func GetActionSeqList(zkConn zkhelper.Conn, productName string) ([]int, error) {
nodes, _, err := zkConn.Children(GetWatchActionPath(productName))
if err != nil {
return nil, errors.Trace(err)
}
return ExtraSeqList(nodes)
}
func ServerGroups(zkConn zkhelper.Conn, productName string) ([]*ServerGroup, error) {
var ret []*ServerGroup
root := fmt.Sprintf("/zk/codis/db_%s/servers", productName)
groups, _, err := zkConn.Children(root)
if err != nil {
return nil, errors.Trace(err)
}
// Buggy :X
//zkhelper.ChildrenRecursive(*zkConn, root)
for _, group := range groups {
// parse group_1 => 1
groupId, err := strconv.Atoi(strings.Split(group, "_")[1])
if err != nil {
return nil, errors.Trace(err)
}
g, err := GetGroup(zkConn, productName, groupId)
if err != nil {
return nil, errors.Trace(err)
}
ret = append(ret, g)
}
return ret, nil
}
func WaitForReceiverWithTimeout(zkConn zkhelper.Conn, productName string, actionZkPath string, proxies []ProxyInfo, timeoutInMs int) error {
if len(proxies) == 0 {
return nil
}
times := 0
proxyIds := make(map[string]struct{})
var offlineProxyIds []string
for _, p := range proxies {
proxyIds[p.Id] = struct{}{}
}
checkTimes := timeoutInMs / 500
// check every 500ms
for times < checkTimes {
if times >= 6 && (times*500)%1000 == 0 {
log.Warnf("abnormal waiting time for receivers: %s %v", actionZkPath, offlineProxyIds)
}
// get confirm ids
nodes, _, err := zkConn.Children(actionZkPath)
if err != nil {
return errors.Trace(err)
}
confirmIds := make(map[string]struct{})
for _, node := range nodes {
id := path.Base(node)
confirmIds[id] = struct{}{}
}
if len(confirmIds) != 0 {
match := true
// check if all proxy have responsed
var notMatchList []string
for id, _ := range proxyIds {
// if proxy id not in confirm ids, means someone didn't response
if _, ok := confirmIds[id]; !ok {
match = false
notMatchList = append(notMatchList, id)
}
}
if match {
return nil
}
offlineProxyIds = notMatchList
}
times += 1
time.Sleep(500 * time.Millisecond)
}
if len(offlineProxyIds) > 0 {
log.Errorf("proxies didn't responed: %v", offlineProxyIds)
}
// set offline proxies
for _, id := range offlineProxyIds {
log.Errorf("mark proxy %s to PROXY_STATE_MARK_OFFLINE", id)
if err := SetProxyStatus(zkConn, productName, id, PROXY_STATE_MARK_OFFLINE); err != nil {
return errors.Trace(err)
}
}
return errors.Trace(ErrReceiverTimeout)
}
func GetFenceProxyMap(zkConn zkhelper.Conn, productName string) (map[string]bool, error) {
children, _, err := zkConn.Children(GetProxyFencePath(productName))
if err != nil {
if err.Error() == zk.ErrNoNode.Error() {
return make(map[string]bool), nil
} else {
return nil, err
}
}
m := make(map[string]bool, len(children))
for _, fenceNode := range children {
m[fenceNode] = true
}
return m, nil
}
func (self *ServerGroup) GetServers(zkConn zkhelper.Conn) ([]*Server, error) {
var ret []*Server
root := fmt.Sprintf("/zk/codis/db_%s/servers/group_%d", self.ProductName, self.Id)
nodes, _, err := zkConn.Children(root)
if err != nil {
return nil, errors.Trace(err)
}
for _, node := range nodes {
nodePath := root + "/" + node
s, err := GetServer(zkConn, nodePath)
if err != nil {
return nil, errors.Trace(err)
}
ret = append(ret, s)
}
return ret, nil
}
func ForceRemoveLock(zkConn zkhelper.Conn, productName string) error {
lockPath := fmt.Sprintf("/zk/codis/db_%s/LOCK", productName)
children, _, err := zkConn.Children(lockPath)
if err != nil && !zkhelper.ZkErrorEqual(err, zk.ErrNoNode) {
return errors.Trace(err)
}
for _, c := range children {
fullPath := path.Join(lockPath, c)
log.Info("deleting..", fullPath)
err := zkConn.Delete(fullPath, 0)
if err != nil {
return errors.Trace(err)
}
}
return nil
}
func WaitForReceiverWithTimeout(zkConn zkhelper.Conn, productName string, actionZkPath string, proxies []ProxyInfo, timeoutInMs int) error {
if len(proxies) == 0 {
return nil
}
times := 0
proxyIds := make(map[string]bool)
for _, p := range proxies {
proxyIds[p.Id] = true
}
