// EnsureCheck is used to create a check or updates it's state
func (s *StateStore) EnsureCheck(index uint64, check *structs.HealthCheck) error {
// Ensure we have a status
if check.Status == "" {
check.Status = structs.HealthUnknown
}
// Start the txn
tx, err := s.tables.StartTxn(false)
if err != nil {
panic(fmt.Errorf("Failed to start txn: %v", err))
}
defer tx.Abort()
// Ensure the node exists
res, err := s.nodeTable.GetTxn(tx, "id", check.Node)
if err != nil {
return err
}
if len(res) == 0 {
return fmt.Errorf("Missing node registration")
}
// Ensure the service exists if specified
if check.ServiceID != "" {
res, err = s.serviceTable.GetTxn(tx, "id", check.Node, check.ServiceID)
if err != nil {
return err
}
if len(res) == 0 {
return fmt.Errorf("Missing service registration")
}
// Ensure we set the correct service
srv := res[0].(*structs.ServiceNode)
check.ServiceName = srv.ServiceName
}
// Invalidate any sessions if status is critical
if check.Status == structs.HealthCritical {
err := s.invalidateCheck(index, tx, check.Node, check.CheckID)
if err != nil {
return err
}
}
// Ensure the check is set
if err := s.checkTable.InsertTxn(tx, check); err != nil {
return err
}
if err := s.checkTable.SetLastIndexTxn(tx, index); err != nil {
return err
}
tx.Defer(func() { s.watch[s.checkTable].Notify() })
return tx.Commit()
}
// AddCheck is used to add a health check to the agent.
// This entry is persistent and the agent will make a best effort to
// ensure it is registered. The Check may include a CheckType which
// is used to automatically update the check status
func (a *Agent) AddCheck(check *structs.HealthCheck, chkType *CheckType, persist bool, token string) error {
if check.CheckID == "" {
return fmt.Errorf("CheckID missing")
}
if chkType != nil && !chkType.Valid() {
return fmt.Errorf("Check type is not valid")
}
if check.ServiceID != "" {
svc, ok := a.state.Services()[check.ServiceID]
if !ok {
return fmt.Errorf("ServiceID %q does not exist", check.ServiceID)
}
check.ServiceName = svc.Service
}
a.checkLock.Lock()
defer a.checkLock.Unlock()
// Check if already registered
if chkType != nil {
if chkType.IsTTL() {
if existing, ok := a.checkTTLs[check.CheckID]; ok {
existing.Stop()
}
ttl := &CheckTTL{
Notify: &a.state,
CheckID: check.CheckID,
TTL: chkType.TTL,
Logger: a.logger,
}
// Restore persisted state, if any
if err := a.loadCheckState(check); err != nil {
a.logger.Printf("[WARN] agent: failed restoring state for check %q: %s",
check.CheckID, err)
}
ttl.Start()
a.checkTTLs[check.CheckID] = ttl
} else if chkType.IsHTTP() {
if existing, ok := a.checkHTTPs[check.CheckID]; ok {
existing.Stop()
}
if chkType.Interval < MinInterval {
a.logger.Println(fmt.Sprintf("[WARN] agent: check '%s' has interval below minimum of %v",
check.CheckID, MinInterval))
chkType.Interval = MinInterval
}
http := &CheckHTTP{
Notify: &a.state,
CheckID: check.CheckID,
HTTP: chkType.HTTP,
Interval: chkType.Interval,
Timeout: chkType.Timeout,
Logger: a.logger,
}
http.Start()
a.checkHTTPs[check.CheckID] = http
} else {
if existing, ok := a.checkMonitors[check.CheckID]; ok {
existing.Stop()
}
if chkType.Interval < MinInterval {
a.logger.Println(fmt.Sprintf("[WARN] agent: check '%s' has interval below minimum of %v",
check.CheckID, MinInterval))
chkType.Interval = MinInterval
}
monitor := &CheckMonitor{
Notify: &a.state,
CheckID: check.CheckID,
Script: chkType.Script,
Interval: chkType.Interval,
Logger: a.logger,
}
monitor.Start()
a.checkMonitors[check.CheckID] = monitor
}
}
// Add to the local state for anti-entropy
a.state.AddCheck(check, token)
// Persist the check
if persist {
return a.persistCheck(check, chkType)
}
return nil
}
// AddCheck is used to add a health check to the agent.
