func NewPostgresKeeper(id string, cfg config, stop chan bool, end chan error) (*PostgresKeeper, error) {
etcdPath := filepath.Join(common.EtcdBasePath, cfg.clusterName)
e, err := etcdm.NewEtcdManager(cfg.etcdEndpoints, etcdPath, common.DefaultEtcdRequestTimeout)
if err != nil {
return nil, fmt.Errorf("cannot create etcd manager: %v", err)
}
clusterConfig, _, err := e.GetClusterConfig()
if err != nil {
return nil, fmt.Errorf("cannot get cluster config: %v", err)
}
log.Debugf(spew.Sprintf("clusterConfig: %#v", clusterConfig))
p := &PostgresKeeper{id: id,
dataDir: cfg.dataDir,
e: e,
listenAddress: cfg.listenAddress,
port: cfg.port,
pgListenAddress: cfg.pgListenAddress,
pgPort: cfg.pgPort,
clusterConfig: clusterConfig,
stop: stop,
end: end,
}
serverParameters := p.createServerParameters()
pgm, err := postgresql.NewManager(id, cfg.pgBinPath, cfg.dataDir, serverParameters, p.getOurConnString(), p.getOurReplConnString(), clusterConfig.PGReplUser, clusterConfig.PGReplPassword, clusterConfig.RequestTimeout)
if err != nil {
return nil, fmt.Errorf("cannot create postgres manager: %v", err)
}
p.pgm = pgm
return p, nil
}
func NewPostgresKeeper(id string, cfg config, stop chan bool, end chan error) (*PostgresKeeper, error) {
etcdPath := filepath.Join(common.EtcdBasePath, cfg.clusterName)
e, err := etcdm.NewEtcdManager(cfg.etcdEndpoints, etcdPath, common.DefaultEtcdRequestTimeout)
if err != nil {
return nil, fmt.Errorf("cannot create etcd manager: %v", err)
}
cd, _, err := e.GetClusterData()
if err != nil {
return nil, fmt.Errorf("error retrieving cluster data: %v", err)
}
var cv *cluster.ClusterView
if cd == nil {
cv = cluster.NewClusterView()
} else {
cv = cd.ClusterView
}
log.Debugf(spew.Sprintf("clusterView: %#v", cv))
clusterConfig := cv.Config.ToConfig()
log.Debugf(spew.Sprintf("clusterConfig: %#v", clusterConfig))
p := &PostgresKeeper{id: id,
dataDir: cfg.dataDir,
e: e,
listenAddress: cfg.listenAddress,
port: cfg.port,
pgListenAddress: cfg.pgListenAddress,
pgPort: cfg.pgPort,
clusterConfig: clusterConfig,
stop: stop,
end: end,
}
followersIDs := cv.GetFollowersIDs(p.id)
pgParameters := p.createPGParameters(followersIDs)
pgm, err := postgresql.NewManager(id, cfg.pgBinPath, cfg.dataDir, cfg.pgConfDir, pgParameters, p.getOurConnString(), p.getOurReplConnString(), clusterConfig.PGReplUser, clusterConfig.PGReplPassword, clusterConfig.RequestTimeout)
if err != nil {
return nil, fmt.Errorf("cannot create postgres manager: %v", err)
}
p.pgm = pgm
return p, nil
}
func (p *PostgresKeeper) Start() {
endSMCh := make(chan struct{})
endPgStatecheckerCh := make(chan struct{})
endApiCh := make(chan error)
var err error
var cd *cluster.ClusterData
// TODO(sgotti) make the postgres manager stateless and instantiate a
// new one at every check loop, this will avoid the need to loop here
// to get the clusterconfig
for {
cd, _, err = p.e.GetClusterData()
if err == nil {
break
}
log.Errorf("error retrieving cluster data: %v", err)
time.Sleep(cluster.DefaultSleepInterval)
}
var cv *cluster.ClusterView
if cd == nil {
cv = cluster.NewClusterView()
} else {
cv = cd.ClusterView
}
log.Debugf(spew.Sprintf("clusterView: %#v", cv))
p.clusterConfig = cv.Config.ToConfig()
log.Debugf(spew.Sprintf("clusterConfig: %#v", p.clusterConfig))
if err := p.loadCVVersion(); err != nil {
p.end <- fmt.Errorf("failed to load cluster version file: %v", err)
return
}
// TODO(sgotti) reconfigure the various configurations options (PGRepl*
