feat: initial watchdog implementation (#1341)

* feat: initial watchdog implementation

Signed-off-by: Ettore Di Giacinto <mudler@users.noreply.github.com>

* fiuxups

* Add more output

* wip: idletime checker

* wire idle watchdog checks

* enlarge watchdog time window

* small fixes

* Use stopmodel

* Always delete process

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

---------

Signed-off-by: Ettore Di Giacinto <mudler@users.noreply.github.com>
Signed-off-by: Ettore Di Giacinto <mudler@localai.io>
This commit is contained in:
Ettore Di Giacinto 2023-11-26 18:36:23 +01:00 committed by GitHub
parent 9482acfdfc
commit 824612f1b4
No known key found for this signature in database
GPG key ID: 4AEE18F83AFDEB23
10 changed files with 341 additions and 13 deletions

View file

@ -121,7 +121,7 @@ func (ml *ModelLoader) grpcModel(backend string, o *Options) func(string, string
// Wait for the service to start up
ready := false
for i := 0; i < o.grpcAttempts; i++ {
if client.GRPC(o.parallelRequests).HealthCheck(context.Background()) {
if client.GRPC(o.parallelRequests, ml.wd).HealthCheck(context.Background()) {
log.Debug().Msgf("GRPC Service Ready")
ready = true
break
@ -140,7 +140,7 @@ func (ml *ModelLoader) grpcModel(backend string, o *Options) func(string, string
log.Debug().Msgf("GRPC: Loading model with options: %+v", options)
res, err := client.GRPC(o.parallelRequests).LoadModel(o.context, &options)
res, err := client.GRPC(o.parallelRequests, ml.wd).LoadModel(o.context, &options)
if err != nil {
return "", fmt.Errorf("could not load model: %w", err)
}
@ -154,11 +154,11 @@ func (ml *ModelLoader) grpcModel(backend string, o *Options) func(string, string
func (ml *ModelLoader) resolveAddress(addr ModelAddress, parallel bool) (*grpc.Client, error) {
if parallel {
return addr.GRPC(parallel), nil
return addr.GRPC(parallel, ml.wd), nil
}
if _, ok := ml.grpcClients[string(addr)]; !ok {
ml.grpcClients[string(addr)] = addr.GRPC(parallel)
ml.grpcClients[string(addr)] = addr.GRPC(parallel, ml.wd)
}
return ml.grpcClients[string(addr)], nil
}

View file

@ -63,12 +63,17 @@ type ModelLoader struct {
models map[string]ModelAddress
grpcProcesses map[string]*process.Process
templates map[TemplateType]map[string]*template.Template
wd *WatchDog
}
type ModelAddress string
func (m ModelAddress) GRPC(parallel bool) *grpc.Client {
return grpc.NewClient(string(m), parallel)
func (m ModelAddress) GRPC(parallel bool, wd *WatchDog) *grpc.Client {
enableWD := false
if wd != nil {
enableWD = true
}
return grpc.NewClient(string(m), parallel, wd, enableWD)
}
func NewModelLoader(modelPath string) *ModelLoader {
@ -79,10 +84,15 @@ func NewModelLoader(modelPath string) *ModelLoader {
templates: make(map[TemplateType]map[string]*template.Template),
grpcProcesses: make(map[string]*process.Process),
}
nml.initializeTemplateMap()
return nml
}
func (ml *ModelLoader) SetWatchDog(wd *WatchDog) {
ml.wd = wd
}
func (ml *ModelLoader) ExistsInModelPath(s string) bool {
return existsInPath(ml.ModelPath, s)
}
@ -139,11 +149,17 @@ func (ml *ModelLoader) LoadModel(modelName string, loader func(string, string) (
func (ml *ModelLoader) ShutdownModel(modelName string) error {
ml.mu.Lock()
defer ml.mu.Unlock()
return ml.StopModel(modelName)
}
func (ml *ModelLoader) StopModel(modelName string) error {
defer ml.deleteProcess(modelName)
if _, ok := ml.models[modelName]; !ok {
return fmt.Errorf("model %s not found", modelName)
}
return ml.deleteProcess(modelName)
return nil
//return ml.deleteProcess(modelName)
}
func (ml *ModelLoader) CheckIsLoaded(s string) ModelAddress {
@ -153,7 +169,7 @@ func (ml *ModelLoader) CheckIsLoaded(s string) ModelAddress {
if c, ok := ml.grpcClients[s]; ok {
client = c
} else {
client = m.GRPC(false)
client = m.GRPC(false, ml.wd)
}
if !client.HealthCheck(context.Background()) {

