(ctx) {
return nil, fmt.Errorf("instance not configured for JIT: %w", runnerErrors.ErrNotFound)
}
instance, err := validateInstanceState(ctx) instance, err := validateInstanceState(ctx)
if err != nil { if err != nil {
slog.With(slog.Any("error", err)).ErrorContext( slog.With(slog.Any("error", err)).ErrorContext(

View file

@ -47,15 +47,15 @@ import (
) )
var ( var (
poolIDLabelprefix = "runner-pool-id:" poolIDLabelprefix = "runner-pool-id"
controllerLabelPrefix = "runner-controller-id:" controllerLabelPrefix = "runner-controller-id"
// We tag runners that have been spawned as a result of a queued job with the job ID // We tag runners that have been spawned as a result of a queued job with the job ID
// that spawned them. There is no way to guarantee that the runner spawned in response to a particular // that spawned them. There is no way to guarantee that the runner spawned in response to a particular
// job, will be picked up by that job. We mark them so as in the very likely event that the runner // job, will be picked up by that job. We mark them so as in the very likely event that the runner
// has picked up a different job, we can clear the lock on the job that spaned it. // has picked up a different job, we can clear the lock on the job that spaned it.
// The job it picked up would already be transitioned to in_progress so it will be ignored by the // The job it picked up would already be transitioned to in_progress so it will be ignored by the
// consume loop. // consume loop.
jobLabelPrefix = "in_response_to_job:" jobLabelPrefix = "in_response_to_job"
) )
const ( const (
@ -296,7 +296,8 @@ func (r *basePoolManager) HandleWorkflowJob(job params.WorkflowJob) error {
func jobIDFromLabels(labels []string) int64 { func jobIDFromLabels(labels []string) int64 {
for _, lbl := range labels { for _, lbl := range labels {
if strings.HasPrefix(lbl, jobLabelPrefix) { if strings.HasPrefix(lbl, jobLabelPrefix) {
jobID, err := strconv.ParseInt(lbl[len(jobLabelPrefix):], 10, 64) trimLength := min(len(jobLabelPrefix)+1, len(lbl))
jobID, err := strconv.ParseInt(lbl[trimLength:], 10, 64)
if err != nil { if err != nil {
return 0 return 0
} }
@ -361,21 +362,21 @@ func (r *basePoolManager) startLoopForFunction(f func() error, interval time.Dur
} }
func (r *basePoolManager) updateTools() error { func (r *basePoolManager) updateTools() error {
// Update tools cache. tools, err := cache.GetGithubToolsCache(r.entity.ID)
tools, err := r.FetchTools()
if err != nil { if err != nil {
slog.With(slog.Any("error", err)).ErrorContext( slog.With(slog.Any("error", err)).ErrorContext(
r.ctx, "failed to update tools for entity", "entity", r.entity.String()) r.ctx, "failed to update tools for entity", "entity", r.entity.String())
r.SetPoolRunningState(false, err.Error()) r.SetPoolRunningState(false, err.Error())
return fmt.Errorf("failed to update tools for entity %s: %w", r.entity.String(), err) return fmt.Errorf("failed to update tools for entity %s: %w", r.entity.String(), err)
} }
r.mux.Lock() r.mux.Lock()
r.tools = tools r.tools = tools
r.mux.Unlock() r.mux.Unlock()
slog.DebugContext(r.ctx, "successfully updated tools") slog.DebugContext(r.ctx, "successfully updated tools")
r.SetPoolRunningState(true, "") r.SetPoolRunningState(true, "")
return err return nil
} }
// cleanupOrphanedProviderRunners compares runners in github with local runners and removes // cleanupOrphanedProviderRunners compares runners in github with local runners and removes
@ -995,11 +996,11 @@ func (r *basePoolManager) paramsWorkflowJobToParamsJob(job params.WorkflowJob) (
} }
func (r *basePoolManager) poolLabel(poolID string) string { func (r *basePoolManager) poolLabel(poolID string) string {
return fmt.Sprintf("%s%s", poolIDLabelprefix, poolID) return fmt.Sprintf("%s=%s", poolIDLabelprefix, poolID)
} }
func (r *basePoolManager) controllerLabel() string { func (r *basePoolManager) controllerLabel() string {
return fmt.Sprintf("%s%s", controllerLabelPrefix, r.controllerInfo.ControllerID.String()) return fmt.Sprintf("%s=%s", controllerLabelPrefix, r.controllerInfo.ControllerID.String())
} }
func (r *basePoolManager) updateArgsFromProviderInstance(providerInstance commonParams.ProviderInstance) params.UpdateInstanceParams { func (r *basePoolManager) updateArgsFromProviderInstance(providerInstance commonParams.ProviderInstance) params.UpdateInstanceParams {
@ -1613,6 +1614,16 @@ func (r *basePoolManager) Start() error {
initialToolUpdate := make(chan struct{}, 1) initialToolUpdate := make(chan struct{}, 1)
go func() { go func() {
slog.Info("running initial tool update") slog.Info("running initial tool update")
for {
slog.DebugContext(r.ctx, "waiting for tools to be available")
hasTools, stopped := r.waitForToolsOrCancel()
if stopped {
return
}
if hasTools {
break
}
}
if err := r.updateTools(); err != nil { if err := r.updateTools(); err != nil {
slog.With(slog.Any("error", err)).Error("failed to update tools") slog.With(slog.Any("error", err)).Error("failed to update tools")
} }
@ -1804,7 +1815,7 @@ func (r *basePoolManager) consumeQueuedJobs() error {
} }
jobLabels := []string{ jobLabels := []string{
fmt.Sprintf("%s%d", jobLabelPrefix, job.ID), fmt.Sprintf("%s=%d", jobLabelPrefix, job.ID),
} }
for i := 0; i < poolRR.Len(); i++ { for i := 0; i < poolRR.Len(); i++ {
pool, err := poolRR.Next() pool, err := poolRR.Next()

