introduce new ServiceV2 API to handle guided restarts (#18826)

New API now verifies any hung disks before restart/stop,
provides a 'per node' break down of the restart/stop results.

Provides also how many blocked syscalls are present on the
drives and what users must do about them.

Adds options to do pre-flight checks to provide information
to the user regarding any hung disks. Provides 'force' option
to forcibly attempt a restart() even with waiting syscalls
on the drives.
This commit is contained in:
Harshavardhana 2024-01-19 14:22:36 -08:00 committed by GitHub
parent 83bf15a703
commit ac81f0248c
No known key found for this signature in database
GPG key ID: B5690EEEBB952194
7 changed files with 203 additions and 20 deletions

View file

@ -227,8 +227,6 @@ func (a adminAPIHandlers) ServerUpdateHandler(w http.ResponseWriter, r *http.Req
// ServiceHandler - POST /minio/admin/v3/service?action={action}
// ----------
// Supports following actions:
// - restart (restarts all the MinIO instances in a setup)
// - stop (stops all the MinIO instances in a setup)
// - freeze (freezes all incoming S3 API calls)
// - unfreeze (unfreezes previously frozen S3 API calls)
func (a adminAPIHandlers) ServiceHandler(w http.ResponseWriter, r *http.Request) {
@ -238,7 +236,85 @@ func (a adminAPIHandlers) ServiceHandler(w http.ResponseWriter, r *http.Request)
action := vars["action"]
var serviceSig serviceSignal
switch madmin.ServiceAction(action) {
switch act := madmin.ServiceAction(action); act {
case madmin.ServiceActionFreeze:
serviceSig = serviceFreeze
case madmin.ServiceActionUnfreeze:
serviceSig = serviceUnFreeze
default:
process := act == madmin.ServiceActionRestart || act == madmin.ServiceActionStop
if process {
apiErr := errorCodes.ToAPIErr(ErrMalformedPOSTRequest)
apiErr.Description = "process actions are not supported via this API anymore, please upgrade 'mc' or madmin-go to use ServiceV2() API"
writeErrorResponseJSON(ctx, w, apiErr, r.URL)
return
}
logger.LogIf(ctx, fmt.Errorf("Unrecognized service action %s requested", action), logger.Application)
writeErrorResponseJSON(ctx, w, errorCodes.ToAPIErr(ErrMalformedPOSTRequest), r.URL)
return
}
objectAPI, _ := validateAdminReq(ctx, w, r, policy.ServiceFreezeAdminAction)
if objectAPI == nil {
return
}
// Notify all other MinIO peers signal service.
for _, nerr := range globalNotificationSys.SignalService(serviceSig) {
if nerr.Err != nil {
logger.GetReqInfo(ctx).SetTags("peerAddress", nerr.Host.String())
logger.LogIf(ctx, nerr.Err)
}
}
// Reply to the client before restarting, stopping MinIO server.
writeSuccessResponseHeadersOnly(w)
switch serviceSig {
case serviceFreeze:
freezeServices()
case serviceUnFreeze:
unfreezeServices()
}
}
type servicePeerResult struct {
Host string `json:"host"`
Err string `json:"err,omitempty"`
WaitingDrives map[string]DiskMetrics `json:"waitingDrives,omitempty"`
}
type serviceResult struct {
Action madmin.ServiceAction `json:"action"`
Forced bool `json:"forced"`
DryRun bool `json:"dryRun"`
Results []servicePeerResult `json:"results,omitempty"`
}
// ServiceV2Handler - POST /minio/admin/v3/service?action={action}&type=2
// ----------
// Supports following actions:
// - restart (restarts all the MinIO instances in a setup)
// - stop (stops all the MinIO instances in a setup)
// - freeze (freezes all incoming S3 API calls)
// - unfreeze (unfreezes previously frozen S3 API calls)
//
// This newer API now returns back status per remote peer and local regarding
// if a "restart/stop" was successful or not. Service signal now supports
// a dry-run that helps skip the nodes that may have hung drives. By default
// restart/stop will ignore the servers that are hung on drives. You can use
// 'force' param to force restart even with hung drives if needed.
func (a adminAPIHandlers) ServiceV2Handler(w http.ResponseWriter, r *http.Request) {
ctx := r.Context()
vars := mux.Vars(r)
action := vars["action"]
dryRun := r.Form.Get("dry-run") == "true"
force := r.Form.Get("force") == "true"
var serviceSig serviceSignal
act := madmin.ServiceAction(action)
switch act {
case madmin.ServiceActionRestart:
serviceSig = serviceRestart
case madmin.ServiceActionStop:
@ -267,15 +343,57 @@ func (a adminAPIHandlers) ServiceHandler(w http.ResponseWriter, r *http.Request)
}
// Notify all other MinIO peers signal service.
for _, nerr := range globalNotificationSys.SignalService(serviceSig) {
if nerr.Err != nil {
logger.GetReqInfo(ctx).SetTags("peerAddress", nerr.Host.String())
logger.LogIf(ctx, nerr.Err)
nerrs := globalNotificationSys.SignalServiceV2(serviceSig, dryRun, force)
srvResult := serviceResult{Action: act, Results: make([]servicePeerResult, 0, len(nerrs))}
process := act == madmin.ServiceActionRestart || act == madmin.ServiceActionStop
var trigger bool
if process {
localhost := globalLocalNodeName
if globalLocalNodeName == "" {
localhost = "127.0.0.1"
}
waitingDrives := canWeRestartNode()
srvResult.Results = append(srvResult.Results, servicePeerResult{
Host: localhost,
WaitingDrives: waitingDrives,
})
trigger = len(waitingDrives) == 0 || force
}
for _, nerr := range nerrs {
if nerr.Err != nil && process {
waitingDrives := map[string]DiskMetrics{}
jerr := json.Unmarshal([]byte(nerr.Err.Error()), &waitingDrives)
if jerr == nil {
srvResult.Results = append(srvResult.Results, servicePeerResult{
Host: nerr.Host.String(),
WaitingDrives: waitingDrives,
})
continue
}
}
errStr := ""
if nerr.Err != nil {
errStr = nerr.Err.Error()
}
srvResult.Results = append(srvResult.Results, servicePeerResult{
Host: nerr.Host.String(),
Err: errStr,
})
}
srvResult.Forced = force
srvResult.DryRun = dryRun
buf, err := json.Marshal(srvResult)
if err != nil {
writeErrorResponseJSON(ctx, w, toAdminAPIErr(ctx, err), r.URL)
return
}
// Reply to the client before restarting, stopping MinIO server.
writeSuccessResponseHeadersOnly(w)
writeSuccessResponseJSON(w, buf)
switch serviceSig {
case serviceFreeze:
@ -283,7 +401,9 @@ func (a adminAPIHandlers) ServiceHandler(w http.ResponseWriter, r *http.Request)
case serviceUnFreeze:
unfreezeServices()
case serviceRestart, serviceStop:
globalServiceSignalCh <- serviceSig
if !dryRun && trigger {
globalServiceSignalCh <- serviceSig
}
}
}

