runner-go (#102)
Reviewed-on: andr3/fyp#102 Co-authored-by: Andre Henriques <andr3h3nriqu3s@gmail.com> Co-committed-by: Andre Henriques <andr3h3nriqu3s@gmail.com>
This commit was merged in pull request #102.
This commit is contained in:
@@ -6,3 +6,19 @@ const (
|
||||
DATA_POINT_MODE_TRAINING DATA_POINT_MODE = 1
|
||||
DATA_POINT_MODE_TESTING = 2
|
||||
)
|
||||
|
||||
type ModelClassStatus int
|
||||
|
||||
const (
|
||||
CLASS_STATUS_TO_TRAIN ModelClassStatus = iota + 1
|
||||
CLASS_STATUS_TRAINING
|
||||
CLASS_STATUS_TRAINED
|
||||
)
|
||||
|
||||
type ModelClass struct {
|
||||
Id string `db:"mc.id" json:"id"`
|
||||
ModelId string `db:"mc.model_id" json:"model_id"`
|
||||
Name string `db:"mc.name" json:"name"`
|
||||
ClassOrder int `db:"mc.class_order" json:"class_order"`
|
||||
Status int `db:"mc.status" json:"status"`
|
||||
}
|
||||
|
||||
95
logic/db_types/definitions.go
Normal file
95
logic/db_types/definitions.go
Normal file
@@ -0,0 +1,95 @@
|
||||
package dbtypes
|
||||
|
||||
import (
|
||||
"time"
|
||||
|
||||
"git.andr3h3nriqu3s.com/andr3/fyp/logic/db"
|
||||
)
|
||||
|
||||
type DefinitionStatus int
|
||||
|
||||
const (
|
||||
DEFINITION_STATUS_CANCELD_TRAINING DefinitionStatus = -4
|
||||
DEFINITION_STATUS_FAILED_TRAINING = -3
|
||||
DEFINITION_STATUS_PRE_INIT = 1
|
||||
DEFINITION_STATUS_INIT = 2
|
||||
DEFINITION_STATUS_TRAINING = 3
|
||||
DEFINITION_STATUS_PAUSED_TRAINING = 6
|
||||
DEFINITION_STATUS_TRANIED = 4
|
||||
DEFINITION_STATUS_READY = 5
|
||||
)
|
||||
|
||||
type Definition struct {
|
||||
Id string `db:"md.id" json:"id"`
|
||||
ModelId string `db:"md.model_id" json:"model_id"`
|
||||
Accuracy float64 `db:"md.accuracy" json:"accuracy"`
|
||||
TargetAccuracy int `db:"md.target_accuracy" json:"target_accuracy"`
|
||||
Epoch int `db:"md.epoch" json:"epoch"`
|
||||
Status int `db:"md.status" json:"status"`
|
||||
CreatedOn time.Time `db:"md.created_on" json:"created"`
|
||||
EpochProgress int `db:"md.epoch_progress" json:"epoch_progress"`
|
||||
}
|
||||
|
||||
type SortByAccuracyDefinitions []*Definition
|
||||
|
||||
func (nf SortByAccuracyDefinitions) Len() int { return len(nf) }
|
||||
func (nf SortByAccuracyDefinitions) Swap(i, j int) { nf[i], nf[j] = nf[j], nf[i] }
|
||||
func (nf SortByAccuracyDefinitions) Less(i, j int) bool {
|
||||
return nf[i].Accuracy < nf[j].Accuracy
|
||||
}
|
||||
|
||||
func GetDefinition(db db.Db, definition_id string) (definition Definition, err error) {
|
||||
err = GetDBOnce(db, &definition, "model_definition as md where id=$1;", definition_id)
|
||||
return
|
||||
}
|
||||
|
||||
func MakeDefenition(db db.Db, model_id string, target_accuracy int) (definition Definition, err error) {
|
||||
var NewDefinition = struct {
|
||||
ModelId string `db:"model_id"`
|
||||
TargetAccuracy int `db:"target_accuracy"`
|
||||
}{ModelId: model_id, TargetAccuracy: target_accuracy}
|
||||
|
||||
id, err := InsertReturnId(db, &NewDefinition, "model_definition", "id")
|
||||
if err != nil {
|
||||
return
|
||||
}
|
||||
return GetDefinition(db, id)
|
||||
}
|
||||
|
||||
func (d Definition) UpdateStatus(db db.Db, status DefinitionStatus) (err error) {
|
||||
_, err = db.Exec("update model_definition set status=$1 where id=$2", status, d.Id)
|
||||
return
|
||||
}
|
||||
|
||||
func (d Definition) MakeLayer(db db.Db, layer_order int, layer_type LayerType, shape string) (layer Layer, err error) {
|
||||
var NewLayer = struct {
|
||||
DefinitionId string `db:"def_id"`
|
||||
LayerOrder int `db:"layer_order"`
|
||||
LayerType LayerType `db:"layer_type"`
|
||||
Shape string `db:"shape"`
|
||||
}{
|
||||
DefinitionId: d.Id,
|
||||
LayerOrder: layer_order,
|
||||
LayerType: layer_type,
|
||||
Shape: shape,
|
||||
}
|
||||
|
||||
id, err := InsertReturnId(db, &NewLayer, "model_definition_layer", "id")
|
||||
if err != nil {
|
||||
return
|
||||
}
|
||||
|
||||
return GetLayer(db, id)
|
||||
}
|
||||
|
||||
func (d Definition) GetLayers(db db.Db, filter string, args ...any) (layer []*Layer, err error) {
|
||||
args = append(args, d.Id)
|
||||
return GetDbMultitple[Layer](db, "model_definition_layer as mdl where mdl.def_id=$1 "+filter, args...)
