fix: race during stop of active backends (#5106)

* chore: drop double call to stop all backends, refactors

Signed-off-by: Ettore Di Giacinto <mudler@localai.io>

* fix: do lock when cycling to models to delete

Signed-off-by: Ettore Di Giacinto <mudler@localai.io>

---------

Signed-off-by: Ettore Di Giacinto <mudler@localai.io>
This commit is contained in:
Ettore Di Giacinto 2025-04-01 00:01:10 +02:00 committed by GitHub
parent 2f9203cd2a
commit 05f7004487
No known key found for this signature in database
GPG key ID: B5690EEEBB952194
4 changed files with 45 additions and 37 deletions

View file

@ -142,26 +142,6 @@ func (ml *ModelLoader) LoadModel(modelID, modelName string, loader func(string,
func (ml *ModelLoader) ShutdownModel(modelName string) error {
ml.mu.Lock()
defer ml.mu.Unlock()
model, ok := ml.models[modelName]
if !ok {
return fmt.Errorf("model %s not found", modelName)
}
retries := 1
for model.GRPC(false, ml.wd).IsBusy() {
log.Debug().Msgf("%s busy. Waiting.", modelName)
dur := time.Duration(retries*2) * time.Second
if dur > retryTimeout {
dur = retryTimeout
}
time.Sleep(dur)
retries++
if retries > 10 && os.Getenv("LOCALAI_FORCE_BACKEND_SHUTDOWN") == "true" {
log.Warn().Msgf("Model %s is still busy after %d retries. Forcing shutdown.", modelName, retries)
break
}
}
return ml.deleteProcess(modelName)
}