View file

@ -5,11 +5,13 @@ import (
"strings" "strings"
"sync" "sync"
"sync/atomic" "sync/atomic"
"time"
"github.com/google/go-github/v71/github" "github.com/google/go-github/v71/github"
runnerErrors "github.com/cloudbase/garm-provider-common/errors" runnerErrors "github.com/cloudbase/garm-provider-common/errors"
commonParams "github.com/cloudbase/garm-provider-common/params" commonParams "github.com/cloudbase/garm-provider-common/params"
"github.com/cloudbase/garm/cache"
dbCommon "github.com/cloudbase/garm/database/common" dbCommon "github.com/cloudbase/garm/database/common"
"github.com/cloudbase/garm/database/watcher" "github.com/cloudbase/garm/database/watcher"
"github.com/cloudbase/garm/params" "github.com/cloudbase/garm/params"
@ -91,7 +93,8 @@ func instanceInList(instanceName string, instances []commonParams.ProviderInstan
func controllerIDFromLabels(labels []string) string { func controllerIDFromLabels(labels []string) string {
for _, lbl := range labels { for _, lbl := range labels {
if strings.HasPrefix(lbl, controllerLabelPrefix) { if strings.HasPrefix(lbl, controllerLabelPrefix) {
return lbl[len(controllerLabelPrefix):] trimLength := min(len(controllerLabelPrefix)+1, len(lbl))
return lbl[trimLength:]
} }
} }
return "" return ""
@ -134,3 +137,19 @@ func composeWatcherFilters(entity params.ForgeEntity) dbCommon.PayloadFilterFunc
watcher.WithForgeCredentialsFilter(entity.Credentials), watcher.WithForgeCredentialsFilter(entity.Credentials),
) )
} }
func (r *basePoolManager) waitForToolsOrCancel() (hasTools, stopped bool) {
ticker := time.NewTicker(1 * time.Second)
defer ticker.Stop()
select {
case <-ticker.C:
if _, err := cache.GetGithubToolsCache(r.entity.ID); err != nil {
return false, false
}
return true, false
case <-r.quit:
return false, true
case <-r.ctx.Done():
return false, true
}
}