View file

@ -17,7 +17,7 @@ import (
func (ml *ModelLoader) StopAllExcept(s string) {
ml.StopGRPC(func(id string, p *process.Process) bool {
if id != s {
for ml.models[id].GRPC(false).IsBusy() {
for ml.models[id].GRPC(false, ml.wd).IsBusy() {
log.Debug().Msgf("%s busy. Waiting.", id)
time.Sleep(2 * time.Second)
}
@ -80,6 +80,11 @@ func (ml *ModelLoader) startProcess(grpcProcess, id string, serverAddress string
process.WithEnvironment(os.Environ()...),
)
if ml.wd != nil {
ml.wd.Add(serverAddress, grpcControlProcess)
ml.wd.AddAddressModelMap(serverAddress, id)
}
ml.grpcProcesses[id] = grpcControlProcess
if err := grpcControlProcess.Run(); err != nil {

155
pkg/model/watchdog.go Normal file
View file

@ -0,0 +1,155 @@
package model
import (
"sync"
"time"
process "github.com/mudler/go-processmanager"
"github.com/rs/zerolog/log"
)
// All GRPC Clients created by ModelLoader should have an associated injected
// watchdog that will keep track of the state of each backend (busy or not)
// and for how much time it has been busy.
// If a backend is busy for too long, the watchdog will kill the process and
// force a reload of the model
// The watchdog runs as a separate go routine,
// and the GRPC client talks to it via a channel to send status updates
type WatchDog struct {
sync.Mutex
timetable map[string]time.Time
idleTime map[string]time.Time
timeout, idletimeout time.Duration
addressMap map[string]*process.Process
addressModelMap map[string]string
pm ProcessManager
stop chan bool
busyCheck, idleCheck bool
}
type ProcessManager interface {
StopModel(modelName string) error
}
func NewWatchDog(pm ProcessManager, timeoutBusy, timeoutIdle time.Duration, busy, idle bool) *WatchDog {
return &WatchDog{
timeout: timeoutBusy,
idletimeout: timeoutIdle,
pm: pm,
timetable: make(map[string]time.Time),
idleTime: make(map[string]time.Time),
addressMap: make(map[string]*process.Process),
busyCheck: busy,
idleCheck: idle,
addressModelMap: make(map[string]string),
}
}
func (wd *WatchDog) Shutdown() {
wd.Lock()
defer wd.Unlock()
wd.stop <- true
}
func (wd *WatchDog) AddAddressModelMap(address string, model string) {
wd.Lock()
defer wd.Unlock()
wd.addressModelMap[address] = model
}
func (wd *WatchDog) Add(address string, p *process.Process) {
wd.Lock()
defer wd.Unlock()
wd.addressMap[address] = p
}
func (wd *WatchDog) Mark(address string) {
wd.Lock()
defer wd.Unlock()
wd.timetable[address] = time.Now()
delete(wd.idleTime, address)
}
func (wd *WatchDog) UnMark(ModelAddress string) {
wd.Lock()
defer wd.Unlock()
delete(wd.timetable, ModelAddress)
wd.idleTime[ModelAddress] = time.Now()
}
func (wd *WatchDog) Run() {
log.Info().Msg("[WatchDog] starting watchdog")
for {
select {
case <-wd.stop:
log.Info().Msg("[WatchDog] Stopping watchdog")
return
case <-time.After(30 * time.Second):
if !wd.busyCheck && !wd.idleCheck {
log.Info().Msg("[WatchDog] No checks enabled, stopping watchdog")
return
}
if wd.busyCheck {
wd.checkBusy()
}
if wd.idleCheck {
wd.checkIdle()
}
}
}
}
func (wd *WatchDog) checkIdle() {
wd.Lock()
defer wd.Unlock()
log.Debug().Msg("[WatchDog] Watchdog checks for idle connections")
for address, t := range wd.idleTime {
log.Debug().Msgf("[WatchDog] %s: idle connection", address)
if time.Since(t) > wd.idletimeout {
log.Warn().Msgf("[WatchDog] Address %s is idle for too long, killing it", address)
p, ok := wd.addressModelMap[address]
if ok {
if err := wd.pm.StopModel(p); err != nil {
log.Error().Msgf("[watchdog] Error shutting down model %s: %v", p, err)
}
delete(wd.idleTime, address)
delete(wd.addressModelMap, address)
delete(wd.addressMap, address)
} else {
log.Warn().Msgf("[WatchDog] Address %s unresolvable", address)
delete(wd.idleTime, address)
}
}
}
}
func (wd *WatchDog) checkBusy() {
wd.Lock()
defer wd.Unlock()
log.Debug().Msg("[WatchDog] Watchdog checks for busy connections")
for address, t := range wd.timetable {
log.Debug().Msgf("[WatchDog] %s: active connection", address)
if time.Since(t) > wd.timeout {
model, ok := wd.addressModelMap[address]
if ok {
log.Warn().Msgf("[WatchDog] Model %s is busy for too long, killing it", model)
if err := wd.pm.StopModel(model); err != nil {
log.Error().Msgf("[watchdog] Error shutting down model %s: %v", model, err)
}
delete(wd.timetable, address)
delete(wd.addressModelMap, address)
delete(wd.addressMap, address)
} else {
log.Warn().Msgf("[WatchDog] Address %s unresolvable", address)
delete(wd.timetable, address)
}
}
}
}