cronsun/node/node.go

474 lines
8.6 KiB
Go
Raw Normal View History

2017-01-04 09:27:36 +00:00
package node
import (
2017-01-05 12:35:41 +00:00
"fmt"
2017-01-05 02:37:48 +00:00
"os"
2017-01-05 12:35:41 +00:00
"strconv"
"time"
2017-01-04 09:27:36 +00:00
client "github.com/coreos/etcd/clientv3"
2017-01-05 02:37:48 +00:00
2017-05-12 06:48:24 +00:00
"github.com/shunfei/cronsun"
2017-05-12 07:38:50 +00:00
"github.com/shunfei/cronsun/conf"
"github.com/shunfei/cronsun/log"
2017-05-09 10:27:32 +00:00
"github.com/shunfei/cronsun/node/cron"
2017-05-10 09:01:52 +00:00
"github.com/shunfei/cronsun/utils"
2017-01-05 12:35:41 +00:00
)
2017-01-04 09:27:36 +00:00
// Node 执行 cron 命令服务的结构体
type Node struct {
2017-05-12 06:48:24 +00:00
*cronsun.Client
*cronsun.Node
*cron.Cron
jobs Jobs // 和结点相关的任务
groups Groups
2017-05-12 06:48:24 +00:00
cmds map[string]*cronsun.Cmd
link
2017-02-17 13:29:04 +00:00
// 删除的 job id用于 group 更新
delIDs map[string]bool
2017-01-05 02:37:48 +00:00
2017-03-14 07:23:53 +00:00
ttl int64
lID client.LeaseID // lease id
2017-01-06 08:44:42 +00:00
done chan struct{}
2017-01-04 09:27:36 +00:00
}
2017-01-06 08:44:42 +00:00
func NewNode(cfg *conf.Conf) (n *Node, err error) {
2017-05-10 09:01:52 +00:00
ip, err := utils.LocalIP()
2017-01-05 02:37:48 +00:00
if err != nil {
return
}
n = &Node{
2017-05-12 06:48:24 +00:00
Client: cronsun.DefalutClient,
Node: &cronsun.Node{
2017-01-09 10:51:52 +00:00
ID: ip.String(),
PID: strconv.Itoa(os.Getpid()),
},
Cron: cron.New(),
2017-01-06 08:44:42 +00:00
2017-03-14 07:23:53 +00:00
jobs: make(Jobs, 8),
2017-05-12 06:48:24 +00:00
cmds: make(map[string]*cronsun.Cmd),
2017-03-14 07:23:53 +00:00
link: newLink(8),
delIDs: make(map[string]bool, 8),
2017-01-09 10:51:52 +00:00
ttl: cfg.Ttl,
2017-01-06 08:44:42 +00:00
done: make(chan struct{}),
2017-01-05 02:37:48 +00:00
}
return
2017-01-04 09:27:36 +00:00
}
// 注册到 /cronsun/node/xx
2017-01-05 12:35:41 +00:00
func (n *Node) Register() (err error) {
2017-01-09 10:51:52 +00:00
pid, err := n.Node.Exist()
2017-01-05 12:35:41 +00:00
if err != nil {
return
}
if pid != -1 {
2017-01-09 10:51:52 +00:00
return fmt.Errorf("node[%s] pid[%d] exist", n.Node.ID, pid)
2017-01-05 12:35:41 +00:00
}
2017-03-14 07:23:53 +00:00
return n.set()
}
func (n *Node) set() error {
2017-03-20 04:03:54 +00:00
resp, err := n.Client.Grant(n.ttl + 2)
2017-01-05 12:35:41 +00:00
if err != nil {
2017-03-14 07:23:53 +00:00
return err
2017-01-05 12:35:41 +00:00
}
2017-01-09 10:51:52 +00:00
if _, err = n.Node.Put(client.WithLease(resp.ID)); err != nil {
2017-03-14 07:23:53 +00:00
return err
2017-01-05 12:35:41 +00:00
}
2017-01-06 08:44:42 +00:00
2017-03-14 07:23:53 +00:00
n.lID = resp.ID
return nil
}
2017-01-05 12:35:41 +00:00
2017-03-14 07:23:53 +00:00
// 断网掉线重新注册
func (n *Node) keepAlive() {
duration := time.Duration(n.ttl) * time.Second
timer := time.NewTimer(duration)
for {
select {
case <-n.done:
return
case <-timer.C:
if n.lID > 0 {
2017-03-20 04:03:54 +00:00
_, err := n.Client.KeepAliveOnce(n.lID)
2017-03-14 07:23:53 +00:00
if err == nil {
timer.Reset(duration)
continue
}