// and RequestTimeout) after a changed cluster config
followersIDs := cv.GetFollowersIDs(p.id)
pgParameters := p.createPGParameters(followersIDs)
pgm := postgresql.NewManager(p.id, cfg.pgBinPath, cfg.dataDir, cfg.pgConfDir, pgParameters, p.getOurConnString(), p.getOurReplConnString(), p.clusterConfig.PGReplUser, p.clusterConfig.PGReplPassword, p.clusterConfig.RequestTimeout)
p.pgm = pgm
p.pgm.Stop(true)
http.HandleFunc("/info", p.infoHandler)
http.HandleFunc("/pgstate", p.pgStateHandler)
go func() {
endApiCh <- http.ListenAndServe(fmt.Sprintf("%s:%s", p.listenAddress, p.port), nil)
}()
ctx, cancel := context.WithCancel(context.Background())
smTimerCh := time.NewTimer(0).C
updatePGStateTimerCh := time.NewTimer(0).C
for true {
select {
case <-p.stop:
log.Debugf("stopping stolon keeper")
cancel()
p.pgm.Stop(true)
p.end <- nil
return
case <-smTimerCh:
go func() {
p.postgresKeeperSM(ctx)
endSMCh <- struct{}{}
}()
case <-endSMCh:
smTimerCh = time.NewTimer(p.clusterConfig.SleepInterval).C
case <-updatePGStateTimerCh:
go func() {
p.updatePGState(ctx)
endPgStatecheckerCh <- struct{}{}
}()
case <-endPgStatecheckerCh:
updatePGStateTimerCh = time.NewTimer(p.clusterConfig.SleepInterval).C
case err := <-endApiCh:
if err != nil {
log.Fatal("ListenAndServe: ", err)
}
close(p.stop)
}
}
}
func (p *PostgresKeeper) Start() {
endSMCh := make(chan struct{})
endPgStatecheckerCh := make(chan struct{})
endUpdateKeeperInfo := make(chan struct{})
var err error
var cd *cluster.ClusterData
cd, _, err = p.e.GetClusterData()
if err != nil {
log.Error("error retrieving cluster data", zap.Error(err))
} else if cd != nil {
if cd.FormatVersion != cluster.CurrentCDFormatVersion {
log.Error("unsupported clusterdata format version", zap.Uint64("version", cd.FormatVersion))
} else if cd.Cluster != nil {
p.sleepInterval = cd.Cluster.Spec.SleepInterval.Duration
p.requestTimeout = cd.Cluster.Spec.RequestTimeout.Duration
}
}
log.Debug("cd dump", zap.String("cd", spew.Sdump(cd)))
// TODO(sgotti) reconfigure the various configurations options
// (RequestTimeout) after a changed cluster config
pgParameters := make(common.Parameters)
pgm := postgresql.NewManager(p.pgBinPath, p.dataDir, pgParameters, p.getLocalConnParams(), p.getOurReplConnParams(), p.pgSUUsername, p.pgSUPassword, p.pgReplUsername, p.pgReplPassword, p.requestTimeout)
p.pgm = pgm
p.pgm.Stop(true)
ctx, cancel := context.WithCancel(context.Background())
smTimerCh := time.NewTimer(0).C
updatePGStateTimerCh := time.NewTimer(0).C
updateKeeperInfoTimerCh := time.NewTimer(0).C
for true {
select {
case <-p.stop:
log.Debug("stopping stolon keeper")
cancel()
p.pgm.Stop(true)
p.end <- nil
return
case <-smTimerCh:
go func() {
p.postgresKeeperSM(ctx)
endSMCh <- struct{}{}
}()
case <-endSMCh:
smTimerCh = time.NewTimer(p.sleepInterval).C
case <-updatePGStateTimerCh:
// updateKeeperInfo two times faster than the sleep interval
go func() {
p.updatePGState(ctx)
endPgStatecheckerCh <- struct{}{}
}()
case <-endPgStatecheckerCh:
// updateKeeperInfo two times faster than the sleep interval
updatePGStateTimerCh = time.NewTimer(p.sleepInterval / 2).C
case <-updateKeeperInfoTimerCh:
go func() {
if err := p.updateKeeperInfo(); err != nil {
log.Error("failed to update keeper info", zap.Error(err))
}
endUpdateKeeperInfo <- struct{}{}
}()
case <-endUpdateKeeperInfo:
updateKeeperInfoTimerCh = time.NewTimer(p.sleepInterval).C
}
}
}