View file

@ -602,7 +602,7 @@ func (r *Runner) validateHookBody(signature, secret string, body []byte) error {
return nil return nil
} }
func (r *Runner) findEndpointForJob(job params.WorkflowJob) (params.ForgeEndpoint, error) { func (r *Runner) findEndpointForJob(job params.WorkflowJob, forgeType params.EndpointType) (params.ForgeEndpoint, error) {
uri, err := url.ParseRequestURI(job.WorkflowJob.HTMLURL) uri, err := url.ParseRequestURI(job.WorkflowJob.HTMLURL)
if err != nil { if err != nil {
return params.ForgeEndpoint{}, errors.Wrap(err, "parsing job URL") return params.ForgeEndpoint{}, errors.Wrap(err, "parsing job URL")
@ -614,12 +614,23 @@ func (r *Runner) findEndpointForJob(job params.WorkflowJob) (params.ForgeEndpoin
// a GHES involved, those users will have just one extra endpoint or 2 (if they also have a // a GHES involved, those users will have just one extra endpoint or 2 (if they also have a
// test env). But there should be a relatively small number, regardless. So we don't really care // test env). But there should be a relatively small number, regardless. So we don't really care
// that much about the performance of this function. // that much about the performance of this function.
endpoints, err := r.store.ListGithubEndpoints(r.ctx) var endpoints []params.ForgeEndpoint
switch forgeType {
case params.GithubEndpointType:
endpoints, err = r.store.ListGithubEndpoints(r.ctx)
case params.GiteaEndpointType:
endpoints, err = r.store.ListGiteaEndpoints(r.ctx)
default:
return params.ForgeEndpoint{}, runnerErrors.NewBadRequestError("unknown forge type %s", forgeType)
}
if err != nil { if err != nil {
return params.ForgeEndpoint{}, errors.Wrap(err, "fetching github endpoints") return params.ForgeEndpoint{}, errors.Wrap(err, "fetching github endpoints")
} }
for _, ep := range endpoints { for _, ep := range endpoints {
if ep.BaseURL == baseURI { slog.DebugContext(r.ctx, "checking endpoint", "base_uri", baseURI, "endpoint", ep.BaseURL)
epBaseURI := strings.TrimSuffix(ep.BaseURL, "/")
if epBaseURI == baseURI {
return ep, nil return ep, nil
} }
} }
@ -627,18 +638,21 @@ func (r *Runner) findEndpointForJob(job params.WorkflowJob) (params.ForgeEndpoin
return params.ForgeEndpoint{}, runnerErrors.NewNotFoundError("no endpoint found for job") return params.ForgeEndpoint{}, runnerErrors.NewNotFoundError("no endpoint found for job")
} }
func (r *Runner) DispatchWorkflowJob(hookTargetType, signature string, jobData []byte) error { func (r *Runner) DispatchWorkflowJob(hookTargetType, signature string, forgeType params.EndpointType, jobData []byte) error {
if len(jobData) == 0 { if len(jobData) == 0 {
slog.ErrorContext(r.ctx, "missing job data")
return runnerErrors.NewBadRequestError("missing job data") return runnerErrors.NewBadRequestError("missing job data")
} }
var job params.WorkflowJob var job params.WorkflowJob
if err := json.Unmarshal(jobData, &job); err != nil { if err := json.Unmarshal(jobData, &job); err != nil {
slog.ErrorContext(r.ctx, "failed to unmarshal job data", "error", err)
return errors.Wrapf(runnerErrors.ErrBadRequest, "invalid job data: %s", err) return errors.Wrapf(runnerErrors.ErrBadRequest, "invalid job data: %s", err)
} }
endpoint, err := r.findEndpointForJob(job) endpoint, err := r.findEndpointForJob(job, forgeType)
if err != nil { if err != nil {
slog.ErrorContext(r.ctx, "failed to find endpoint for job", "error", err)
return errors.Wrap(err, "finding endpoint for job") return errors.Wrap(err, "finding endpoint for job")
} }
@ -867,15 +881,17 @@ func (r *Runner) DeleteRunner(ctx context.Context, instanceName string, forceDel
} }
if err != nil { if err != nil {
if errors.Is(err, runnerErrors.ErrUnauthorized) && instance.PoolID != "" { if !errors.Is(err, runnerErrors.ErrNotFound) {
poolMgr, err := r.getPoolManagerFromInstance(ctx, instance) if errors.Is(err, runnerErrors.ErrUnauthorized) && instance.PoolID != "" {
if err != nil { poolMgr, err := r.getPoolManagerFromInstance(ctx, instance)
return errors.Wrap(err, "fetching pool manager for instance") if err != nil {
return errors.Wrap(err, "fetching pool manager for instance")
}
poolMgr.SetPoolRunningState(false, fmt.Sprintf("failed to remove runner: %q", err))
}
if !bypassGithubUnauthorized {
return errors.Wrap(err, "removing runner from github")
} }
poolMgr.SetPoolRunningState(false, fmt.Sprintf("failed to remove runner: %q", err))
}
if !bypassGithubUnauthorized {
return errors.Wrap(err, "removing runner from github")
} }
} }
} }

