lotus/node/health.go

118 lines
2.8 KiB
Go
Raw Normal View History

2022-05-21 01:38:17 +00:00
package node
import (
"context"
"net/http"
2022-05-23 19:04:13 +00:00
"sync/atomic"
2022-05-21 01:38:17 +00:00
"time"
2022-05-23 16:29:11 +00:00
logging "github.com/ipfs/go-log/v2"
2022-08-25 18:20:41 +00:00
"github.com/libp2p/go-libp2p/core/network"
2022-06-14 15:00:51 +00:00
lapi "github.com/filecoin-project/lotus/api"
2022-05-21 01:38:17 +00:00
)
2022-05-23 16:29:11 +00:00
var healthlog = logging.Logger("healthcheck")
2022-05-21 01:38:17 +00:00
type HealthHandler struct {
healthy int32
2022-05-21 01:38:17 +00:00
}
func (h *HealthHandler) SetHealthy(healthy bool) {
2022-05-23 19:04:13 +00:00
var hi32 int32
if healthy {
2022-05-23 19:04:13 +00:00
hi32 = 1
}
2022-05-23 19:04:13 +00:00
atomic.StoreInt32(&h.healthy, hi32)
2022-05-21 01:38:17 +00:00
}
func (h *HealthHandler) ServeHTTP(w http.ResponseWriter, r *http.Request) {
if atomic.LoadInt32(&h.healthy) != 1 {
2022-05-21 01:38:17 +00:00
w.WriteHeader(http.StatusServiceUnavailable)
return
}
w.WriteHeader(http.StatusOK)
}
2022-05-23 19:04:13 +00:00
// Check that the node is still working. That is, that it's still processing the chain.
// If there have been no recent changes, consider the node to be dead.
2022-05-21 01:38:17 +00:00
func NewLiveHandler(api lapi.FullNode) *HealthHandler {
ctx := context.Background()
h := HealthHandler{}
go func() {
2022-05-24 17:24:37 +00:00
const (
reset int32 = 5
maxbackoff time.Duration = time.Minute
minbackoff time.Duration = time.Second
)
var (
countdown int32
headCh <-chan []*lapi.HeadChange
2024-05-30 19:45:41 +00:00
backoff = minbackoff
2022-05-24 17:24:37 +00:00
err error
)
2022-05-21 01:38:17 +00:00
minutely := time.NewTicker(time.Minute)
for {
2022-05-24 17:24:37 +00:00
if headCh == nil {
healthlog.Infof("waiting %v before starting ChainNotify channel", backoff)
<-time.After(backoff)
headCh, err = api.ChainNotify(ctx)
if err != nil {
healthlog.Warnf("failed to instantiate ChainNotify channel; cannot determine liveness. %s", err)
h.SetHealthy(false)
nextbackoff := 2 * backoff
if nextbackoff > maxbackoff {
nextbackoff = maxbackoff
}
backoff = nextbackoff
continue
}
2024-05-30 19:45:41 +00:00
healthlog.Infof("started ChainNotify channel")
backoff = minbackoff
2022-05-24 17:24:37 +00:00
}
2022-05-21 01:38:17 +00:00
select {
case <-minutely.C:
2022-05-23 19:04:13 +00:00
atomic.AddInt32(&countdown, -1)
if countdown <= 0 {
2022-05-21 01:38:17 +00:00
h.SetHealthy(false)
}
2022-05-24 17:24:37 +00:00
case _, ok := <-headCh:
if !ok { // channel is closed, enter reconnect loop.
h.SetHealthy(false)
headCh = nil
continue
}
2022-05-23 19:04:13 +00:00
atomic.StoreInt32(&countdown, reset)
2022-05-21 01:38:17 +00:00
h.SetHealthy(true)
}
}
}()
return &h
}
// Check if we are ready to handle traffic.
2022-05-23 19:04:13 +00:00
// 1. sync workers are reasonably up to date.
// 2. libp2p is servicable
2022-05-21 01:38:17 +00:00
func NewReadyHandler(api lapi.FullNode) *HealthHandler {
ctx := context.Background()
h := HealthHandler{}
go func() {
const heightTolerance = uint64(5)
var nethealth, synchealth bool
minutely := time.NewTicker(time.Minute)
for {
select {
case <-minutely.C:
netstat, err := api.NetAutoNatStatus(ctx)
nethealth = err == nil && netstat.Reachability != network.ReachabilityUnknown
nodestat, err := api.NodeStatus(ctx, false)
synchealth = err == nil && nodestat.SyncStatus.Behind < heightTolerance
h.SetHealthy(nethealth && synchealth)
}
}
}()
return &h
}