|
||||
}
|
||||
|
||||
func (d *Definition) UpdateAfterEpoch(db db.Db, accuracy float64, epoch int) (err error) {
|
||||
d.Accuracy = accuracy
|
||||
d.Epoch += epoch
|
||||
_, err = db.Exec("update model_definition set epoch=$1, accuracy=$2 where id=$3", d.Epoch, d.Accuracy, d.Id)
|
||||
return
|
||||
}
|
||||
69
logic/db_types/layer.go
Normal file
69
logic/db_types/layer.go
Normal file
@@ -0,0 +1,69 @@
|
||||
package dbtypes
|
||||
|
||||
import (
|
||||
"encoding/json"
|
||||
"fmt"
|
||||
|
||||
"git.andr3h3nriqu3s.com/andr3/fyp/logic/db"
|
||||
"github.com/charmbracelet/log"
|
||||
)
|
||||
|
||||
type LayerType int
|
||||
|
||||
const (
|
||||
LAYER_INPUT LayerType = 1
|
||||
LAYER_DENSE = 2
|
||||
LAYER_FLATTEN = 3
|
||||
LAYER_SIMPLE_BLOCK = 4
|
||||
)
|
||||
|
||||
type Layer struct {
|
||||
Id string `db:"mdl.id" json:"id"`
|
||||
DefinitionId string `db:"mdl.def_id" json:"definition_id"`
|
||||
LayerOrder int `db:"mdl.layer_order" json:"layer_order"`
|
||||
LayerType LayerType `db:"mdl.layer_type" json:"layer_type"`
|
||||
Shape string `db:"mdl.shape" json:"shape"`
|
||||
ExpType int `db:"mdl.exp_type" json:"exp_type"`
|
||||
}
|
||||
|
||||
func (x *Layer) ShapeToSize() {
|
||||
v := x.GetShape()
|
||||
switch x.LayerType {
|
||||
case LAYER_INPUT:
|
||||
x.Shape = fmt.Sprintf("%d,%d", v[1], v[2])
|
||||
case LAYER_DENSE:
|
||||
x.Shape = fmt.Sprintf("(%d)", v[0])
|
||||
default:
|
||||
x.Shape = "ERROR"
|
||||
}
|
||||
}
|
||||
|
||||
func ShapeToString(args ...int) string {
|
||||
text, err := json.Marshal(args)
|
||||
if err != nil {
|
||||
log.Error("json err!", "err", err)
|
||||
panic("Could not generate Shape")
|
||||
}
|
||||
return string(text)
|
||||
}
|
||||
|
||||
func StringToShape(str string) (shape []int64) {
|
||||
err := json.Unmarshal([]byte(str), &shape)
|
||||
if err != nil {
|
||||
log.Error("json err!", "err", err)
|
||||
panic("Could not parse Shape")
|
||||
}
|
||||
return
|
||||
}
|
||||
|
||||
func (l Layer) GetShape() []int64 {
|
||||
if l.Shape == "" {
|
||||
return []int64{}
|
||||
}
|
||||
return StringToShape(l.Shape)
|
||||
}
|
||||
|
||||
func GetLayer(db db.Db, layer_id string) (layer Layer, err error) {
|
||||
err = GetDBOnce(db, &layer, "model_definition_layer as mdl where mdl.id=$1", layer_id)
|
||||
return
|
||||
}
|
||||
@@ -2,23 +2,25 @@ package dbtypes
|
||||
|
||||
import (
|
||||
"errors"
|
||||
"fmt"
|
||||
|
||||
"git.andr3h3nriqu3s.com/andr3/fyp/logic/db"
|
||||
)
|
||||
|
||||
const (
|
||||
FAILED_TRAINING = -4
|
||||
FAILED_PREPARING_TRAINING = -3
|
||||
FAILED_PREPARING_ZIP_FILE = -2
|
||||
FAILED_PREPARING = -1
|
||||
type ModelStatus int
|
||||
|
||||
PREPARING = 1
|
||||
CONFIRM_PRE_TRAINING = 2
|
||||
PREPARING_ZIP_FILE = 3
|
||||
TRAINING = 4
|
||||
READY = 5
|
||||
READY_ALTERATION = 6
|
||||
READY_ALTERATION_FAILED = -6
|
||||
const (
|
||||
FAILED_TRAINING ModelStatus = -4
|
||||
FAILED_PREPARING_TRAINING = -3
|
||||
FAILED_PREPARING_ZIP_FILE = -2
|
||||
FAILED_PREPARING = -1
|
||||
PREPARING = 1
|
||||
CONFIRM_PRE_TRAINING = 2
|
||||
PREPARING_ZIP_FILE = 3
|
||||
TRAINING = 4
|
||||
READY = 5
|
||||
READY_ALTERATION = 6
|
||||
READY_ALTERATION_FAILED = -6
|
||||
|
||||
READY_RETRAIN = 7
|
||||
READY_RETRAIN_FAILED = -7
|
||||
@@ -26,15 +28,6 @@ const (
|
||||
|
||||
type ModelDefinitionStatus int
|
||||
|
||||
type LayerType int
|
||||
|
||||
const (
|
||||
LAYER_INPUT LayerType = 1
|
||||
LAYER_DENSE = 2
|
||||
LAYER_FLATTEN = 3
|
||||
LAYER_SIMPLE_BLOCK = 4
|
||||
)
|
||||
|
||||
const (
|
||||
MODEL_DEFINITION_STATUS_CANCELD_TRAINING ModelDefinitionStatus = -4
|
||||
MODEL_DEFINITION_STATUS_FAILED_TRAINING = -3
|
||||
@@ -46,14 +39,6 @@ const (
|
||||
MODEL_DEFINITION_STATUS_READY = 5
|
||||
)
|
||||
|
||||
type ModelClassStatus int
|
||||
|
||||
const (
|
||||
MODEL_CLASS_STATUS_TO_TRAIN ModelClassStatus = 1
|
||||
MODEL_CLASS_STATUS_TRAINING = 2
|
||||
MODEL_CLASS_STATUS_TRAINED = 3
|
||||
)
|
||||
|
||||
type ModelHeadStatus int
|
||||
|
||||
const (
|
||||
@@ -65,17 +50,16 @@ const (
|
||||
)
|
||||
|
||||
type BaseModel struct {
|
||||
Name string
|
||||
Status int
|
||||
Id string
|
||||
|
||||
ModelType int `db:"model_type"`
|
||||
ImageModeRaw string `db:"color_mode"`
|
||||
ImageMode int `db:"0"`
|
||||
Width int
|
||||
Height int
|
||||
Format string
|
||||
CanTrain int `db:"can_train"`
|
||||
Name string `json:"name"`
|
||||
Status int `json:"status"`
|
||||
Id string `json:"id"`
|
||||
ModelType int `db:"model_type" json:"model_type"`
|
||||
ImageModeRaw string `db:"color_mode" json:"image_more_raw"`
|
||||
ImageMode int `db:"0" json:"image_mode"`
|
||||
Width int `json:"width"`
|
||||
Height int `json:"height"`
|
||||
Format string `json:"format"`
|
||||
CanTrain int `db:"can_train" json:"can_train"`
|
||||
}
|
||||
|
||||
var ModelNotFoundError = errors.New("Model not found error")
|
||||
@@ -97,6 +81,59 @@ func (m BaseModel) CanEval() bool {
|
||||
return true
|
||||
}
|
||||
|
||||
// DO NOT Pass un filtered data on filters
|
||||
func (m BaseModel) GetDefinitions(db db.Db, filters string, args ...any) ([]*Definition, error) {
|
||||
n_args := []any{m.Id}
|
||||
n_args = append(n_args, args...)
|
||||
return GetDbMultitple[Definition](db, fmt.Sprintf("model_definition as md where md.model_id=$1 %s", filters), n_args...)
|
||||
}
|
||||
|
||||
func (m BaseModel) GetClasses(db db.Db, filters string, args ...any) ([]*ModelClass, error) {
|
||||
n_args := []any{m.Id}
|
||||
n_args = append(n_args, args...)
|
||||
return GetDbMultitple[ModelClass](db, fmt.Sprintf("model_classes as mc where mc.model_id=$1 %s", filters), n_args...)