// check every 500ms
for times < timeoutInMs/500 {
if times >= 6 && (times*500)%1000 == 0 {
log.Warnf("abnormal waiting time for receivers: %s %v", actionZkPath, proxyIds)
}
// get confirm ids
nodes, _, err := zkConn.Children(actionZkPath)
if err != nil {
return errors.Trace(err)
}
for _, node := range nodes {
id := path.Base(node)
delete(proxyIds, id)
}
if len(proxyIds) == 0 {
return nil
}
times++
time.Sleep(500 * time.Millisecond)
}
log.Warn("proxies didn't responed: ", proxyIds)
// set offline proxies
/*
for id, _ := range proxyIds {
log.Errorf("mark proxy %s to PROXY_STATE_MARK_OFFLINE", id)
if err := SetProxyStatus(zkConn, productName, id, PROXY_STATE_MARK_OFFLINE); err != nil {
return errors.Trace(err)
}
}
*/
return ErrReceiverTimeout
}
func ProxyList(zkConn zkhelper.Conn, productName string, filter func(*ProxyInfo) bool) ([]ProxyInfo, error) {
ret := make([]ProxyInfo, 0)
root := GetProxyPath(productName)
proxies, _, err := zkConn.Children(root)
if err != nil && !zkhelper.ZkErrorEqual(err, zk.ErrNoNode) {
return nil, errors.Trace(err)
}
for _, proxyName := range proxies {
pi, err := GetProxyInfo(zkConn, productName, proxyName)
if err != nil {
return nil, errors.Trace(err)
}
if filter == nil || filter(pi) == true {
ret = append(ret, *pi)
}
}
return ret, nil
}
func Slots(zkConn zkhelper.Conn, productName string) ([]*Slot, error) {
zkPath := GetSlotBasePath(productName)
children, _, err := zkConn.Children(zkPath)
if err != nil {
return nil, errors.Trace(err)
}
var slots []*Slot
for _, p := range children {
data, _, err := zkConn.Get(path.Join(zkPath, p))
if err != nil {
return nil, errors.Trace(err)
}
slot := &Slot{}
if err := json.Unmarshal(data, &slot); err != nil {
return nil, errors.Trace(err)
}
slots = append(slots, slot)
}
return slots, nil
}
func ActionGC(zkConn zkhelper.Conn, productName string, gcType int, keep int) error {
prefix := GetWatchActionPath(productName)
respPrefix := GetActionResponsePath(productName)
exists, err := zkhelper.NodeExists(zkConn, prefix)
if err != nil {
return errors.Trace(err)
}
if !exists {
// if action path not exists just return nil
return nil
}
actions, _, err := zkConn.Children(prefix)
if err != nil {
return errors.Trace(err)
}
var act Action
currentTs := time.Now().Unix()
if gcType == GC_TYPE_N {
sort.Strings(actions)
// keep last 500 actions
if len(actions)-500 <= keep {
return nil
}
for _, action := range actions[:len(actions)-keep-500] {
if err := zkhelper.DeleteRecursive(zkConn, path.Join(prefix, action), -1); err != nil {
return errors.Trace(err)
}
err := zkhelper.DeleteRecursive(zkConn, path.Join(respPrefix, action), -1)
if err != nil && !zkhelper.ZkErrorEqual(err, zk.ErrNoNode) {
return errors.Trace(err)
}
}
} else if gcType == GC_TYPE_SEC {
secs := keep
for _, action := range actions {
b, _, err := zkConn.Get(path.Join(prefix, action))
if err != nil {
return errors.Trace(err)
}
if err := json.Unmarshal(b, &act); err != nil {
return errors.Trace(err)
}
log.Infof("action = %s, timestamp = %s", action, act.Ts)
ts, _ := strconv.ParseInt(act.Ts, 10, 64)
if currentTs-ts > int64(secs) {
if err := zkhelper.DeleteRecursive(zkConn, path.Join(prefix, action), -1); err != nil {
return errors.Trace(err)
}
err := zkhelper.DeleteRecursive(zkConn, path.Join(respPrefix, action), -1)
if err != nil && !zkhelper.ZkErrorEqual(err, zk.ErrNoNode) {
return errors.Trace(err)
}
}
}
actionResps, _, err := zkConn.Children(respPrefix)
if err != nil {
return errors.Trace(err)
}
for _, action := range actionResps {
b, _, err := zkConn.Get(path.Join(respPrefix, action))
if err != nil {
return errors.Trace(err)
}
if err := json.Unmarshal(b, &act); err != nil {
return errors.Trace(err)
}
log.Infof("action = %s, timestamp = %s", action, act.Ts)
ts, _ := strconv.ParseInt(act.Ts, 10, 64)
if currentTs-ts > int64(secs) {
if err := zkhelper.DeleteRecursive(zkConn, path.Join(respPrefix, action), -1); err != nil {
return errors.Trace(err)
}
}
}
}
return nil
}