// This entry is persistent and the agent will make a best effort to
// ensure it is registered. The Check may include a CheckType which
// is used to automatically update the check status
func (a *Agent) AddCheck(check *structs.HealthCheck, chkType *CheckType, persist bool, token string) error {
if check.CheckID == "" {
return fmt.Errorf("CheckID missing")
}
if chkType != nil && !chkType.Valid() {
return fmt.Errorf("Check type is not valid")
}
if check.ServiceID != "" {
svc, ok := a.state.Services()[check.ServiceID]
if !ok {
return fmt.Errorf("ServiceID %q does not exist", check.ServiceID)
}
check.ServiceName = svc.Service
}
a.checkLock.Lock()
defer a.checkLock.Unlock()
// Check if already registered
if chkType != nil {
if chkType.IsTTL() {
if existing, ok := a.checkTTLs[check.CheckID]; ok {
existing.Stop()
}
ttl := &CheckTTL{
Notify: &a.state,
CheckID: check.CheckID,
TTL: chkType.TTL,
Logger: a.logger,
}
// Restore persisted state, if any
if err := a.loadCheckState(check); err != nil {
a.logger.Printf("[WARN] agent: failed restoring state for check %q: %s",
check.CheckID, err)
}
ttl.Start()
a.checkTTLs[check.CheckID] = ttl
} else if chkType.IsHTTP() {
if existing, ok := a.checkHTTPs[check.CheckID]; ok {
existing.Stop()
}
if chkType.Interval < MinInterval {
a.logger.Println(fmt.Sprintf("[WARN] agent: check '%s' has interval below minimum of %v",
check.CheckID, MinInterval))
chkType.Interval = MinInterval
}
http := &CheckHTTP{
Notify: &a.state,
CheckID: check.CheckID,
HTTP: chkType.HTTP,
Interval: chkType.Interval,
Timeout: chkType.Timeout,
Logger: a.logger,
}
http.Start()
a.checkHTTPs[check.CheckID] = http
} else if chkType.IsTCP() {
if existing, ok := a.checkTCPs[check.CheckID]; ok {
existing.Stop()
}
if chkType.Interval < MinInterval {
a.logger.Println(fmt.Sprintf("[WARN] agent: check '%s' has interval below minimum of %v",
check.CheckID, MinInterval))
chkType.Interval = MinInterval
}
tcp := &CheckTCP{
Notify: &a.state,
CheckID: check.CheckID,
TCP: chkType.TCP,
Interval: chkType.Interval,
Timeout: chkType.Timeout,
Logger: a.logger,
}
tcp.Start()
a.checkTCPs[check.CheckID] = tcp
} else if chkType.IsDocker() {
if existing, ok := a.checkDockers[check.CheckID]; ok {
existing.Stop()
}
if chkType.Interval < MinInterval {
a.logger.Println(fmt.Sprintf("[WARN] agent: check '%s' has interval below minimum of %v",
check.CheckID, MinInterval))
chkType.Interval = MinInterval
}
dockerCheck := &CheckDocker{
Notify: &a.state,
CheckID: check.CheckID,
DockerContainerID: chkType.DockerContainerID,
Shell: chkType.Shell,
Script: chkType.Script,
Interval: chkType.Interval,
Logger: a.logger,
}
if err := dockerCheck.Init(); err != nil {
return err
}
dockerCheck.Start()
a.checkDockers[check.CheckID] = dockerCheck
} else if chkType.IsMonitor() {
if existing, ok := a.checkMonitors[check.CheckID]; ok {
existing.Stop()
}
if chkType.Interval < MinInterval {
a.logger.Println(fmt.Sprintf("[WARN] agent: check '%s' has interval below minimum of %v",
check.CheckID, MinInterval))
chkType.Interval = MinInterval
}
monitor := &CheckMonitor{
Notify: &a.state,
CheckID: check.CheckID,
Script: chkType.Script,
Interval: chkType.Interval,
Timeout: chkType.Timeout,