|
||||
}
|
||||
|
||||
func (m *BaseModel) UpdateStatus(db db.Db, status ModelStatus) (err error) {
|
||||
_, err = db.Exec("update models set status=$1 where id=$2", status, m.Id)
|
||||
return
|
||||
}
|
||||
|
||||
type DataPoint struct {
|
||||
Id string `json:"id"`
|
||||
Class int `json:"class"`
|
||||
Path string `json:"path"`
|
||||
}
|
||||
|
||||
func (m BaseModel) DataPoints(db db.Db, mode DATA_POINT_MODE) (data []DataPoint, err error) {
|
||||
rows, err := db.Query(
|
||||
"select mdp.id, mc.class_order, mdp.file_path from model_data_point as mdp inner "+
|
||||
"join model_classes as mc on mc.id = mdp.class_id "+
|
||||
"where mc.model_id = $1 and mdp.model_mode=$2;",
|
||||
m.Id, mode)
|
||||
if err != nil {
|
||||
return
|
||||
}
|
||||
defer rows.Close()
|
||||
|
||||
data = []DataPoint{}
|
||||
|
||||
for rows.Next() {
|
||||
var id string
|
||||
var class_order int
|
||||
var file_path string
|
||||
if err = rows.Scan(&id, &class_order, &file_path); err != nil {
|
||||
return
|
||||
}
|
||||
data = append(data, DataPoint{
|
||||
Id: id,
|
||||
Path: file_path,
|
||||
Class: class_order,
|
||||
})
|
||||
}
|
||||
return
|
||||
}
|
||||
|
||||
func StringToImageMode(colorMode string) int {
|
||||
switch colorMode {
|
||||
case "greyscale":
|
||||
|
||||
@@ -14,10 +14,19 @@ const (
|
||||
)
|
||||
|
||||
type User struct {
|
||||
Id string `db:"u.id"`
|
||||
Username string `db:"u.username"`
|
||||
Email string `db:"u.email"`
|
||||
UserType int `db:"u.user_type"`
|
||||
Id string `db:"u.id" json:"id"`
|
||||
Username string `db:"u.username" json:"username"`
|
||||
Email string `db:"u.email" json:"email"`
|
||||
UserType int `db:"u.user_type" json:"user_type"`
|
||||
}
|
||||
|
||||
func UserFromId(db db.Db, id string) (*User, error) {
|
||||
var user User
|
||||
err := GetDBOnce(db, &user, "users as u where u.id=$1", id)
|
||||
if err != nil {
|
||||
return nil, err
|
||||
}
|
||||
return &user, nil
|
||||
}
|
||||
|
||||
func UserFromToken(db db.Db, token string) (*User, error) {
|
||||
|
||||
@@ -7,15 +7,15 @@ import (
|
||||
. "git.andr3h3nriqu3s.com/andr3/fyp/logic/db_types"
|
||||
)
|
||||
|
||||
type ModelClass struct {
|
||||
type ModelClassJSON struct {
|
||||
Id string `json:"id"`
|
||||
ModelId string `json:"model_id" db:"model_id"`
|
||||
Name string `json:"name"`
|
||||
Status int `json:"status"`
|
||||
}
|
||||
|
||||
func ListClasses(c BasePack, model_id string) (cls []*ModelClass, err error) {
|
||||
return GetDbMultitple[ModelClass](c.GetDb(), "model_classes where model_id=$1", model_id)
|
||||
func ListClasses(c BasePack, model_id string) (cls []*ModelClassJSON, err error) {
|
||||
return GetDbMultitple[ModelClassJSON](c.GetDb(), "model_classes where model_id=$1", model_id)
|
||||
}
|
||||
|
||||
func ModelHasDataPoints(db db.Db, model_id string) (result bool, err error) {
|
||||
|
||||
@@ -495,7 +495,7 @@ func handleDataUpload(handle *Handle) {
|
||||
return c.E500M("Could not create class", err)
|
||||
}
|
||||
|
||||
var modelClass model_classes.ModelClass
|
||||
var modelClass model_classes.ModelClassJSON
|
||||
err = GetDBOnce(c, &modelClass, "model_classes where id=$1;", id)
|
||||
if err != nil {
|
||||
return c.E500M("Failed to get class information but class was creted", err)
|
||||
@@ -704,7 +704,7 @@ func handleDataUpload(handle *Handle) {
|
||||
return c.Error500(err)
|
||||
}
|
||||
} else {
|
||||
_, err = handle.Db.Exec("delete from model_classes where model_id=$1 and status=$2;", model.Id, MODEL_CLASS_STATUS_TO_TRAIN)
|
||||
_, err = handle.Db.Exec("delete from model_classes where model_id=$1 and status=$2;", model.Id, CLASS_STATUS_TO_TRAIN)
|
||||
if err != nil {
|
||||
return c.Error500(err)
|
||||
}
|
||||
|
||||
@@ -51,7 +51,7 @@ func handleDelete(handle *Handle) {
|
||||
return c.E500M("Faield to get model", err)
|
||||
}
|
||||
|
||||
switch model.Status {
|
||||
switch ModelStatus(model.Status) {
|
||||
case FAILED_TRAINING:
|
||||
fallthrough
|
||||
case FAILED_PREPARING_ZIP_FILE:
|
||||
|
||||
@@ -35,9 +35,9 @@ func handleEdit(handle *Handle) {
|
||||
}
|
||||
|
||||
type ReturnType struct {
|
||||
Classes []*model_classes.ModelClass `json:"classes"`
|
||||
HasData bool `json:"has_data"`
|
||||
NumberOfInvalidImages int `json:"number_of_invalid_images"`
|
||||
Classes []*model_classes.ModelClassJSON `json:"classes"`
|
||||
HasData bool `json:"has_data"`
|
||||
NumberOfInvalidImages int `json:"number_of_invalid_images"`
|
||||
}
|
||||
|
||||
c.ShowMessage = false
|
||||
|
||||
118
logic/models/train/remote_train.go
Normal file
118
logic/models/train/remote_train.go
Normal file
@@ -0,0 +1,118 @@
|
||||
package models_train
|
||||
|
||||
import (
|
||||
"errors"
|
||||
|
||||
. "git.andr3h3nriqu3s.com/andr3/fyp/logic/db_types"
|
||||
. "git.andr3h3nriqu3s.com/andr3/fyp/logic/tasks/utils"
|
||||
. "git.andr3h3nriqu3s.com/andr3/fyp/logic/utils"
|
||||
"github.com/charmbracelet/log"
|
||||
"github.com/goccy/go-json"
|
||||
)
|
||||
|
||||
func PrepareTraining(handler *Handle, b BasePack, task Task, runner_id string) (err error) {
|
||||
l := b.GetLogger()
|
||||
|
||||
model, err := GetBaseModel(b.GetDb(), *task.ModelId)
|
||||
if err != nil {
|
||||
task.UpdateStatusLog(b, TASK_FAILED_RUNNING, "Failed to get model information")
|
||||
l.Error("Failed to get model information", "err", err)
|
||||
return err
|
||||
}
|
||||
|
||||
if model.Status != TRAINING {
|
||||
task.UpdateStatusLog(b, TASK_FAILED_RUNNING, "Model not in the correct status for training")
|
||||
return errors.New("Model not in the right status")
|
||||
}
|
||||
|
||||
// TODO do this when the runner says it's OK
|
||||
//task.UpdateStatusLog(b, TASK_RUNNING, "Training model")
|
||||
|
||||
// TODO move this to the runner part as well
|
||||
var dat struct {
|
||||
NumberOfModels int
|
||||
Accuracy int
|
||||
}
|
||||
|
||||
err = json.Unmarshal([]byte(task.ExtraTaskInfo), &dat)
|
||||
if err != nil {
|
||||
task.UpdateStatusLog(b, TASK_FAILED_RUNNING, "Failed to get model extra information")
|
||||
}
|
||||
|
||||
if model.ModelType == 2 {
|
||||
full_error := generateExpandableDefinitions(b, model, dat.Accuracy, dat.NumberOfModels)
|
||||
if full_error != nil {
|
||||
l.Error("Failed to generate defintions", "err", full_error)
|
||||
task.UpdateStatusLog(b, TASK_FAILED_RUNNING, "Failed generate model")
|
||||
return errors.New("Failed to generate definitions")
|
||||
}
|
||||
} else {
|
||||
error := generateDefinitions(b, model, dat.Accuracy, dat.NumberOfModels)
|
||||
if error != nil {
|
||||
task.UpdateStatusLog(b, TASK_FAILED_RUNNING, "Failed generate model")
|
||||
return errors.New("Failed to generate definitions")
|
||||
}
|
||||
}
|
||||
|
||||
runners := handler.DataMap["runners"].(map[string]interface{})