log.Warnf("%s lid[%x] keepAlive err: %s, try to reset...", n.String(), n.lID, err.Error())
n.lID = 0
}
if err := n.set(); err != nil {
log.Warnf("%s set lid err: %s, try to reset after %d seconds...", n.String(), err.Error(), n.ttl)
} else {
2017-05-12 07:38:50 +00:00
log.Infof("%s set lid[%x] success", n.String(), n.lID)
2017-03-14 07:23:53 +00:00
}
timer.Reset(duration)
}
}
2017-01-04 09:27:36 +00:00
}
func (n *Node) loadJobs() (err error) {
2017-05-12 06:48:24 +00:00
if n.groups, err = cronsun.GetGroups(""); err != nil {
return
}
2017-05-12 06:48:24 +00:00
jobs, err := cronsun.GetJobs()
2017-02-17 13:29:04 +00:00
if err != nil {
return
}
2017-02-17 13:29:04 +00:00
if len(jobs) == 0 {
2017-02-17 10:16:47 +00:00
return
}
2017-02-17 13:29:04 +00:00
for _, job := range jobs {
job.Init(n.ID)
2017-02-17 10:16:47 +00:00
n.addJob(job, false)
}
2017-02-17 13:29:04 +00:00
return
}
2017-05-12 06:48:24 +00:00
func (n *Node) addJob(job *cronsun.Job, notice bool) {
2017-02-23 04:13:14 +00:00
n.link.addJob(job)
if job.IsRunOn(n.ID, n.groups) {
n.jobs[job.ID] = job
}
2017-02-17 13:29:04 +00:00
2017-02-17 10:16:47 +00:00
cmds := job.Cmds(n.ID, n.groups)
if len(cmds) == 0 {
return
}
2017-02-17 10:16:47 +00:00
for _, cmd := range cmds {
n.addCmd(cmd, notice)
}
2017-02-17 10:16:47 +00:00
return
}
func (n *Node) delJob(id string) {
n.delIDs[id] = true
job, ok := n.jobs[id]
// 之前此任务没有在当前结点执行
if !ok {
return
}
2017-02-17 13:29:04 +00:00
delete(n.jobs, id)
2017-02-23 04:13:14 +00:00
n.link.delJob(job)
2017-02-17 13:29:04 +00:00
2017-02-17 10:16:47 +00:00
cmds := job.Cmds(n.ID, n.groups)
if len(cmds) == 0 {
return
}
for _, cmd := range cmds {
n.delCmd(cmd)
}
return
}
2017-05-12 06:48:24 +00:00
func (n *Node) modJob(job *cronsun.Job) {
oJob, ok := n.jobs[job.ID]
// 之前此任务没有在当前结点执行,直接增加任务
if !ok {
n.addJob(job, true)
return
}
2017-02-23 04:13:14 +00:00
n.link.delJob(oJob)
prevCmds := oJob.Cmds(n.ID, n.groups)
job.Count = oJob.Count
*oJob = *job
cmds := oJob.Cmds(n.ID, n.groups)
2017-02-17 10:16:47 +00:00
for id, cmd := range cmds {
2017-06-13 08:00:42 +00:00
n.modCmd(cmd, true)
2017-02-17 10:16:47 +00:00
delete(prevCmds, id)
}
for _, cmd := range prevCmds {
n.delCmd(cmd)
}
2017-02-17 13:29:04 +00:00
2017-02-23 04:13:14 +00:00
n.link.addJob(oJob)
2017-02-17 10:16:47 +00:00
}
2017-05-12 06:48:24 +00:00
func (n *Node) addCmd(cmd *cronsun.Cmd, notice bool) {
2017-06-13 08:00:42 +00:00
n.Cron.Schedule(cmd.JobRule.Schedule, cmd)
n.cmds[cmd.GetID()] = cmd
2017-06-13 08:00:42 +00:00
if notice {
log.Infof("job[%s] group[%s] rule[%s] timer[%s] has added", cmd.Job.ID, cmd.Job.Group, cmd.JobRule.ID, cmd.JobRule.Timer)
}
2017-06-13 08:00:42 +00:00
return
}
2017-06-13 08:00:42 +00:00
func (n *Node) modCmd(cmd *cronsun.Cmd, notice bool) {
c, ok := n.cmds[cmd.GetID()]
if !ok {
2017-06-13 08:00:42 +00:00
n.addCmd(cmd, notice)
return
}
sch := c.JobRule.Timer
*c = *cmd
// 节点执行时间改变,更新 cron
// 否则不用更新 cron
if c.JobRule.Timer != sch {
n.Cron.Schedule(c.JobRule.Schedule, c)
2017-01-19 11:03:18 +00:00
}
2017-02-17 10:16:47 +00:00
if notice {
2017-06-13 08:00:42 +00:00
log.Infof("job[%s] group[%s] rule[%s] timer[%s] has updated", c.Job.ID, c.Job.Group, c.JobRule.ID, c.JobRule.Timer)
