runner now done

This commit is contained in:
2024-05-09 00:46:42 +01:00
parent 9eba93cd3c
commit a0aed71b3c
5 changed files with 436 additions and 102 deletions

View File

@@ -10,6 +10,7 @@ import (
. "git.andr3h3nriqu3s.com/andr3/fyp/logic/models/train"
. "git.andr3h3nriqu3s.com/andr3/fyp/logic/tasks/utils"
. "git.andr3h3nriqu3s.com/andr3/fyp/logic/utils"
"github.com/charmbracelet/log"
)
func verifyRunner(c *Context, dat *JustId) (runner *Runner, e *Error) {
@@ -34,6 +35,12 @@ type VerifyTask struct {
TaskId string `json:"taskId" validate:"required"`
}
type RunnerTrainDef struct {
Id string `json:"id" validate:"required"`
TaskId string `json:"taskId" validate:"required"`
DefId string `json:"defId" validate:"required"`
}
func verifyTask(x *Handle, c *Context, dat *VerifyTask) (task *Task, error *Error) {
mutex := x.DataMap["runners_mutex"].(*sync.Mutex)
mutex.Lock()
@@ -53,6 +60,18 @@ func verifyTask(x *Handle, c *Context, dat *VerifyTask) (task *Task, error *Erro
return runner_data["task"].(*Task), nil
}
func clearRunnerTask(x *Handle, runner_id string) {
mutex := x.DataMap["runners_mutex"].(*sync.Mutex)
mutex.Lock()
defer mutex.Unlock()
var runners map[string]interface{} = x.DataMap["runners"].(map[string]interface{})
var runner_data map[string]interface{} = runners[runner_id].(map[string]interface{})
runner_data["task"] = nil
runners[runner_id] = runner_data
x.DataMap["runners"] = runners
}
func handleRemoteRunner(x *Handle) {
type RegisterRunner struct {
@@ -202,6 +221,8 @@ func handleRemoteRunner(x *Handle) {
switch task.TaskType {
case int(TASK_TYPE_TRAINING):
CleanUpFailed(c, task)
case int(TASK_TYPE_RETRAINING):
CleanUpFailedRetrain(c, task)
case int(TASK_TYPE_CLASSIFICATION):
// DO nothing
default:
@@ -237,6 +258,8 @@ func handleRemoteRunner(x *Handle) {
switch task.TaskType {
case int(TASK_TYPE_TRAINING):
status = DEFINITION_STATUS_INIT
case int(TASK_TYPE_RETRAINING):
fallthrough
case int(TASK_TYPE_CLASSIFICATION):
status = DEFINITION_STATUS_READY
default:
@@ -268,27 +291,35 @@ func handleRemoteRunner(x *Handle) {
return error
}
switch task.TaskType {
case int(TASK_TYPE_TRAINING):
//DO NOTHING
case int(TASK_TYPE_CLASSIFICATION):
//DO NOTHING
default:
c.Logger.Error("Task not is not the right type to get the definitions", "task type", task.TaskType)
return c.JsonBadRequest("Task is not the right type go get the definitions")
}
model, err := GetBaseModel(c, *task.ModelId)
if err != nil {
return c.E500M("Failed to get model information", err)
}
classes, err := model.GetClasses(c, "and status=$2 order by mc.class_order asc", CLASS_STATUS_TO_TRAIN)
if err != nil {
return c.E500M("Failed to get the model classes", err)
switch task.TaskType {
case int(TASK_TYPE_TRAINING):
classes, err := model.GetClasses(c, "and status in ($2, $3) order by mc.class_order asc", CLASS_STATUS_TO_TRAIN, CLASS_STATUS_TRAINING)
if err != nil {
return c.E500M("Failed to get the model classes", err)
}
return c.SendJSON(classes)
case int(TASK_TYPE_RETRAINING):
classes, err := model.GetClasses(c, "and status=$2 order by mc.class_order asc", CLASS_STATUS_TRAINING)
if err != nil {
return c.E500M("Failed to get the model classes", err)
}
return c.SendJSON(classes)
case int(TASK_TYPE_CLASSIFICATION):
classes, err := model.GetClasses(c, "and status=$2 order by mc.class_order asc", CLASS_STATUS_TRAINED)
if err != nil {
return c.E500M("Failed to get the model classes", err)
}
return c.SendJSON(classes)
default:
c.Logger.Error("Task not is not the right type to get the definitions", "task type", task.TaskType)
return c.JsonBadRequest("Task is not the right type go get the definitions")
}
return c.SendJSON(classes)
})
type RunnerTrainDefStatus struct {
@@ -326,12 +357,13 @@ func handleRemoteRunner(x *Handle) {
return c.SendJSON("Ok")
})
type RunnerTrainDefLayers struct {
Id string `json:"id" validate:"required"`
TaskId string `json:"taskId" validate:"required"`
DefId string `json:"defId" validate:"required"`
type RunnerTrainDefHeadStatus struct {
Id string `json:"id" validate:"required"`
TaskId string `json:"taskId" validate:"required"`
DefId string `json:"defId" validate:"required"`
Status ModelHeadStatus `json:"status" validate:"required"`
}
PostAuthJson(x, "/tasks/runner/train/def/layers", User_Normal, func(c *Context, dat *RunnerTrainDefLayers) *Error {
PostAuthJson(x, "/tasks/runner/train/def/head/status", User_Normal, func(c *Context, dat *RunnerTrainDefHeadStatus) *Error {
_, error := verifyRunner(c, &JustId{Id: dat.Id})
if error != nil {
return error
@@ -352,6 +384,69 @@ func handleRemoteRunner(x *Handle) {
return c.E500M("Failed to get definition information", err)
}
_, err = c.Exec("update exp_model_head set status=$1 where def_id=$2;", dat.Status, def.Id)
if err != nil {
log.Error("Failed to train definition!")