|
||||
runner := runners[runner_id].(map[string]interface{})
|
||||
runner["task"] = &task
|
||||
runners[runner_id] = runner
|
||||
handler.DataMap["runners"] = runners
|
||||
|
||||
return
|
||||
}
|
||||
|
||||
func CleanUpFailed(b BasePack, task *Task) {
|
||||
db := b.GetDb()
|
||||
l := b.GetLogger()
|
||||
model, err := GetBaseModel(db, *task.ModelId)
|
||||
if err != nil {
|
||||
l.Error("Failed to get model", "err", err)
|
||||
} else {
|
||||
err = model.UpdateStatus(db, FAILED_TRAINING)
|
||||
if err != nil {
|
||||
l.Error("Failed to get status", err)
|
||||
}
|
||||
}
|
||||
// Set the class status to trained
|
||||
err = SetModelClassStatus(b, CLASS_STATUS_TO_TRAIN, "model_id=$1 and status=$2;", model.Id, CLASS_STATUS_TRAINING)
|
||||
if err != nil {
|
||||
l.Error("Failed to set class status")
|
||||
return
|
||||
}
|
||||
}
|
||||
|
||||
func CleanUpFailedRetrain(b BasePack, task *Task) {
|
||||
db := b.GetDb()
|
||||
l := b.GetLogger()
|
||||
|
||||
model, err := GetBaseModel(db, *task.ModelId)
|
||||
if err != nil {
|
||||
l.Error("Failed to get model", "err", err)
|
||||
} else {
|
||||
err = model.UpdateStatus(db, FAILED_TRAINING)
|
||||
if err != nil {
|
||||
l.Error("Failed to get status", err)
|
||||
}
|
||||
}
|
||||
|
||||
ResetClasses(b, model)
|
||||
ModelUpdateStatus(b, model.Id, READY_RETRAIN_FAILED)
|
||||
|
||||
var defData struct {
|
||||
Id string `db:"md.id"`
|
||||
TargetAcuuracy float64 `db:"md.target_accuracy"`
|
||||
}
|
||||
|
||||
err = GetDBOnce(db, &defData, "models as m inner join model_definition as md on m.id = md.model_id where m.id=$1;", task.ModelId)
|
||||
if err != nil {
|
||||
log.Error("failed to get def data", err)
|
||||
return
|
||||
}
|
||||
|
||||
_, err_ := db.Exec("delete from exp_model_head where def_id=$1 and status in (2,3)", defData.Id)
|
||||
if err_ != nil {
|
||||
panic(err_)
|
||||
}
|
||||
}
|
||||
@@ -17,7 +17,7 @@ func handleRest(handle *Handle) {
|
||||
return c.E500M("Failed to get model", err)
|
||||
}
|
||||
|
||||
if model.Status != FAILED_PREPARING_TRAINING && model.Status != FAILED_TRAINING {
|
||||
if model.Status != FAILED_PREPARING_TRAINING && model.Status != int(FAILED_TRAINING) {
|
||||
return c.JsonBadRequest("Model is not in status that be reset")
|
||||
}
|
||||
|
||||
|
||||
File diff suppressed because it is too large
Load Diff
@@ -8,4 +8,6 @@ func HandleTasks(handle *Handle) {
|
||||
handleUpload(handle)
|
||||
handleList(handle)
|
||||
handleRequests(handle)
|
||||
handleRemoteRunner(handle)
|
||||
handleRunnerData(handle)
|
||||
}
|
||||
|
||||
960
logic/tasks/runner.go
Normal file
960
logic/tasks/runner.go
Normal file
@@ -0,0 +1,960 @@
|
||||
package tasks
|
||||
|
||||
import (
|
||||
"os"
|
||||
"path"
|
||||
"sync"
|
||||
"time"
|
||||
|
||||
. "git.andr3h3nriqu3s.com/andr3/fyp/logic/db_types"
|
||||
. "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) {
|
||||
runner, err := GetRunner(c, dat.Id)
|
||||
if err == NotFoundError {
|
||||
e = c.JsonBadRequest("Could not find runner, please register runner first")
|
||||
return
|
||||
} else if err != nil {
|
||||
e = c.E500M("Failed to get information about the runner", err)
|
||||
return
|
||||
}
|
||||
|
||||
if runner.Token != *c.Token {
|
||||
return nil, c.SendJSONStatus(401, "Only runners can use this funcion")
|
||||
}
|
||||
|
||||
return
|
||||
}
|
||||
|
||||
type VerifyTask struct {
|
||||
Id string `json:"id" validate:"required"`
|
||||
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()
|
||||
defer mutex.Unlock()
|
||||
|
||||
var runners map[string]interface{} = x.DataMap["runners"].(map[string]interface{})
|
||||
if runners[dat.Id] == nil {
|
||||
return nil, c.JsonBadRequest("Runner not active")
|
||||
}
|
||||
|
||||
var runner_data map[string]interface{} = runners[dat.Id].(map[string]interface{})
|
||||
|
||||
if runner_data["task"] == nil {
|
||||
return nil, c.SendJSONStatus(404, "No active task")
|
||||
}
|
||||
|
||||
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 {
|
||||
Token string `json:"token" validate:"required"`
|
||||
Type RunnerType `json:"type" validate:"required"`
|
||||
}
|
||||
PostAuthJson(x, "/tasks/runner/register", User_Normal, func(c *Context, dat *RegisterRunner) *Error {
|
||||
if *c.Token != dat.Token {
|
||||
// TODO do admin
|
||||
return c.E500M("Please make sure that the token is the same that is being registered", nil)
|
||||
}
|
||||
|
||||
c.Logger.Info("test", "dat", dat)
|
||||
|
||||
var runner Runner
|
||||
err := GetDBOnce(c, &runner, "remote_runner as ru where token=$1", dat.Token)
|
||||
if err != NotFoundError && err != nil {
|
||||
return c.E500M("Failed to get information remote runners", err)
|
||||
}
|
||||
if err != NotFoundError {
|
||||
return c.JsonBadRequest("Token is already registered by a runner")
|
||||
}
|
||||
|
||||
// TODO get id from token passed by when doing admin
|
||||
var userId = c.User.Id
|
||||
|
||||
var new_runner = struct {
|
||||
Type RunnerType
|
||||
UserId string `db:"user_id"`
|
||||
Token string
|
||||
}{
|
||||
Type: dat.Type,
|
||||
Token: dat.Token,
|
||||
UserId: userId,
|
||||
}
|
||||
|
||||
id, err := InsertReturnId(c, &new_runner, "remote_runner", "id")
|
||||
if err != nil {
|
||||
return c.E500M("Failed to create remote runner", err)
|
||||
}
|
||||
|
||||
return c.SendJSON(struct {
|
||||
Id string `json:"id"`
|
||||
}{
|
||||
Id: id,
|
||||
})
|
||||
})
|
||||
|
||||
// TODO remove runner
|
||||
|
||||
PostAuthJson(x, "/tasks/runner/init", User_Normal, func(c *Context, dat *JustId) *Error {
|
||||
runner, error := verifyRunner(c, dat)
|
||||
if error != nil {
|
||||
return error
|
||||
}
|
||||
|
||||
mutex := x.DataMap["runners_mutex"].(*sync.Mutex)
|
||||
mutex.Lock()
|
||||
defer mutex.Unlock()
|
||||
|
||||
var runners map[string]interface{} = x.DataMap["runners"].(map[string]interface{})
|
||||
if runners[dat.Id] != nil {
|
||||
c.Logger.Info("Logger trying to register but already registerd")
|
||||
c.ShowMessage = false
|
||||
return c.SendJSON("Ok")
|
||||
}
|
||||
|
||||
var new_runner = map[string]interface{}{}
|
||||
new_runner["last_time_check"] = time.Now()
|
||||
new_runner["runner_info"] = runner
|
||||
|
||||
runners[dat.Id] = new_runner
|
||||
|
||||
x.DataMap["runners"] = runners
|
||||
|
||||
return c.SendJSON("Ok")
|
||||
})
|
||||
|
||||
PostAuthJson(x, "/tasks/runner/active", User_Normal, func(c *Context, dat *JustId) *Error {
|
||||
_, error := verifyRunner(c, dat)
|
||||
if error != nil {
|
||||
return error
|
||||
}
|
||||
|
||||
mutex := x.DataMap["runners_mutex"].(*sync.Mutex)
|
||||
mutex.Lock()
|
||||
defer mutex.Unlock()
|
||||
|
||||
var runners map[string]interface{} = x.DataMap["runners"].(map[string]interface{})