2017-02-17 10:16:47 +00:00
}
}
2017-05-12 06:48:24 +00:00
func (n *Node) delCmd(cmd *cronsun.Cmd) {
2017-02-17 10:16:47 +00:00
delete(n.cmds, cmd.GetID())
n.Cron.DelJob(cmd)
2017-06-13 08:00:42 +00:00
log.Infof("job[%s] group[%s] rule[%s] timer[%s] has deleted", cmd.Job.ID, cmd.Job.Group, cmd.JobRule.ID, cmd.JobRule.Timer)
2017-02-17 10:16:47 +00:00
}
2017-05-12 06:48:24 +00:00
func (n *Node) addGroup(g *cronsun.Group) {
2017-02-17 13:29:04 +00:00
n.groups[g.ID] = g
}
func (n *Node) delGroup(id string) {
delete(n.groups, id)
n.link.delGroup(id)
2017-02-17 13:29:04 +00:00
job, ok := n.jobs[id]
// 之前此任务没有在当前结点执行
if !ok {
return
}
2017-01-20 09:42:50 +00:00
2017-02-17 13:29:04 +00:00
cmds := job.Cmds(n.ID, n.groups)
if len(cmds) == 0 {
return
2017-01-20 09:42:50 +00:00
}
2017-02-17 13:29:04 +00:00
for _, cmd := range cmds {
n.delCmd(cmd)
}
return
}
2017-05-12 06:48:24 +00:00
func (n *Node) modGroup(g *cronsun.Group) {
2017-02-17 13:29:04 +00:00
oGroup, ok := n.groups[g.ID]
if !ok {
n.addGroup(g)
2017-01-20 09:42:50 +00:00
return
}
2017-02-17 13:29:04 +00:00
// 都包含/都不包含当前节点,对当前节点任务无影响
if (oGroup.Included(n.ID) && g.Included(n.ID)) || (!oGroup.Included(n.ID) && !g.Included(n.ID)) {
*oGroup = *g
return
}
// 增加当前节点
if !oGroup.Included(n.ID) && g.Included(n.ID) {
2017-02-23 07:59:21 +00:00
n.groupAddNode(g)
return
}
// 移除当前节点
n.groupRmNode(g, oGroup)
return
}
2017-05-12 06:48:24 +00:00
func (n *Node) groupAddNode(g *cronsun.Group) {
2017-02-23 07:59:21 +00:00
n.groups[g.ID] = g
jls := n.link[g.ID]
if len(jls) == 0 {
return
}
2017-02-17 13:29:04 +00:00
2017-02-23 07:59:21 +00:00
var err error
for jid, jl := range jls {
job, ok := n.jobs[jid]
if !ok {
// job 已删除
if n.delIDs[jid] {
n.link.delGroupJob(g.ID, jid)
continue
2017-02-17 13:29:04 +00:00
}
2017-05-12 06:48:24 +00:00
if job, err = cronsun.GetJob(jl.gname, jid); err != nil {
2017-02-23 07:59:21 +00:00
log.Warnf("get job[%s][%s] err: %s", jl.gname, jid, err.Error())
n.link.delGroupJob(g.ID, jid)
continue
2017-02-17 13:29:04 +00:00
}
job.Init(n.ID)
2017-02-17 13:29:04 +00:00
}
2017-02-23 07:59:21 +00:00
cmds := job.Cmds(n.ID, n.groups)
for _, cmd := range cmds {
n.addCmd(cmd, true)
}
2017-02-17 13:29:04 +00:00
}
2017-02-23 07:59:21 +00:00
return
}
2017-02-17 13:29:04 +00:00
2017-05-12 06:48:24 +00:00
func (n *Node) groupRmNode(g, og *cronsun.Group) {
jls := n.link[g.ID]
if len(jls) == 0 {
2017-02-17 13:29:04 +00:00
n.groups[g.ID] = g
return
}
for jid, _ := range jls {
2017-02-17 13:29:04 +00:00
job, ok := n.jobs[jid]
// 之前此任务没有在当前结点执行
2017-02-17 13:29:04 +00:00
if !ok {
2017-02-23 04:13:14 +00:00
n.link.delGroupJob(g.ID, jid)
2017-02-17 13:29:04 +00:00
continue
}
2017-02-23 07:59:21 +00:00
n.groups[og.ID] = og
2017-02-17 13:29:04 +00:00
prevCmds := job.Cmds(n.ID, n.groups)
n.groups[g.ID] = g
cmds := job.Cmds(n.ID, n.groups)
for id, cmd := range cmds {
n.addCmd(cmd, true)
delete(prevCmds, id)
}
for _, cmd := range prevCmds {
n.delCmd(cmd)
}
}
n.groups[g.ID] = g
}
2017-01-19 11:03:18 +00:00
func (n *Node) watchJobs() {
2017-05-12 06:48:24 +00:00
rch := cronsun.WatchJobs()