return c.E500M("Failed to train definition", err)
}
return c.SendJSON("Ok")
})
type RunnerRetrainDefHeadStatus struct {
Id string `json:"id" validate:"required"`
TaskId string `json:"taskId" validate:"required"`
HeadId string `json:"defId" validate:"required"`
Status ModelHeadStatus `json:"status" validate:"required"`
}
PostAuthJson(x, "/tasks/runner/retrain/def/head/status", User_Normal, func(c *Context, dat *RunnerRetrainDefHeadStatus) *Error {
_, error := verifyRunner(c, &JustId{Id: dat.Id})
if error != nil {
return error
}
task, error := verifyTask(x, c, &VerifyTask{Id: dat.Id, TaskId: dat.TaskId})
if error != nil {
return error
}
if task.TaskType != int(TASK_TYPE_RETRAINING) {
c.Logger.Error("Task not is not the right type to get the definitions", "task type", task.TaskType)
return c.JsonBadRequest("Task is not the right type go get the definitions")
}
if err := UpdateStatus(c.GetDb(), "exp_model_head", dat.HeadId, MODEL_DEFINITION_STATUS_TRAINING); err != nil {
return c.E500M("Failed to update head status", err)
}
return c.SendJSON("Ok")
})
PostAuthJson(x, "/tasks/runner/train/def/layers", User_Normal, func(c *Context, dat *RunnerTrainDef) *Error {
_, error := verifyRunner(c, &JustId{Id: dat.Id})
if error != nil {
return error
}
task, error := verifyTask(x, c, &VerifyTask{Id: dat.Id, TaskId: dat.TaskId})
if error != nil {
return error
}
switch task.TaskType {
case int(TASK_TYPE_TRAINING):
// Do nothing
case int(TASK_TYPE_RETRAINING):
// Do nothing
default:
c.Logger.Error("Task not is not the right type to get the layers", "task type", task.TaskType)
return c.JsonBadRequest("Task is not the right type go get the layers")
}
def, err := GetDefinition(c, dat.DefId)
if err != nil {
return c.E500M("Failed to get definition information", err)
}
layers, err := def.GetLayers(c, " order by layer_order asc")
if err != nil {
return c.E500M("Failed to get layers", err)
@@ -371,7 +466,12 @@ func handleRemoteRunner(x *Handle) {
return error
}
if task.TaskType != int(TASK_TYPE_TRAINING) {
switch task.TaskType {
case int(TASK_TYPE_TRAINING):
// DO nothing
case int(TASK_TYPE_RETRAINING):
// DO nothing
default:
c.Logger.Error("Task not is not the right type to get the definitions", "task type", task.TaskType)
return c.JsonBadRequest("Task is not the right type go get the definitions")
}
@@ -486,6 +586,8 @@ func handleRemoteRunner(x *Handle) {
switch task.TaskType {
case int(TASK_TYPE_TRAINING):
//DO NOTHING
case int(TASK_TYPE_RETRAINING):
//DO NOTHING
case int(TASK_TYPE_CLASSIFICATION):
//DO NOTHING
default:
@@ -501,6 +603,74 @@ func handleRemoteRunner(x *Handle) {
return c.SendJSON(model)
})
PostAuthJson(x, "/tasks/runner/heads", User_Normal, func(c *Context, dat *RunnerTrainDef) *Error {
_, error := verifyRunner(c, &JustId{Id: dat.Id})
if error != nil {
return error
}
task, error := verifyTask(x, c, &VerifyTask{Id: dat.Id, TaskId: dat.TaskId})
if error != nil {
return error
}
type ExpHead struct {
Id string `json:"id"`
Start int `db:"range_start" json:"start"`
End int `db:"range_end" json:"end"`
}
switch task.TaskType {
case int(TASK_TYPE_TRAINING):
fallthrough
case int(TASK_TYPE_RETRAINING):
// status = 2 (INIT) 3 (TRAINING)
heads, err := GetDbMultitple[ExpHead](c, "exp_model_head where def_id=$1 and status in (2,3)", dat.DefId)
if err != nil {
return c.E500M("Failed getting active heads", err)
}
return c.SendJSON(heads)
case int(TASK_TYPE_CLASSIFICATION):
heads, err := GetDbMultitple[ExpHead](c, "exp_model_head where def_id=$1", dat.DefId)