|
||||
if runners[dat.Id] == nil {
|
||||
return c.JsonBadRequest("Runner not active")
|
||||
}
|
||||
|
||||
var runner_data map[string]interface{} = runners[dat.Id].(map[string]interface{})
|
||||
|
||||
if runner_data["task"] == nil {
|
||||
c.ShowMessage = false
|
||||
return c.SendJSONStatus(404, "No active task")
|
||||
}
|
||||
|
||||
c.ShowMessage = false
|
||||
// This should be a task obj
|
||||
return c.SendJSON(runner_data["task"])
|
||||
})
|
||||
|
||||
PostAuthJson(x, "/tasks/runner/ready", 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
|
||||
}
|
||||
|
||||
err := task.UpdateStatus(c, TASK_RUNNING, "Task Running on Runner")
|
||||
if err != nil {
|
||||
return c.E500M("Failed to set task status", err)
|
||||
}
|
||||
|
||||
return c.SendJSON("Ok")
|
||||
})
|
||||
|
||||
type TaskFail struct {
|
||||
Id string `json:"id" validate:"required"`
|
||||
TaskId string `json:"taskId" validate:"required"`
|
||||
Reason string `json:"reason" validate:"required"`
|
||||
}
|
||||
PostAuthJson(x, "/tasks/runner/fail", User_Normal, func(c *Context, dat *TaskFail) *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
|
||||
}
|
||||
|
||||
err := task.UpdateStatus(c, TASK_FAILED_RUNNING, dat.Reason)
|
||||
if err != nil {
|
||||
return c.E500M("Failed to set task status", err)
|
||||
}
|
||||
|
||||
// Do extra clean up on tasks
|
||||
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:
|
||||
panic("Do not know how to handle this")
|
||||
}
|
||||
|
||||
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/defs", 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
|
||||
}
|
||||
|
||||
var status DefinitionStatus
|
||||
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:
|
||||
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)
|
||||
}
|
||||
|
||||
defs, err := model.GetDefinitions(c, "and md.status=$2", status)
|
||||
if err != nil {
|
||||
return c.E500M("Failed to get the model definitions", err)
|
||||
}
|
||||
|
||||
return c.SendJSON(defs)
|
||||
})
|
||||
|
||||
PostAuthJson(x, "/tasks/runner/classes", 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
|
||||
}
|
||||
|
||||
model, err := GetBaseModel(c, *task.ModelId)
|
||||
if err != nil {
|
||||
return c.E500M("Failed to get model information", 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")
|
||||
}
|
||||
|
||||
})
|
||||
|
||||
type RunnerTrainDefStatus struct {
|
||||
Id string `json:"id" validate:"required"`
|
||||
TaskId string `json:"taskId" validate:"required"`
|
||||
DefId string `json:"defId" validate:"required"`
|
||||
Status DefinitionStatus `json:"status" validate:"required"`
|
||||
}
|
||||
PostAuthJson(x, "/tasks/runner/train/def/status", User_Normal, func(c *Context, dat *RunnerTrainDefStatus) *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_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")
|
||||
}
|
||||
|
||||
def, err := GetDefinition(c, dat.DefId)
|
||||
if err != nil {
|
||||
return c.E500M("Failed to get definition information", err)
|
||||
}
|
||||
|
||||
err = def.UpdateStatus(c, dat.Status)
|
||||
if err != nil {
|
||||
return c.E500M("Failed to update model status", err)
|
||||
}
|
||||
|
||||
return c.SendJSON("Ok")
|
||||
})
|
||||
|
||||
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/head/status", User_Normal, func(c *Context, dat *RunnerTrainDefHeadStatus) *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_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")
|
||||
}
|
||||
|
||||
def, err := GetDefinition(c, dat.DefId)
|
||||
if err != nil {
|
||||
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)
|
||||
}
|
||||
|
||||
return c.SendJSON(layers)
|
||||
})
|
||||
|
||||
PostAuthJson(x, "/tasks/runner/train/datapoints", 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
|
||||
}
|
||||
|
||||
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")
|
||||
}
|
||||
|
||||
model, err := GetBaseModel(c, *task.ModelId)
|
||||
if err != nil {
|
||||
return c.E500M("Failed to get model information", err)
|
||||
}
|
||||
|
||||
training_points, err := model.DataPoints(c, DATA_POINT_MODE_TRAINING)
|
||||
if err != nil {
|
||||
return c.E500M("Failed to get the model classes", err)
|
||||
}
|
||||
testing_points, err := model.DataPoints(c, DATA_POINT_MODE_TRAINING)
|
||||
if err != nil {
|
||||
return c.E500M("Failed to get the model classes", err)
|
||||
}
|
||||
|
||||
return c.SendJSON(struct {
|
||||
Testing []DataPoint `json:"testing"`
|
||||
Training []DataPoint `json:"training"`
|
||||
}{
|
||||
Testing: testing_points,
|
||||
Training: training_points,
|
||||
})
|
||||
})
|
||||
|
||||
type RunnerTrainDefEpoch struct {
|
||||
Id string `json:"id" validate:"required"`
|
||||
TaskId string `json:"taskId" validate:"required"`
|
||||
DefId string `json:"defId" validate:"required"`
|
||||
Epoch int `json:"epoch" validate:"required"`
|
||||
Accuracy float64 `json:"accuracy" validate:"required"`
|
||||
}
|
||||
PostAuthJson(x, "/tasks/runner/train/epoch", User_Normal, func(c *Context, dat *RunnerTrainDefEpoch) *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_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")
|
||||
}
|
||||
|
||||
def, err := GetDefinition(c, dat.DefId)
|
||||
if err != nil {
|
||||
return c.E500M("Failed to get definition information", err)
|
||||
}
|
||||
|
||||
err = def.UpdateAfterEpoch(c, dat.Accuracy, dat.Epoch)
|
||||
if err != nil {
|
||||
return c.E500M("Failed to update model", err)
|
||||
}
|
||||
|
||||
return c.SendJSON("Ok")
|
||||
})
|
||||
|
||||
PostAuthJson(x, "/tasks/runner/train/mark-failed", 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, &VerifyTask{
|
||||
Id: dat.Id,
|
||||
TaskId: dat.TaskId,
|
||||
})
|
||||
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")
|
||||
}
|
||||
|
||||
_, err := c.Exec(
|
||||
"update model_definition set status=$1 "+
|
||||
"where model_id=$2 and status in ($3, $4)",
|
||||
MODEL_DEFINITION_STATUS_CANCELD_TRAINING,
|
||||
task.ModelId,
|
||||
MODEL_DEFINITION_STATUS_TRAINING,
|
||||
MODEL_DEFINITION_STATUS_PAUSED_TRAINING,
|
||||
)
|
||||
if err != nil {
|
||||
return c.E500M("Failed to mark definition as failed", err)
|
||||
}
|
||||
|
||||
return c.SendJSON("Ok")
|
||||
})
|
||||
|
||||
PostAuthJson(x, "/tasks/runner/model", 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
|
||||
}
|
||||
|
||||
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:
|
||||
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)
|
||||
}
|
||||
|
||||
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 {
|
||||
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)
|
||||
}
|
||||
|
||||
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!")