Logger: a.logger,
ReapLock: &a.reapLock,
}
monitor.Start()
a.checkMonitors[check.CheckID] = monitor
} else {
return fmt.Errorf("Check type is not valid")
}
if chkType.DeregisterCriticalServiceAfter > 0 {
timeout := chkType.DeregisterCriticalServiceAfter
if timeout < a.config.CheckDeregisterIntervalMin {
timeout = a.config.CheckDeregisterIntervalMin
a.logger.Println(fmt.Sprintf("[WARN] agent: check '%s' has deregister interval below minimum of %v",
check.CheckID, a.config.CheckDeregisterIntervalMin))
}
a.checkReapAfter[check.CheckID] = timeout
} else {
delete(a.checkReapAfter, check.CheckID)
}
}
// Add to the local state for anti-entropy
a.state.AddCheck(check, token)
// Persist the check
if persist && !a.config.DevMode {
return a.persistCheck(check, chkType)
}
return nil
}
// ensureCheckTransaction is used as the inner method to handle inserting
// a health check into the state store. It ensures safety against inserting
// checks with no matching node or service.
func (s *StateStore) ensureCheckTxn(tx *memdb.Txn, idx uint64, watches *DumbWatchManager,
hc *structs.HealthCheck) error {
// Check if we have an existing health check
existing, err := tx.First("checks", "id", hc.Node, string(hc.CheckID))
if err != nil {
return fmt.Errorf("failed health check lookup: %s", err)
}
// Set the indexes
if existing != nil {
hc.CreateIndex = existing.(*structs.HealthCheck).CreateIndex
hc.ModifyIndex = idx
} else {
hc.CreateIndex = idx
hc.ModifyIndex = idx
}
// Use the default check status if none was provided
if hc.Status == "" {
hc.Status = structs.HealthCritical
}
// Get the node
node, err := tx.First("nodes", "id", hc.Node)
if err != nil {
return fmt.Errorf("failed node lookup: %s", err)
}
if node == nil {
return ErrMissingNode
}
// If the check is associated with a service, check that we have
// a registration for the service.
if hc.ServiceID != "" {
service, err := tx.First("services", "id", hc.Node, hc.ServiceID)
if err != nil {
return fmt.Errorf("failed service lookup: %s", err)
}
if service == nil {
return ErrMissingService
}
// Copy in the service name
hc.ServiceName = service.(*structs.ServiceNode).ServiceName
}
// Delete any sessions for this check if the health is critical.
if hc.Status == structs.HealthCritical {
mappings, err := tx.Get("session_checks", "node_check", hc.Node, string(hc.CheckID))
if err != nil {
return fmt.Errorf("failed session checks lookup: %s", err)
}
var ids []string
for mapping := mappings.Next(); mapping != nil; mapping = mappings.Next() {
ids = append(ids, mapping.(*sessionCheck).Session)
}
// Delete the session in a separate loop so we don't trash the
// iterator.
watches := NewDumbWatchManager(s.tableWatches)
for _, id := range ids {
if err := s.deleteSessionTxn(tx, idx, watches, id); err != nil {
return fmt.Errorf("failed deleting session: %s", err)
}
}
tx.Defer(func() { watches.Notify() })
}
// Persist the check registration in the db.
if err := tx.Insert("checks", hc); err != nil {
return fmt.Errorf("failed inserting check: %s", err)
}
if err := tx.Insert("index", &IndexEntry{"checks", idx}); err != nil {
return fmt.Errorf("failed updating index: %s", err)
}
watches.Arm("checks")
return nil
}