View file

@ -229,6 +229,7 @@ func (g *githubClient) ListEntityRunnerApplicationDownloads(ctx context.Context)
} }
func parseError(response *github.Response, err error) error { func parseError(response *github.Response, err error) error {
slog.Debug("parsing error", "status_code", response.StatusCode, "response", response, "error", err)
switch response.StatusCode { switch response.StatusCode {
case http.StatusNotFound: case http.StatusNotFound:
return runnerErrors.ErrNotFound return runnerErrors.ErrNotFound
@ -251,6 +252,10 @@ func parseError(response *github.Response, err error) error {
case http.StatusUnprocessableEntity: case http.StatusUnprocessableEntity:
return runnerErrors.ErrBadRequest return runnerErrors.ErrBadRequest
default: default:
// ugly hack. Gitea returns 500 if we try to remove a runner that does not exist.
if strings.Contains(err.Error(), "does not exist") {
return runnerErrors.ErrNotFound
}
return err return err
} }
} }

View file

@ -20,6 +20,18 @@ const (
GiteaRunnerMinimumVersion = "v0.2.12" GiteaRunnerMinimumVersion = "v0.2.12"
) )
var (
githubArchMapping map[string]string = map[string]string{
"x86_64": "x64",
"amd64": "x64",
"armv7l": "arm",
"aarch64": "arm64",
"x64": "x64",
"arm": "arm",
"arm64": "arm64",
}
)
var nightlyActRunner = GiteaEntityTool{ var nightlyActRunner = GiteaEntityTool{
TagName: "nightly", TagName: "nightly",
Name: "nightly", Name: "nightly",
@ -50,36 +62,39 @@ type GiteaToolsAssets struct {
DownloadURL string `json:"browser_download_url"` DownloadURL string `json:"browser_download_url"`
} }
func (g GiteaToolsAssets) GetOS() *string { func (g GiteaToolsAssets) GetOS() (*string, error) {
if g.Name == "" { if g.Name == "" {
return nil return nil, fmt.Errorf("gitea tools name is empty")
} }
parts := strings.SplitN(g.Name, "-", 4) parts := strings.SplitN(g.Name, "-", 4)
if len(parts) != 4 { if len(parts) != 4 {
return nil return nil, fmt.Errorf("could not parse asset name")
} }
os := parts[2] os := parts[2]
return &os return &os, nil
} }
func (g GiteaToolsAssets) GetArch() *string { func (g GiteaToolsAssets) GetArch() (*string, error) {
if g.Name == "" { if g.Name == "" {
return nil return nil, fmt.Errorf("gitea tools name is empty")
} }
parts := strings.SplitN(g.Name, "-", 4) parts := strings.SplitN(g.Name, "-", 4)
if len(parts) != 4 { if len(parts) != 4 {
return nil return nil, fmt.Errorf("could not parse asset name")
} }
archParts := strings.SplitN(parts[3], ".", 2) archParts := strings.SplitN(parts[3], ".", 2)
if len(archParts) == 0 { if len(archParts) == 0 {
return nil return nil, fmt.Errorf("unexpected asset name format")
} }
arch := archParts[0] arch := githubArchMapping[archParts[0]]
return &arch if arch == "" {