|
||||
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 := c.Query("select id from model_definition where status != $1 and model_id=$2", MODEL_DEFINITION_STATUS_READY, model.Id)
|
||||
if 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)
|
||||
}
|
||||
defer to_delete.Close()
|
||||
|
||||
for to_delete.Next() {
|
||||
var id string
|
||||
if err = to_delete.Scan(&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)
|
||||
}
|
||||
os.RemoveAll(path.Join("savedData", model.Id, "defs", 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)
|
||||
}
|
||||
|
||||
// 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")
|
||||
return c.E500M("Failed to update status of model", err)
|
||||
}
|
||||
|
||||
task.UpdateStatusLog(c, TASK_DONE, "Model finished training")
|
||||
|
||||
clearRunnerTask(x, dat.Id)
|
||||
return c.SendJSON("Ok")
|
||||
})
|
||||
|
||||
type RunnerClassDone struct {
|
||||
Id string `json:"id" validate:"required"`
|
||||
TaskId string `json:"taskId" validate:"required"`
|
||||
Result string `json:"result" validate:"required"`
|
||||
}
|
||||
PostAuthJson(x, "/tasks/runner/class/done", User_Normal, func(c *Context, dat *RunnerClassDone) *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_CLASSIFICATION) {
|
||||
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")
|
||||
}
|
||||
|
||||
err := task.SetResultText(c, dat.Result)
|
||||
if err != nil {
|
||||
return c.E500M("Failed to update the task", err)
|
||||
}
|
||||
|
||||
err = task.UpdateStatus(c, TASK_DONE, "Task completed")
|
||||
if err != nil {
|
||||
return c.E500M("Failed to update task", err)
|
||||
}
|
||||
|
||||
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/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)
|
||||
}
|
||||
|
||||
defs, err := model.GetDefinitions(c, "")
|
||||
if err != nil {
|
||||
return c.E500M("Failed to get definitions", err)
|
||||
}
|
||||
|
||||
_, err = c.Exec("update exp_model_head set status=$1 where status=$2 and def_id=$3", MODEL_HEAD_STATUS_READY, MODEL_HEAD_STATUS_TRAINING, defs[0].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")
|
||||
})
|
||||
|
||||
}
|
||||
@@ -5,6 +5,7 @@ import (
|
||||
"math"
|
||||
"os"
|
||||
"runtime/debug"
|
||||
"sync"
|
||||
"time"
|
||||
|
||||
"github.com/charmbracelet/log"
|
||||
@@ -90,6 +91,57 @@ func runner(config Config, db db.Db, task_channel chan Task, index int, back_cha
|
||||
}
|
||||
}
|
||||
|
||||
/**
|
||||
* Handle remote runner
|
||||
*/
|
||||
func handleRemoteTask(handler *Handle, base BasePack, runner_id string, task Task) {
|
||||
logger := log.NewWithOptions(os.Stdout, log.Options{
|
||||
ReportCaller: true,
|
||||
ReportTimestamp: true,
|
||||
TimeFormat: time.Kitchen,
|
||||
Prefix: fmt.Sprintf("Runner pre %s", runner_id),
|
||||
})
|
||||
defer func() {
|
||||
if r := recover(); r != nil {
|
||||
logger.Error("Runner failed to setup for runner", "due to", r, "stack", string(debug.Stack()))
|
||||
// TODO maybe create better failed task
|
||||
task.UpdateStatusLog(base, TASK_FAILED_RUNNING, "Failed to setup task for runner")
|
||||
}
|
||||
}()
|
||||
|
||||
err := task.UpdateStatus(base, TASK_PICKED_UP, "Failed to setup task for runner")
|
||||
if err != nil {
|
||||
logger.Error("Failed to mark task as PICK UP")
|
||||
return
|
||||
}
|
||||
|
||||
mutex := handler.DataMap["runners_mutex"].(*sync.Mutex)
|
||||
mutex.Lock()
|
||||
defer mutex.Unlock()
|
||||
|
||||
switch task.TaskType {
|
||||
case int(TASK_TYPE_RETRAINING):
|
||||
runners := handler.DataMap["runners"].(map[string]interface{})
|
||||
runner := runners[runner_id].(map[string]interface{})
|
||||
runner["task"] = &task
|
||||
runners[runner_id] = runner
|
||||
handler.DataMap["runners"] = runners
|
||||
case int(TASK_TYPE_TRAINING):
|
||||
if err := PrepareTraining(handler, base, task, runner_id); err != nil {
|
||||
logger.Error("Failed to prepare for training", "err", err)
|
||||
}
|
||||
case int(TASK_TYPE_CLASSIFICATION):
|
||||
runners := handler.DataMap["runners"].(map[string]interface{})
|
||||
runner := runners[runner_id].(map[string]interface{})
|
||||
runner["task"] = &task
|
||||
runners[runner_id] = runner
|
||||
handler.DataMap["runners"] = runners
|
||||
default:
|
||||
logger.Error("Not sure what to do panicing", "taskType", task.TaskType)
|
||||
panic("not sure what to do")
|
||||
}
|
||||
}
|
||||
|
||||
/**
|
||||
* Tells the orcchestator to look at the task list from time to time
|
||||
*/
|
||||
@@ -125,7 +177,7 @@ func attentionSeeker(config Config, back_channel chan int) {
|
||||
/**
|
||||
* Manages what worker should to Work
|
||||
*/
|
||||
func RunnerOrchestrator(db db.Db, config Config) {
|
||||
func RunnerOrchestrator(db db.Db, config Config, handler *Handle) {
|
||||
logger := log.NewWithOptions(os.Stdout, log.Options{
|
||||
ReportCaller: true,
|
||||
ReportTimestamp: true,
|
||||
@@ -133,6 +185,14 @@ func RunnerOrchestrator(db db.Db, config Config) {
|
||||
Prefix: "Runner Orchestrator Logger",
|
||||
})
|
||||
|
||||
setupHandle(handler)
|
||||
|
||||
base := BasePackStruct{
