diff --git a/api/agent.go b/api/agent.go index e56a18dcd2..2b950d0a3e 100644 --- a/api/agent.go +++ b/api/agent.go @@ -68,6 +68,7 @@ type AgentServiceCheck struct { Timeout string `json:",omitempty"` TTL string `json:",omitempty"` HTTP string `json:",omitempty"` + TCP string `json:",omitempty"` Status string `json:",omitempty"` } type AgentServiceChecks []*AgentServiceCheck diff --git a/command/agent/agent.go b/command/agent/agent.go index 449b03c476..397f616a43 100644 --- a/command/agent/agent.go +++ b/command/agent/agent.go @@ -75,6 +75,9 @@ type Agent struct { // checkHTTPs maps the check ID to an associated HTTP check checkHTTPs map[string]*CheckHTTP + // checkTCPs maps the check ID to an associated TCP check + checkTCPs map[string]*CheckTCP + // checkTTLs maps the check ID to an associated check TTL checkTTLs map[string]*CheckTTL @@ -145,6 +148,7 @@ func Create(config *Config, logOutput io.Writer) (*Agent, error) { checkMonitors: make(map[string]*CheckMonitor), checkTTLs: make(map[string]*CheckTTL), checkHTTPs: make(map[string]*CheckHTTP), + checkTCPs: make(map[string]*CheckTCP), eventCh: make(chan serf.UserEvent, 1024), eventBuf: make([]*UserEvent, 256), shutdownCh: make(chan struct{}), @@ -440,6 +444,10 @@ func (a *Agent) Shutdown() error { chk.Stop() } + for _, chk := range a.checkTCPs { + chk.Stop() + } + a.logger.Println("[INFO] agent: requesting shutdown") var err error if a.server != nil { @@ -801,6 +809,27 @@ func (a *Agent) AddCheck(check *structs.HealthCheck, chkType *CheckType, persist 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 existing, ok := a.checkMonitors[check.CheckID]; ok { existing.Stop() @@ -857,6 +886,10 @@ func (a *Agent) RemoveCheck(checkID string, persist bool) error { check.Stop() delete(a.checkHTTPs, checkID) } + if check, ok := a.checkTCPs[checkID]; ok { + check.Stop() + delete(a.checkTCPs, checkID) + } if check, ok := a.checkTTLs[checkID]; ok { check.Stop() delete(a.checkTTLs, checkID) diff --git a/command/agent/check.go b/command/agent/check.go index 6677483886..97ac592eb4 100644 --- a/command/agent/check.go +++ b/command/agent/check.go @@ -4,6 +4,7 @@ import ( "fmt" "io/ioutil" "log" + "net" "net/http" "os/exec" "sync" @@ -31,13 +32,14 @@ const ( // CheckType is used to create either the CheckMonitor // or the CheckTTL. -// Three types are supported: Script, HTTP, and TTL -// Script and HTTP both require Interval +// Four types are supported: Script, HTTP, TCP and TTL +// Script, HTTP and TCP all require Interval // Only one of the types needs to be provided -// TTL or Script/Interval or HTTP/Interval +// TTL or Script/Interval or HTTP/Interval or TCP/Interval type CheckType struct { Script string HTTP string + TCP string Interval time.Duration Timeout time.Duration @@ -51,7 +53,7 @@ type CheckTypes []*CheckType // Valid checks if the CheckType is valid func (c *CheckType) Valid() bool { - return c.IsTTL() || c.IsMonitor() || c.IsHTTP() + return c.IsTTL() || c.IsMonitor() || c.IsHTTP() || c.IsTCP() } // IsTTL checks if this is a TTL type @@ -69,6 +71,11 @@ func (c *CheckType) IsHTTP() bool { return c.HTTP != "" && c.Interval != 0 } +// IsTCP checks if this is a TCP type +func (c *CheckType) IsTCP() bool { + return c.TCP != "" && c.Interval != 0 +} + // CheckNotifier interface is used by the CheckMonitor // to notify when a check has a status update. The update // should take care to be idempotent. @@ -402,3 +409,86 @@ func (c *CheckHTTP) check() { c.Notify.UpdateCheck(c.CheckID, structs.HealthCritical, result) } } + +// CheckTCP is used to periodically make an TCP/UDP connection to +// determine the health of a given check. +// The check is passing if the connection succeeds +// The check is critical if the connection returns an error +type CheckTCP struct { + Notify CheckNotifier + CheckID string + TCP string + Interval time.Duration + Timeout time.Duration + Logger *log.Logger + + dialer *net.Dialer + stop bool + stopCh chan struct{} + stopLock sync.Mutex +} + +// Start is used to start a TCP check. +// The check runs until stop is called +func (c *CheckTCP) Start() { + c.stopLock.Lock() + defer c.stopLock.Unlock() + + if c.dialer == nil { + // Create the socket dialer + c.dialer = &net.Dialer{DualStack: true} + + // For long (>10s) interval checks the socket timeout is 10s, otherwise + // the timeout is the interval. This means that a check *should* return + // before the next check begins. + if c.Timeout > 0 && c.Timeout < c.Interval { + c.dialer.Timeout = c.Timeout + } else if c.Interval < 10*time.Second { + c.dialer.Timeout = c.Interval + } + } + + c.stop = false + c.stopCh = make(chan struct{}) + go c.run() +} + +// Stop is used to stop a TCP check. +func (c *CheckTCP) Stop() { + c.stopLock.Lock() + defer c.stopLock.Unlock() + if !c.stop { + c.stop = true + close(c.stopCh) + } +} + +// run is invoked by a goroutine to run until Stop() is called +func (c *CheckTCP) run() { + // Get the randomized initial pause time + initialPauseTime := randomStagger(c.Interval) + c.Logger.Printf("[DEBUG] agent: pausing %v before first socket connection of %s", initialPauseTime, c.TCP) + next := time.After(initialPauseTime) + for { + select { + case <-next: + c.check() + next = time.After(c.Interval) + case <-c.stopCh: + return + } + } +} + +// check is invoked periodically to perform the TCP check +func (c *CheckTCP) check() { + conn, err := c.dialer.Dial(`tcp`, c.TCP) + if err != nil { + c.Logger.Printf("[WARN] agent: socket connection failed '%s': %s", c.TCP, err) + c.Notify.UpdateCheck(c.CheckID, structs.HealthCritical, err.Error()) + return + } + conn.Close() + c.Logger.Printf("[DEBUG] agent: check '%v' is passing", c.CheckID) + c.Notify.UpdateCheck(c.CheckID, structs.HealthPassing, fmt.Sprintf("TCP connect %s: Success", c.TCP)) +} diff --git a/command/agent/check_test.go b/command/agent/check_test.go index 3fbd0ff97a..6b9f59df7e 100644 --- a/command/agent/check_test.go +++ b/command/agent/check_test.go @@ -3,6 +3,7 @@ package agent import ( "fmt" "log" + "net" "net/http" "net/http/httptest" "os" @@ -321,3 +322,74 @@ func TestCheckHTTP_disablesKeepAlives(t *testing.T) { t.Fatalf("should have disabled keepalives") } } + +func mockTCPServer(network string) net.Listener { + var ( + addr string + ) + + if network == `tcp6` { + addr = `[::1]:0` + } else { + addr = `127.0.0.1:0` + } + + listener, err := net.Listen(network, addr) + if err != nil { + panic(err) + } + + return listener +} + +func expectTCPStatus(t *testing.T, tcp string, status string) { + mock := &MockNotify{ + state: make(map[string]string), + updates: make(map[string]int), + output: make(map[string]string), + } + check := &CheckTCP{ + Notify: mock, + CheckID: "foo", + TCP: tcp, + Interval: 10 * time.Millisecond, + Logger: log.New(os.Stderr, "", log.LstdFlags), + } + check.Start() + defer check.Stop() + + time.Sleep(50 * time.Millisecond) + + // Should have at least 2 updates + if mock.updates["foo"] < 2 { + t.Fatalf("should have 2 updates %v", mock.updates) + } + + if mock.state["foo"] != status { + t.Fatalf("should be %v %v", status, mock.state) + } +} + +func TestCheckTCPCritical(t *testing.T) { + var ( + tcpServer net.Listener + ) + + tcpServer = mockTCPServer(`tcp`) + expectTCPStatus(t, `127.0.0.1:0`, "critical") + tcpServer.Close() +} + +func TestCheckTCPPassing(t *testing.T) { + var ( + tcpServer net.Listener + ) + + tcpServer = mockTCPServer(`tcp`) + expectTCPStatus(t, tcpServer.Addr().String(), "passing") + tcpServer.Close() + + tcpServer = mockTCPServer(`tcp6`) + expectTCPStatus(t, tcpServer.Addr().String(), "passing") + tcpServer.Close() +}