View file

@ -168,6 +168,7 @@ func testServiceSignalReceiver(cmd cmdType, t *testing.T) {
func getServiceCmdRequest(cmd cmdType, cred auth.Credentials) (*http.Request, error) {
queryVal := url.Values{}
queryVal.Set("action", string(cmd.toServiceAction()))
queryVal.Set("type", "2")
resource := adminPathPrefix + adminAPIVersionPrefix + "/service?" + queryVal.Encode()
req, err := newTestRequest(http.MethodPost, resource, 0, nil)
if err != nil {
@ -220,12 +221,18 @@ func testServicesCmdHandler(cmd cmdType, t *testing.T) {
rec := httptest.NewRecorder()
adminTestBed.router.ServeHTTP(rec, req)
resp, _ := io.ReadAll(rec.Body)
if rec.Code != http.StatusOK {
resp, _ := io.ReadAll(rec.Body)
t.Errorf("Expected to receive %d status code but received %d. Body (%s)",
http.StatusOK, rec.Code, string(resp))
}
result := &serviceResult{}
if err := json.Unmarshal(resp, result); err != nil {
t.Error(err)
}
_ = result
// Wait until testServiceSignalReceiver() called in a goroutine quits.
wg.Wait()
}

View file

@ -154,8 +154,12 @@ func registerAdminRouter(router *mux.Router, enableConfigOps bool) {
}
for _, adminVersion := range adminVersions {
// Restart and stop MinIO service type=2
adminRouter.Methods(http.MethodPost).Path(adminVersion+"/service").HandlerFunc(adminMiddleware(adminAPI.ServiceV2Handler, traceAllFlag)).Queries("action", "{action:.*}", "type", "2")
// Restart and stop MinIO service.
adminRouter.Methods(http.MethodPost).Path(adminVersion+"/service").HandlerFunc(adminMiddleware(adminAPI.ServiceHandler, traceAllFlag)).Queries("action", "{action:.*}")
// Update MinIO servers.
adminRouter.Methods(http.MethodPost).Path(adminVersion+"/update").HandlerFunc(adminMiddleware(adminAPI.ServerUpdateHandler, traceAllFlag)).Queries("updateURL", "{updateURL:.*}")