|
||||
Db: db,
|
||||
Logger: logger,
|
||||
Host: config.Hostname,
|
||||
}
|
||||
|
||||
gpu_workers := config.GpuWorker.NumberOfWorkers
|
||||
|
||||
logger.Info("Starting runners")
|
||||
@@ -144,12 +204,12 @@ func RunnerOrchestrator(db db.Db, config Config) {
|
||||
|
||||
defer func() {
|
||||
if r := recover(); r != nil {
|
||||
logger.Error("Recovered in Orchestrator restarting", "due to", r)
|
||||
logger.Error("Recovered in Orchestrator restarting", "due to", r, "stack", string(debug.Stack()))
|
||||
for x := range task_runners {
|
||||
close(task_runners[x])
|
||||
}
|
||||
close(back_channel)
|
||||
go RunnerOrchestrator(db, config)
|
||||
go RunnerOrchestrator(db, config, handler)
|
||||
}
|
||||
}()
|
||||
|
||||
@@ -159,6 +219,10 @@ func RunnerOrchestrator(db db.Db, config Config) {
|
||||
for i := 0; i < gpu_workers; i++ {
|
||||
task_runners[i] = make(chan Task, 10)
|
||||
task_runners_used[i] = false
|
||||
AddLocalRunner(handler, LocalRunner{
|
||||
RunnerNum: i + 1,
|
||||
Task: nil,
|
||||
})
|
||||
go runner(config, db, task_runners[i], i+1, back_channel)
|
||||
}
|
||||
|
||||
@@ -166,14 +230,17 @@ func RunnerOrchestrator(db db.Db, config Config) {
|
||||
|
||||
for i := range back_channel {
|
||||
|
||||
if i > 0 {
|
||||
logger.Info("Runner freed", "runner", i)
|
||||
task_runners_used[i-1] = false
|
||||
} else if i < 0 {
|
||||
logger.Error("Runner died! Restarting!", "runner", i)
|
||||
i = int(math.Abs(float64(i)) - 1)
|
||||
task_runners_used[i] = false
|
||||
go runner(config, db, task_runners[i], i+1, back_channel)
|
||||
if i != 0 {
|
||||
if i > 0 {
|
||||
logger.Info("Runner freed", "runner", i)
|
||||
task_runners_used[i-1] = false
|
||||
} else if i < 0 {
|
||||
logger.Error("Runner died! Restarting!", "runner", i)
|
||||
i = int(math.Abs(float64(i)) - 1)
|
||||
task_runners_used[i] = false
|
||||
go runner(config, db, task_runners[i], i+1, back_channel)
|
||||
}
|
||||
AddLocalTask(handler, int(math.Abs(float64(i))), nil)
|
||||
}
|
||||
|
||||
if task_to_dispatch == nil {
|
||||
@@ -197,11 +264,38 @@ func RunnerOrchestrator(db db.Db, config Config) {
|
||||
}
|
||||
}
|
||||
|
||||
if task_to_dispatch != nil && task_to_dispatch.TaskType != int(TASK_TYPE_DELETE_USER) {
|
||||
// TODO split tasks into cpu tasks and GPU tasks
|
||||
mutex := handler.DataMap["runners_mutex"].(*sync.Mutex)
|
||||
mutex.Lock()
|
||||
remote_runners := handler.DataMap["runners"].(map[string]interface{})
|
||||
|
||||
for k, v := range remote_runners {
|
||||
runner_data := v.(map[string]interface{})
|
||||
runner_info := runner_data["runner_info"].(*Runner)
|
||||
|
||||
if runner_data["task"] != nil {
|
||||
continue
|
||||
}
|
||||
|
||||
if runner_info.UserId != task_to_dispatch.UserId {
|
||||
continue
|
||||
}
|
||||
|
||||
go handleRemoteTask(handler, base, k, *task_to_dispatch)
|
||||
task_to_dispatch = nil
|
||||
break
|
||||
}
|
||||
|
||||
mutex.Unlock()
|
||||
}
|
||||
|
||||
if task_to_dispatch != nil {
|
||||
for i := 0; i < len(task_runners_used); i += 1 {
|
||||
if !task_runners_used[i] {
|
||||
task_runners[i] <- *task_to_dispatch
|
||||
task_runners_used[i] = true
|
||||
AddLocalTask(handler, i+1, task_to_dispatch)
|
||||
task_to_dispatch = nil
|
||||
break
|
||||
}
|
||||
@@ -211,6 +305,6 @@ func RunnerOrchestrator(db db.Db, config Config) {
|
||||
}
|
||||
}
|
||||
|
||||
func StartRunners(db db.Db, config Config) {
|
||||
go RunnerOrchestrator(db, config)
|
||||
func StartRunners(db db.Db, config Config, handler *Handle) {
|
||||
go RunnerOrchestrator(db, config, handler)
|
||||
}
|
||||
|
||||
51
logic/tasks/runner/utils.go
Normal file
51
logic/tasks/runner/utils.go
Normal file
@@ -0,0 +1,51 @@
|
||||
package task_runner
|
||||
|
||||
import (
|
||||
"sync"
|
||||
|
||||
. "git.andr3h3nriqu3s.com/andr3/fyp/logic/tasks/utils"
|
||||
. "git.andr3h3nriqu3s.com/andr3/fyp/logic/utils"
|
||||
)
|
||||
|
||||
type LocalRunner struct {
|
||||
RunnerNum int `json:"id"`
|
||||
Task *Task `json:"task"`
|
||||
}
|
||||
|
||||
type LocalRunners map[int]*LocalRunner
|
||||
|
||||
func LockRunners(handler *Handle, t string) *sync.Mutex {
|
||||
req := t + "_runners_mutex"
|
||||
if t == "" {
|
||||
req = "runners_mutex"
|
||||
}
|
||||
mutex := handler.DataMap[req].(*sync.Mutex)
|
||||
mutex.Lock()
|
||||
return mutex
|
||||
}
|
||||
|
||||
func setupHandle(handler *Handle) {
|
||||
// Setup Remote Runner data
|
||||
handler.DataMap["runners"] = map[string]interface{}{}
|
||||
handler.DataMap["runners_mutex"] = &sync.Mutex{}
|
||||
|
||||
// Setup Local Runner data
|
||||
handler.DataMap["local_runners"] = &LocalRunners{}
|
||||
handler.DataMap["local_runners_mutex"] = &sync.Mutex{}
|
||||
}
|
||||
|
||||
func AddLocalRunner(handler *Handle, runner LocalRunner) {
|
||||
mutex := LockRunners(handler, "local")
|
||||
defer mutex.Unlock()
|
||||
|
||||
runners := handler.DataMap["local_runners"].(*LocalRunners)
|
||||
(*runners)[runner.RunnerNum] = &runner
|
||||
}
|
||||
|
||||
func AddLocalTask(handler *Handle, runner_id int, task *Task) {
|
||||