if err != nil {
return c.E500M("Failed getting active heads", err)
}
return c.SendJSON(heads)
default:
c.Logger.Error("Task not is not the right type to get the definitions", "task type", task.TaskType)
return c.JsonBadRequest("Task is not the right type go get the definitions")
}
})
PostAuthJson(x, "/tasks/runner/train_exp/class/status/train", User_Normal, func(c *Context, dat *VerifyTask) *Error {
_, error := verifyRunner(c, &JustId{Id: dat.Id})
if error != nil {
return error
}
task, error := verifyTask(x, c, dat)
if error != nil {
return error
}
if task.TaskType != int(TASK_TYPE_TRAINING) {
c.Logger.Error("Task not is not the right type to get the definitions", "task type", task.TaskType)
return c.JsonBadRequest("Task is not the right type go get the definitions")
}
model, err := GetBaseModel(c, *task.ModelId)
if err != nil {
return c.E500M("Failed to get model", err)
}
err = SetModelClassStatus(c, CLASS_STATUS_TRAINING, "model_id=$1 and status=$2;", model.Id, CLASS_STATUS_TO_TRAIN)
if err != nil {
return c.E500M("Failed update status", err)
}
return c.SendJSON("Ok")
})
PostAuthJson(x, "/tasks/runner/train/done", User_Normal, func(c *Context, dat *VerifyTask) *Error {
_, error := verifyRunner(c, &JustId{Id: dat.Id})
if error != nil {
@@ -568,6 +738,13 @@ func handleRemoteRunner(x *Handle) {
return c.E500M("Failed to delete unsed definitions", err)
}
// Set the class status to trained
err = SetModelClassStatus(c, CLASS_STATUS_TRAINED, "model_id=$1;", model.Id)
if err != nil {
c.Logger.Error("Failed to set class status")
return c.E500M("Failed to set class status", err)
}
if err = model.UpdateStatus(c, READY); err != nil {
model.UpdateStatus(c, FAILED_TRAINING)
task.UpdateStatusLog(c, TASK_FAILED_RUNNING, "Failed to delete unsed definitions")
@@ -576,16 +753,7 @@ func handleRemoteRunner(x *Handle) {
task.UpdateStatusLog(c, TASK_DONE, "Model finished training")
mutex := x.DataMap["runners_mutex"].(*sync.Mutex)
mutex.Lock()
defer mutex.Unlock()
var runners map[string]interface{} = x.DataMap["runners"].(map[string]interface{})
var runner_data map[string]interface{} = runners[dat.Id].(map[string]interface{})
runner_data["task"] = nil
runners[dat.Id] = runner_data
x.DataMap["runners"] = runners
clearRunnerTask(x, dat.Id)
return c.SendJSON("Ok")
})
@@ -635,4 +803,153 @@ func handleRemoteRunner(x *Handle) {
return c.SendJSON("Ok")
})
PostAuthJson(x, "/tasks/runner/train_exp/done", User_Normal, func(c *Context, dat *VerifyTask) *Error {
_, error := verifyRunner(c, &JustId{Id: dat.Id})
if error != nil {
return error
}
task, error := verifyTask(x, c, dat)
if error != nil {
return error
}
if task.TaskType != int(TASK_TYPE_TRAINING) {
c.Logger.Error("Task not is not the right type to get the definitions", "task type", task.TaskType)
return c.JsonBadRequest("Task is not the right type go get the definitions")
}
model, err := GetBaseModel(c, *task.ModelId)
if err != nil {
c.Logger.Error("Failed to get model", "err", err)
return c.E500M("Failed to get mode", err)
}
// TODO add check the to the model
var def Definition
err = GetDBOnce(c, &def, "model_definition as md where model_id=$1 and status=$2 order by accuracy desc limit 1;", task.ModelId, DEFINITION_STATUS_TRANIED)
if err == NotFoundError {
// TODO Make the Model status have a message
c.Logger.Error("All definitions failed to train!")
model.UpdateStatus(c, FAILED_TRAINING)
task.UpdateStatusLog(c, TASK_FAILED_RUNNING, "All definition failed to train!")