return nil, fmt.Errorf("could not find arch for %s", archParts[0])
}
return &arch, nil
} }
type GiteaEntityTool struct { type GiteaEntityTool struct {
@ -140,9 +155,17 @@ func getTools() ([]commonParams.RunnerApplicationDownload, error) {
ret := []commonParams.RunnerApplicationDownload{} ret := []commonParams.RunnerApplicationDownload{}
for _, asset := range latest.Assets { for _, asset := range latest.Assets {
arch, err := asset.GetArch()
if err != nil {
return nil, fmt.Errorf("getting arch: %w", err)
}
os, err := asset.GetOS()
if err != nil {
return nil, fmt.Errorf("getting os: %w", err)
}
ret = append(ret, commonParams.RunnerApplicationDownload{ ret = append(ret, commonParams.RunnerApplicationDownload{
OS: asset.GetOS(), OS: os,
Architecture: asset.GetArch(), Architecture: arch,
DownloadURL: &asset.DownloadURL, DownloadURL: &asset.DownloadURL,
Filename: &asset.Name, Filename: &asset.Name,
}) })

View file

@ -49,7 +49,7 @@ func (t *toolsUpdater) Start() error {
t.running = true t.running = true
t.quit = make(chan struct{}) t.quit = make(chan struct{})
slog.DebugContext(t.ctx, "starting tools updater", "entity", t.entity.String(), "forge_type", t.entity.Credentials) slog.DebugContext(t.ctx, "starting tools updater", "entity", t.entity.String(), "forge_type", t.entity.Credentials.ForgeType)
switch t.entity.Credentials.ForgeType { switch t.entity.Credentials.ForgeType {
case params.GithubEndpointType: case params.GithubEndpointType:

View file

@ -148,9 +148,9 @@ func (i *instanceManager) handleCreateInstanceInProvider(instance params.Instanc
if err != nil { if err != nil {
return fmt.Errorf("creating instance token: %w", err) return fmt.Errorf("creating instance token: %w", err)
} }
tools, ok := cache.GetGithubToolsCache(entity.ID) tools, err := cache.GetGithubToolsCache(entity.ID)
if !ok { if err != nil {
return fmt.Errorf("tools not found in cache for entity %s", entity.String()) return fmt.Errorf("tools not found in cache for entity %s: %w", entity.String(), err)
} }
bootstrapArgs := commonParams.BootstrapInstance{ bootstrapArgs := commonParams.BootstrapInstance{

View file

@ -776,8 +776,11 @@ func (w *Worker) waitForToolsOrCancel() (hasTools, stopped bool) {
if err != nil { if err != nil {
slog.ErrorContext(w.ctx, "error getting entity", "error", err) slog.ErrorContext(w.ctx, "error getting entity", "error", err)
} }
_, ok := cache.GetGithubToolsCache(entity.ID) if _, err := cache.GetGithubToolsCache(entity.ID); err != nil {
return ok, false slog.DebugContext(w.ctx, "tools not found in cache; waiting for tools")
return false, false
}
return true, false
case <-w.quit: case <-w.quit:
return false, true return false, true
case <-w.ctx.Done(): case <-w.ctx.Done():