mutex := LockRunners(handler, "local")
|
||||
defer mutex.Unlock()
|
||||
|
||||
runners := handler.DataMap["local_runners"].(*LocalRunners)
|
||||
(*(*runners)[runner_id]).Task = task
|
||||
}
|
||||
25
logic/tasks/runner_data.go
Normal file
25
logic/tasks/runner_data.go
Normal file
@@ -0,0 +1,25 @@
|
||||
package tasks
|
||||
|
||||
import (
|
||||
. "git.andr3h3nriqu3s.com/andr3/fyp/logic/db_types"
|
||||
. "git.andr3h3nriqu3s.com/andr3/fyp/logic/tasks/runner"
|
||||
. "git.andr3h3nriqu3s.com/andr3/fyp/logic/utils"
|
||||
)
|
||||
|
||||
func handleRunnerData(x *Handle) {
|
||||
type NonType struct{}
|
||||
PostAuthJson(x, "/tasks/runner/info", User_Admin, func(c *Context, dat *NonType) *Error {
|
||||
mutex_remote := LockRunners(x, "")
|
||||
defer mutex_remote.Unlock()
|
||||
mutex_local := LockRunners(x, "local")
|
||||
defer mutex_local.Unlock()
|
||||
|
||||
return c.SendJSON(struct {
|
||||
RemoteRunners map[string]interface{} `json:"remoteRunners"`
|
||||
LocalRunner *LocalRunners `json:"localRunners"`
|
||||
}{
|
||||
RemoteRunners: x.DataMap["runners"].(map[string]interface{}),
|
||||
LocalRunner: x.DataMap["local_runners"].(*LocalRunners),
|
||||
})
|
||||
})
|
||||
}
|
||||
29
logic/tasks/utils/runner.go
Normal file
29
logic/tasks/utils/runner.go
Normal file
@@ -0,0 +1,29 @@
|
||||
package tasks_utils
|
||||
|
||||
import (
|
||||
"time"
|
||||
|
||||
"git.andr3h3nriqu3s.com/andr3/fyp/logic/db"
|
||||
dbtypes "git.andr3h3nriqu3s.com/andr3/fyp/logic/db_types"
|
||||
)
|
||||
|
||||
type RunnerType int64
|
||||
|
||||
const (
|
||||
RUNNER_TYPE_GPU RunnerType = iota + 1
|
||||
)
|
||||
|
||||
type Runner struct {
|
||||
Id string `json:"id" db:"ru.id"`
|
||||
UserId string `json:"user_id" db:"ru.user_id"`
|
||||
Token string `json:"token" db:"ru.token"`
|
||||
Type RunnerType `json:"type" db:"ru.type"`
|
||||
CreateOn time.Time `json:"createOn" db:"ru.created_on"`
|
||||
}
|
||||
|
||||
func GetRunner(db db.Db, id string) (ru *Runner, err error) {
|
||||
var runner Runner
|
||||
err = dbtypes.GetDBOnce(db, &runner, "remote_runner as ru where ru.id=$1", id)
|
||||
ru = &runner
|
||||
return
|
||||
}
|
||||
@@ -101,7 +101,11 @@ func (t Task) SetResult(base BasePack, result any) (err error) {
|
||||
if err != nil {
|
||||
return
|
||||
}
|
||||
_, err = base.GetDb().Exec("update tasks set result=$1 where id=$2", text, t.Id)
|
||||
return t.SetResultText(base, string(text))
|
||||
}
|
||||
|
||||
func (t Task) SetResultText(base BasePack, text string) (err error) {
|
||||
_, err = base.GetDb().Exec("update tasks set result=$1 where id=$2", []byte(text), t.Id)
|
||||
return
|
||||
}
|
||||
|
||||
|
||||
@@ -241,6 +241,17 @@ func UsersEndpints(db db.Db, handle *Handle) {
|
||||
return c.SendJSON(userReturn)
|
||||
})
|
||||
|
||||
PostAuthJson(handle, "/user/info/get", User_Admin, func(c *Context, dat *JustId) *Error {
|
||||
var user *User
|
||||
user, err := UserFromId(c, dat.Id)
|
||||
if err == NotFoundError {
|
||||
return c.SendJSONStatus(404, "User not found")
|
||||
} else if err != nil {
|
||||
return c.E500M("Could not get user", err)
|
||||
}
|
||||
return c.SendJSON(user)
|
||||
})
|
||||
|
||||
// Handles updating users
|
||||
type UpdateUserData struct {
|
||||
Id string `json:"id"`
|
||||
|
||||
@@ -114,12 +114,16 @@ func (c *Config) Cleanup(db db.Db) {
|
||||
tasks[i].UpdateStatus(base, TASK_FAILED_RUNNING, "Task inturupted by server restart please try again")
|
||||
_, err = db.Exec("update models set status=$1 where id=$2", READY_RETRAIN_FAILED, tasks[i].ModelId)
|
||||
failLog(err)
|
||||
_, err = db.Exec("update model_classes set status=$1 where model_id=$2 and status=$3", CLASS_STATUS_TO_TRAIN, tasks[i].ModelId, CLASS_STATUS_TRAINING)
|
||||
failLog(err)
|
||||
continue
|
||||
}
|
||||
if tasks[i].TaskType == int(TASK_TYPE_TRAINING) {
|
||||
tasks[i].UpdateStatus(base, TASK_FAILED_RUNNING, "Task inturupted by server restart please try again")
|
||||
_, err = db.Exec("update models set status=$1 where id=$2", FAILED_TRAINING, tasks[i].ModelId)
|
||||
failLog(err)
|
||||
_, err = db.Exec("update model_classes set status=$1 where model_id=$2 and status=$3", CLASS_STATUS_TO_TRAIN, tasks[i].ModelId, CLASS_STATUS_TRAINING)
|
||||
failLog(err)
|
||||
continue
|
||||
}
|
||||
}
|
||||
|
||||
@@ -374,7 +374,7 @@ func (c Context) JsonBadRequest(dat any) *Error {
|
||||
c.SetReportCaller(true)
|
||||
c.Logger.Warn("Request failed with a bad request", "dat", dat)
|
||||
c.SetReportCaller(false)
|
||||
return c.ErrorCode(nil, 404, dat)
|
||||
return c.SendJSONStatus(http.StatusBadRequest, dat)
|
||||
}
|
||||
|
||||
func (c Context) JsonErrorBadRequest(err error, dat any) *Error {
|
||||
@@ -449,7 +449,7 @@ func (x Handle) createContext(handler *Handle, r *http.Request, w http.ResponseW
|
||||
logger := log.NewWithOptions(os.Stdout, log.Options{
|
||||
ReportCaller: true,
|
||||
ReportTimestamp: true,
|
||||
TimeFormat: time.Kitchen,
|
||||
TimeFormat: time.DateTime,
|
||||
Prefix: r.URL.Path,
|
||||
})
|
||||
|
||||
|
||||
Reference in New Issue
Block a user