clearRunnerTask(x, dat.Id)
return c.SendJSON("Ok")
} else if err != nil {
model.UpdateStatus(c, FAILED_TRAINING)
task.UpdateStatusLog(c, TASK_FAILED_RUNNING, "Failed to get model definition")
return c.E500M("Failed to get model definition", err)
}
if err = def.UpdateStatus(c, DEFINITION_STATUS_READY); err != nil {
model.UpdateStatus(c, FAILED_TRAINING)
task.UpdateStatusLog(c, TASK_FAILED_RUNNING, "Failed to update model definition")
return c.E500M("Failed to update model definition", err)
}
to_delete, err := GetDbMultitple[JustId](c, "model_definition where status!=$1 and model_id=$2", MODEL_DEFINITION_STATUS_READY, model.Id)
if err != nil {
c.GetLogger().Error("Failed to select model_definition to delete")
return c.E500M("Failed to select model definition to delete", err)
}
for _, d := range to_delete {
os.RemoveAll(path.Join("savedData", model.Id, "defs", d.Id))
}
// TODO Check if returning also works here
if _, err = c.Exec("delete from model_definition where status!=$1 and model_id=$2;", MODEL_DEFINITION_STATUS_READY, model.Id); err != nil {
model.UpdateStatus(c, FAILED_TRAINING)
task.UpdateStatusLog(c, TASK_FAILED_RUNNING, "Failed to delete unsed definitions")
return c.E500M("Failed to delete unsed definitions", err)
}
if err = SplitModel(c, model); err != nil {
err = SetModelClassStatus(c, CLASS_STATUS_TO_TRAIN, "model_id=$1 and status=$2;", model.Id, CLASS_STATUS_TRAINING)
if err != nil {
c.Logger.Error("Failed to split the model! And Failed to set class status")
return c.E500M("Failed to split the model", err)
}
c.Logger.Error("Failed to split the model")
return c.E500M("Failed to split the model", err)
}
// Set the class status to trained
err = SetModelClassStatus(c, CLASS_STATUS_TRAINED, "model_id=$1 and status=$2;", model.Id, CLASS_STATUS_TRAINING)
if err != nil {
c.Logger.Error("Failed to set class status")
return c.E500M("Failed to set class status", err)
}
c.Logger.Warn("Removing base model for", "model", model.Id, "def", def.Id)
os.RemoveAll(path.Join("savedData", model.Id, "defs", def.Id, "model"))
os.RemoveAll(path.Join("savedData", model.Id, "defs", def.Id, "model.keras"))
if err = model.UpdateStatus(c, READY); err != nil {
model.UpdateStatus(c, FAILED_TRAINING)
task.UpdateStatusLog(c, TASK_FAILED_RUNNING, "Failed to delete unsed definitions")
return c.E500M("Failed to update status of model", err)
}
task.UpdateStatusLog(c, TASK_DONE, "Model finished training")
mutex := x.DataMap["runners_mutex"].(*sync.Mutex)
mutex.Lock()
defer mutex.Unlock()
var runners map[string]interface{} = x.DataMap["runners"].(map[string]interface{})
var runner_data map[string]interface{} = runners[dat.Id].(map[string]interface{})
runner_data["task"] = nil
runners[dat.Id] = runner_data
x.DataMap["runners"] = runners
return c.SendJSON("Ok")
})
PostAuthJson(x, "/tasks/runner/retrain/done", User_Normal, func(c *Context, dat *VerifyTask) *Error {
_, error := verifyRunner(c, &JustId{Id: dat.Id})
if error != nil {
return error
}
task, error := verifyTask(x, c, dat)
if error != nil {
return error
}
if task.TaskType != int(TASK_TYPE_RETRAINING) {
c.Logger.Error("Task not is not the right type to get the definitions", "task type", task.TaskType)
return c.JsonBadRequest("Task is not the right type go get the definitions")
}
model, err := GetBaseModel(c, *task.ModelId)
if err != nil {
c.Logger.Error("Failed to get model", "err", err)
return c.E500M("Failed to get mode", err)
}
err = SetModelClassStatus(c, CLASS_STATUS_TRAINED, "model_id=$1 and status=$2;", model.Id, CLASS_STATUS_TRAINING)
if err != nil {
return c.E500M("Failed to set class status", err)
}
_, err = c.Exec("update exp_model_head set status=$1 where status=$2 and model_id=$3", MODEL_HEAD_STATUS_READY, MODEL_HEAD_STATUS_TRAINING, model.Id)
if err != nil {
return c.E500M("Failed to set head status", err)
}
err = model.UpdateStatus(c, READY)
if err != nil {
return c.E500M("Failed to set class status", err)
}
task.UpdateStatusLog(c, TASK_DONE, "Model finished training")
clearRunnerTask(x, dat.Id)
return c.SendJSON("Ok")
})
}