2017-01-19 11:03:18 +00:00
for wresp := range rch {
for _, ev := range wresp.Events {
switch {
case ev.IsCreate():
2017-05-12 06:48:24 +00:00
job, err := cronsun.GetJobFromKv(ev.Kv)
2017-01-19 11:03:18 +00:00
if err != nil {
2017-03-03 08:00:01 +00:00
log.Warnf("err: %s, kv: %s", err.Error(), ev.Kv.String())
2017-01-19 11:03:18 +00:00
continue
}
job.Init(n.ID)
2017-02-17 10:16:47 +00:00
n.addJob(job, true)
2017-01-19 11:03:18 +00:00
case ev.IsModify():
2017-05-12 06:48:24 +00:00
job, err := cronsun.GetJobFromKv(ev.Kv)
2017-01-19 11:03:18 +00:00
if err != nil {
2017-03-03 08:00:01 +00:00
log.Warnf("err: %s, kv: %s", err.Error(), ev.Kv.String())
2017-01-19 11:03:18 +00:00
continue
}
job.Init(n.ID)
n.modJob(job)
2017-01-19 11:03:18 +00:00
case ev.Type == client.EventTypeDelete:
2017-05-12 06:48:24 +00:00
n.delJob(cronsun.GetIDFromKey(string(ev.Kv.Key)))
2017-01-19 11:03:18 +00:00
default:
log.Warnf("unknown event type[%v] from job[%s]", ev.Type, string(ev.Kv.Key))
}
}
}
}
func (n *Node) watchGroups() {
2017-05-12 06:48:24 +00:00
rch := cronsun.WatchGroups()
for wresp := range rch {
for _, ev := range wresp.Events {
switch {
case ev.IsCreate():
2017-05-12 06:48:24 +00:00
g, err := cronsun.GetGroupFromKv(ev.Kv)
if err != nil {
2017-03-03 08:00:01 +00:00
log.Warnf("err: %s, kv: %s", err.Error(), ev.Kv.String())
continue
}
n.addGroup(g)
case ev.IsModify():
2017-05-12 06:48:24 +00:00
g, err := cronsun.GetGroupFromKv(ev.Kv)
2017-01-20 09:42:50 +00:00
if err != nil {
2017-03-03 08:00:01 +00:00
log.Warnf("err: %s, kv: %s", err.Error(), ev.Kv.String())
2017-01-20 09:42:50 +00:00
continue
}
2017-02-17 13:29:04 +00:00
n.modGroup(g)
case ev.Type == client.EventTypeDelete:
2017-05-12 06:48:24 +00:00
n.delGroup(cronsun.GetIDFromKey(string(ev.Kv.Key)))
default:
log.Warnf("unknown event type[%v] from group[%s]", ev.Type, string(ev.Kv.Key))
}
}
}
}
2017-01-19 11:03:18 +00:00
2017-03-09 09:43:08 +00:00
func (n *Node) watchOnce() {
2017-05-12 06:48:24 +00:00
rch := cronsun.WatchOnce()
2017-03-09 09:43:08 +00:00
for wresp := range rch {
for _, ev := range wresp.Events {
switch {
case ev.IsCreate(), ev.IsModify():
if len(ev.Kv.Value) != 0 && string(ev.Kv.Value) != n.ID {
continue
}
2017-05-12 06:48:24 +00:00
job, ok := n.jobs[cronsun.GetIDFromKey(string(ev.Kv.Key))]
2017-03-10 08:36:44 +00:00
if !ok || !job.IsRunOn(n.ID, n.groups) {
2017-03-09 09:43:08 +00:00
continue
}
go job.RunWithRecovery()
}
}
}
}
2017-01-05 02:37:48 +00:00
// 启动服务
func (n *Node) Run() (err error) {
2017-01-06 08:44:42 +00:00
go n.keepAlive()
defer func() {
if err != nil {
n.Stop(nil)
}
}()
if err = n.loadJobs(); err != nil {
return
}
n.Cron.Start()
2017-02-17 10:16:47 +00:00
go n.watchJobs()
2017-02-17 13:29:04 +00:00
go n.watchGroups()
2017-03-10 08:36:44 +00:00
go n.watchOnce()
n.Node.On()
return
2017-01-06 08:44:42 +00:00
}
2017-01-18 08:27:35 +00:00
// 停止服务
2017-01-05 02:37:48 +00:00
func (n *Node) Stop(i interface{}) {
n.Node.Down()
2017-01-06 08:44:42 +00:00
close(n.done)
2017-01-09 10:51:52 +00:00
n.Node.Del()
2017-01-05 02:37:48 +00:00
n.Client.Close()
n.Cron.Stop()
2017-01-05 02:37:48 +00:00
}