View file

@ -106,7 +106,7 @@ func (g *NotificationGroup) Go(ctx context.Context, f func() error, index int, a
if i == g.retryCount-1 {
reqInfo := (&logger.ReqInfo{}).AppendTags("peerAddress", addr.String())
ctx := logger.SetReqInfo(ctx, reqInfo)
logger.LogIf(ctx, err)
logger.LogOnceIf(ctx, err, addr.String())
}
// Wait for a minimum of 100ms and dynamically increase this based on number of attempts.
if i < g.retryCount-1 {
@ -376,7 +376,7 @@ func (sys *NotificationSys) SignalConfigReload(subSys string) []NotificationPeer
}
client := client
ng.Go(GlobalContext, func() error {
return client.SignalService(serviceReloadDynamic, subSys)
return client.SignalService(serviceReloadDynamic, subSys, false, true)
}, idx, *client.host)
}
return ng.Wait()
@ -391,7 +391,23 @@ func (sys *NotificationSys) SignalService(sig serviceSignal) []NotificationPeerE
}
client := client
ng.Go(GlobalContext, func() error {
return client.SignalService(sig, "")
// force == true preserves the current behavior
return client.SignalService(sig, "", false, true)
}, idx, *client.host)
}
return ng.Wait()
}
// SignalServiceV2 - calls signal service RPC call on all peers with v2 API
func (sys *NotificationSys) SignalServiceV2(sig serviceSignal, dryRun, force bool) []NotificationPeerErr {
ng := WithNPeers(len(sys.peerClients))
for idx, client := range sys.peerClients {
if client == nil {
continue
}
client := client
ng.Go(GlobalContext, func() error {
return client.SignalService(sig, "", dryRun, force)
}, idx, *client.host)
}
return ng.Wait()
@ -1275,7 +1291,7 @@ func (sys *NotificationSys) ServiceFreeze(ctx context.Context, freeze bool) []No
}
client := client
ng.Go(GlobalContext, func() error {
return client.SignalService(serviceSig, "")
return client.SignalService(serviceSig, "", false, true)
}, idx, *client.host)
}
nerrs := ng.Wait()

View file

@ -513,9 +513,11 @@ func (client *peerRESTClient) CommitBinary(ctx context.Context) error {
}
// SignalService - sends signal to peer nodes.
func (client *peerRESTClient) SignalService(sig serviceSignal, subSys string) error {
func (client *peerRESTClient) SignalService(sig serviceSignal, subSys string, dryRun, force bool) error {
values := make(url.Values)
values.Set(peerRESTSignal, strconv.Itoa(int(sig)))
values.Set(peerRESTDryRun, strconv.FormatBool(dryRun))
values.Set(peerRESTForce, strconv.FormatBool(force))
values.Set(peerRESTSubSys, subSys)
respBody, err := client.call(peerRESTMethodSignalService, values, nil, -1)
if err != nil {

View file

@ -18,7 +18,7 @@
package cmd
const (
peerRESTVersion = "v34" // Add metrics flag to LocalStorageInfo call
peerRESTVersion = "v35" // Add new service restart behavior
peerRESTVersionPrefix = SlashSeparator + peerRESTVersion
peerRESTPrefix = minioReservedBucketPath + "/peer"
@ -108,6 +108,8 @@ const (
peerRESTDepID = "depID"
peerRESTStartRebalance = "start-rebalance"
peerRESTMetrics = "metrics"
peerRESTDryRun = "dry-run"
peerRESTForce = "force"
peerRESTListenBucket = "bucket"
peerRESTListenPrefix = "prefix"

View file

@ -20,6 +20,7 @@ package cmd
import (
"context"
"encoding/gob"
"encoding/json"
"errors"
"fmt"
"io"
@ -863,6 +864,21 @@ func (s *peerRESTServer) CommitBinaryHandler(w http.ResponseWriter, r *http.Requ
var errUnsupportedSignal = fmt.Errorf("unsupported signal")
func canWeRestartNode() map[string]DiskMetrics {
errs := make([]error, len(globalLocalDrives))
infos := make([]DiskInfo, len(globalLocalDrives))
for i, drive := range globalLocalDrives {
infos[i], errs[i] = drive.DiskInfo(GlobalContext, false)
}
infoMaps := make(map[string]DiskMetrics)
for i := range infos {
if infos[i].Metrics.TotalWaiting >= 1 && errors.Is(errs[i], errFaultyDisk) {
infoMaps[infos[i].Endpoint] = infos[i].Metrics
}
}
return infoMaps
}
// SignalServiceHandler - signal service handler.
func (s *peerRESTServer) SignalServiceHandler(w http.ResponseWriter, r *http.Request) {
if !s.IsValid(w, r) {
@ -883,10 +899,26 @@ func (s *peerRESTServer) SignalServiceHandler(w http.ResponseWriter, r *http.Req
}
signal := serviceSignal(si)
switch signal {
case serviceRestart:
globalServiceSignalCh <- signal
case serviceStop:
globalServiceSignalCh <- signal
case serviceRestart, serviceStop:
dryRun := r.Form.Get("dry-run") == "true" // This is only supported for `restart/stop`
force := r.Form.Get("force") == "true"
waitingDisks := canWeRestartNode()
if len(waitingDisks) > 0 {
buf, err := json.Marshal(waitingDisks)
if err != nil {
s.writeErrorResponse(w, err)
return
}
s.writeErrorResponse(w, errors.New(string(buf)))
// if its forced we signal the process anyway.
if !force {
return
}
}
if !dryRun {
globalServiceSignalCh <- signal
}
case serviceFreeze:
freezeServices()
case serviceUnFreeze: