2014-01-21 02:44:23 +00:00
|
|
|
package agent
|
|
|
|
|
|
|
|
import (
|
2014-01-30 21:39:02 +00:00
|
|
|
"fmt"
|
2015-01-12 22:35:28 +00:00
|
|
|
"io/ioutil"
|
2014-01-21 02:44:23 +00:00
|
|
|
"log"
|
2015-01-09 22:43:24 +00:00
|
|
|
"net/http"
|
2014-01-21 02:44:23 +00:00
|
|
|
"os/exec"
|
|
|
|
"sync"
|
|
|
|
"syscall"
|
|
|
|
"time"
|
2015-01-13 20:18:18 +00:00
|
|
|
|
|
|
|
"github.com/armon/circbuf"
|
|
|
|
"github.com/hashicorp/consul/consul/structs"
|
2014-01-21 02:44:23 +00:00
|
|
|
)
|
|
|
|
|
2014-04-21 21:42:42 +00:00
|
|
|
const (
|
|
|
|
// Do not allow for a interval below this value.
|
|
|
|
// Otherwise we risk fork bombing a system.
|
|
|
|
MinInterval = time.Second
|
2014-04-29 22:28:56 +00:00
|
|
|
|
|
|
|
// Limit the size of a check's output to the
|
|
|
|
// last CheckBufSize. Prevents an enormous buffer
|
|
|
|
// from being captured
|
|
|
|
CheckBufSize = 4 * 1024 // 4KB
|
2015-05-18 17:12:10 +00:00
|
|
|
|
|
|
|
// Use this user agent when doing requests for
|
|
|
|
// HTTP health checks.
|
|
|
|
HttpUserAgent = "Consul Health Check"
|
2014-04-21 21:42:42 +00:00
|
|
|
)
|
|
|
|
|
2014-01-30 21:18:05 +00:00
|
|
|
// CheckType is used to create either the CheckMonitor
|
2015-01-09 22:43:24 +00:00
|
|
|
// or the CheckTTL.
|
|
|
|
// Three types are supported: Script, HTTP, and TTL
|
|
|
|
// Script and HTTP both require Interval
|
|
|
|
// Only one of the types needs to be provided
|
|
|
|
// TTL or Script/Interval or HTTP/Interval
|
2014-01-30 21:18:05 +00:00
|
|
|
type CheckType struct {
|
|
|
|
Script string
|
2015-01-09 22:43:24 +00:00
|
|
|
HTTP string
|
2014-01-30 21:18:05 +00:00
|
|
|
Interval time.Duration
|
|
|
|
|
2015-01-29 06:37:48 +00:00
|
|
|
Timeout time.Duration
|
|
|
|
TTL time.Duration
|
2014-11-07 02:24:04 +00:00
|
|
|
|
2015-04-12 00:53:48 +00:00
|
|
|
Status string
|
|
|
|
|
2014-11-07 02:24:04 +00:00
|
|
|
Notes string
|
2014-01-30 21:18:05 +00:00
|
|
|
}
|
2015-01-14 01:52:17 +00:00
|
|
|
type CheckTypes []*CheckType
|
2014-01-30 21:18:05 +00:00
|
|
|
|
|
|
|
// Valid checks if the CheckType is valid
|
|
|
|
func (c *CheckType) Valid() bool {
|
2015-01-09 22:43:24 +00:00
|
|
|
return c.IsTTL() || c.IsMonitor() || c.IsHTTP()
|
2014-01-30 21:18:05 +00:00
|
|
|
}
|
|
|
|
|
|
|
|
// IsTTL checks if this is a TTL type
|
|
|
|
func (c *CheckType) IsTTL() bool {
|
|
|
|
return c.TTL != 0
|
|
|
|
}
|
|
|
|
|
|
|
|
// IsMonitor checks if this is a Monitor type
|
|
|
|
func (c *CheckType) IsMonitor() bool {
|
|
|
|
return c.Script != "" && c.Interval != 0
|
|
|
|
}
|
|
|
|
|
2015-01-09 22:43:24 +00:00
|
|
|
// IsHTTP checks if this is a HTTP type
|
|
|
|
func (c *CheckType) IsHTTP() bool {
|
|
|
|
return c.HTTP != "" && c.Interval != 0
|
|
|
|
}
|
|
|
|
|
2014-01-21 02:44:23 +00:00
|
|
|
// CheckNotifier interface is used by the CheckMonitor
|
|
|
|
// to notify when a check has a status update. The update
|
|
|
|
// should take care to be idempotent.
|
|
|
|
type CheckNotifier interface {
|
2014-04-21 23:20:22 +00:00
|
|
|
UpdateCheck(checkID, status, output string)
|
2014-01-21 02:44:23 +00:00
|
|
|
}
|
|
|
|
|
|
|
|
// CheckMonitor is used to periodically invoke a script to
|
|
|
|
// determine the health of a given check. It is compatible with
|
|
|
|
// nagios plugins and expects the output in the same format.
|
|
|
|
type CheckMonitor struct {
|
|
|
|
Notify CheckNotifier
|
|
|
|
CheckID string
|
|
|
|
Script string
|
|
|
|
Interval time.Duration
|
|
|
|
Logger *log.Logger
|
|
|
|
|
|
|
|
stop bool
|
|
|
|
stopCh chan struct{}
|
|
|
|
stopLock sync.Mutex
|
|
|
|
}
|
|
|
|
|
|
|
|
// Start is used to start a check monitor.
|
|
|
|
// Monitor runs until stop is called
|
|
|
|
func (c *CheckMonitor) Start() {
|
|
|
|
c.stopLock.Lock()
|
|
|
|
defer c.stopLock.Unlock()
|
|
|
|
c.stop = false
|
|
|
|
c.stopCh = make(chan struct{})
|
|
|
|
go c.run()
|
|
|
|
}
|
|
|
|
|
|
|
|
// Stop is used to stop a check monitor.
|
|
|
|
func (c *CheckMonitor) 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 *CheckMonitor) run() {
|
2014-12-18 02:44:12 +00:00
|
|
|
// Get the randomized initial pause time
|
2014-12-18 03:24:41 +00:00
|
|
|
initialPauseTime := randomStagger(c.Interval)
|
2014-12-18 14:00:51 +00:00
|
|
|
c.Logger.Printf("[DEBUG] agent: pausing %v before first invocation of %s", initialPauseTime, c.Script)
|
2014-12-18 02:44:12 +00:00
|
|
|
next := time.After(initialPauseTime)
|
2014-01-21 02:46:01 +00:00
|
|
|
for {
|
|
|
|
select {
|
|
|
|
case <-next:
|
|
|
|
c.check()
|
|
|
|
next = time.After(c.Interval)
|
|
|
|
case <-c.stopCh:
|
|
|
|
return
|
|
|
|
}
|
2014-01-21 02:44:23 +00:00
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
// check is invoked periodically to perform the script check
|
|
|
|
func (c *CheckMonitor) check() {
|
|
|
|
// Create the command
|
2014-08-21 21:28:16 +00:00
|
|
|
cmd, err := ExecScript(c.Script)
|
|
|
|
if err != nil {
|
|
|
|
c.Logger.Printf("[ERR] agent: failed to setup invoke '%s': %s", c.Script, err)
|
2014-10-15 17:14:46 +00:00
|
|
|
c.Notify.UpdateCheck(c.CheckID, structs.HealthCritical, err.Error())
|
2014-08-21 21:28:16 +00:00
|
|
|
return
|
|
|
|
}
|
2014-01-21 02:44:23 +00:00
|
|
|
|
|
|
|
// Collect the output
|
2014-04-29 22:28:56 +00:00
|
|
|
output, _ := circbuf.NewBuffer(CheckBufSize)
|
|
|
|
cmd.Stdout = output
|
|
|
|
cmd.Stderr = output
|
2014-01-21 02:44:23 +00:00
|
|
|
|
|
|
|
// Start the check
|
|
|
|
if err := cmd.Start(); err != nil {
|
|
|
|
c.Logger.Printf("[ERR] agent: failed to invoke '%s': %s", c.Script, err)
|
2014-10-15 17:14:46 +00:00
|
|
|
c.Notify.UpdateCheck(c.CheckID, structs.HealthCritical, err.Error())
|
2014-01-21 02:44:23 +00:00
|
|
|
return
|
|
|
|
}
|
|
|
|
|
|
|
|
// Wait for the check to complete
|
2014-01-30 21:39:02 +00:00
|
|
|
errCh := make(chan error, 2)
|
|
|
|
go func() {
|
|
|
|
errCh <- cmd.Wait()
|
|
|
|
}()
|
|
|
|
go func() {
|
|
|
|
time.Sleep(30 * time.Second)
|
|
|
|
errCh <- fmt.Errorf("Timed out running check '%s'", c.Script)
|
|
|
|
}()
|
2014-08-21 21:28:16 +00:00
|
|
|
err = <-errCh
|
2014-01-30 21:39:02 +00:00
|
|
|
|
2014-04-29 22:28:56 +00:00
|
|
|
// Get the output, add a message about truncation
|
2014-04-21 23:20:22 +00:00
|
|
|
outputStr := string(output.Bytes())
|
2014-04-29 22:28:56 +00:00
|
|
|
if output.TotalWritten() > output.Size() {
|
|
|
|
outputStr = fmt.Sprintf("Captured %d of %d bytes\n...\n%s",
|
|
|
|
output.Size(), output.TotalWritten(), outputStr)
|
|
|
|
}
|
|
|
|
|
2014-01-21 02:44:23 +00:00
|
|
|
c.Logger.Printf("[DEBUG] agent: check '%s' script '%s' output: %s",
|
2014-04-21 23:20:22 +00:00
|
|
|
c.CheckID, c.Script, outputStr)
|
2014-01-21 02:44:23 +00:00
|
|
|
|
|
|
|
// Check if the check passed
|
|
|
|
if err == nil {
|
2015-01-13 20:18:18 +00:00
|
|
|
c.Logger.Printf("[DEBUG] agent: Check '%v' is passing", c.CheckID)
|
2014-04-21 23:20:22 +00:00
|
|
|
c.Notify.UpdateCheck(c.CheckID, structs.HealthPassing, outputStr)
|
2014-01-21 02:44:23 +00:00
|
|
|
return
|
|
|
|
}
|
|
|
|
|
|
|
|
// If the exit code is 1, set check as warning
|
|
|
|
exitErr, ok := err.(*exec.ExitError)
|
|
|
|
if ok {
|
|
|
|
if status, ok := exitErr.Sys().(syscall.WaitStatus); ok {
|
|
|
|
code := status.ExitStatus()
|
|
|
|
if code == 1 {
|
2015-01-13 20:18:18 +00:00
|
|
|
c.Logger.Printf("[WARN] agent: Check '%v' is now warning", c.CheckID)
|
2014-04-21 23:20:22 +00:00
|
|
|
c.Notify.UpdateCheck(c.CheckID, structs.HealthWarning, outputStr)
|
2014-01-21 02:44:23 +00:00
|
|
|
return
|
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
// Set the health as critical
|
2015-01-13 20:18:18 +00:00
|
|
|
c.Logger.Printf("[WARN] agent: Check '%v' is now critical", c.CheckID)
|
2014-04-21 23:20:22 +00:00
|
|
|
c.Notify.UpdateCheck(c.CheckID, structs.HealthCritical, outputStr)
|
2014-01-21 02:44:23 +00:00
|
|
|
}
|
2014-01-21 03:12:40 +00:00
|
|
|
|
|
|
|
// CheckTTL is used to apply a TTL to check status,
|
|
|
|
// and enables clients to set the status of a check
|
|
|
|
// but upon the TTL expiring, the check status is
|
|
|
|
// automatically set to critical.
|
|
|
|
type CheckTTL struct {
|
|
|
|
Notify CheckNotifier
|
|
|
|
CheckID string
|
|
|
|
TTL time.Duration
|
|
|
|
Logger *log.Logger
|
|
|
|
|
|
|
|
timer *time.Timer
|
|
|
|
|
|
|
|
stop bool
|
|
|
|
stopCh chan struct{}
|
|
|
|
stopLock sync.Mutex
|
|
|
|
}
|
|
|
|
|
|
|
|
// Start is used to start a check ttl, runs until Stop()
|
|
|
|
func (c *CheckTTL) Start() {
|
|
|
|
c.stopLock.Lock()
|
|
|
|
defer c.stopLock.Unlock()
|
|
|
|
c.stop = false
|
|
|
|
c.stopCh = make(chan struct{})
|
|
|
|
c.timer = time.NewTimer(c.TTL)
|
|
|
|
go c.run()
|
|
|
|
}
|
|
|
|
|
|
|
|
// Stop is used to stop a check ttl.
|
|
|
|
func (c *CheckTTL) Stop() {
|
|
|
|
c.stopLock.Lock()
|
|
|
|
defer c.stopLock.Unlock()
|
|
|
|
if !c.stop {
|
|
|
|
c.timer.Stop()
|
|
|
|
c.stop = true
|
|
|
|
close(c.stopCh)
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
// run is used to handle TTL expiration and to update the check status
|
|
|
|
func (c *CheckTTL) run() {
|
|
|
|
for {
|
|
|
|
select {
|
|
|
|
case <-c.timer.C:
|
2015-01-13 20:18:18 +00:00
|
|
|
c.Logger.Printf("[WARN] agent: Check '%v' missed TTL, is now critical",
|
2014-01-21 03:12:40 +00:00
|
|
|
c.CheckID)
|
2014-01-21 03:19:20 +00:00
|
|
|
c.Notify.UpdateCheck(c.CheckID, structs.HealthCritical, "TTL expired")
|
2014-01-21 03:12:40 +00:00
|
|
|
|
|
|
|
case <-c.stopCh:
|
|
|
|
return
|
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
// SetStatus is used to update the status of the check,
|
|
|
|
// and to renew the TTL. If expired, TTL is restarted.
|
2014-04-21 23:20:22 +00:00
|
|
|
func (c *CheckTTL) SetStatus(status, output string) {
|
2015-01-13 20:18:18 +00:00
|
|
|
c.Logger.Printf("[DEBUG] agent: Check '%v' status is now %v",
|
2014-01-21 03:12:40 +00:00
|
|
|
c.CheckID, status)
|
2014-04-21 23:20:22 +00:00
|
|
|
c.Notify.UpdateCheck(c.CheckID, status, output)
|
2014-01-21 03:12:40 +00:00
|
|
|
c.timer.Reset(c.TTL)
|
|
|
|
}
|
2014-11-29 20:25:01 +00:00
|
|
|
|
|
|
|
// persistedCheck is used to serialize a check and write it to disk
|
|
|
|
// so that it may be restored later on.
|
|
|
|
type persistedCheck struct {
|
|
|
|
Check *structs.HealthCheck
|
|
|
|
ChkType *CheckType
|
2015-04-28 02:01:02 +00:00
|
|
|
Token string
|
2014-11-29 20:25:01 +00:00
|
|
|
}
|
2015-01-09 22:43:24 +00:00
|
|
|
|
|
|
|
// CheckHTTP is used to periodically make an HTTP request to
|
|
|
|
// determine the health of a given check.
|
2015-01-13 20:18:18 +00:00
|
|
|
// The check is passing if the response code is 2XX.
|
|
|
|
// The check is warning if the response code is 429.
|
2015-01-09 22:43:24 +00:00
|
|
|
// The check is critical if the response code is anything else
|
|
|
|
// or if the request returns an error
|
|
|
|
type CheckHTTP struct {
|
|
|
|
Notify CheckNotifier
|
|
|
|
CheckID string
|
|
|
|
HTTP string
|
|
|
|
Interval time.Duration
|
2015-01-29 06:37:48 +00:00
|
|
|
Timeout time.Duration
|
2015-01-09 22:43:24 +00:00
|
|
|
Logger *log.Logger
|
|
|
|
|
|
|
|
httpClient *http.Client
|
|
|
|
stop bool
|
|
|
|
stopCh chan struct{}
|
|
|
|
stopLock sync.Mutex
|
|
|
|
}
|
|
|
|
|
|
|
|
// Start is used to start an HTTP check.
|
|
|
|
// The check runs until stop is called
|
|
|
|
func (c *CheckHTTP) Start() {
|
|
|
|
c.stopLock.Lock()
|
|
|
|
defer c.stopLock.Unlock()
|
2015-01-12 23:58:25 +00:00
|
|
|
|
|
|
|
if c.httpClient == nil {
|
2015-03-15 20:30:50 +00:00
|
|
|
// Create the transport. We disable HTTP Keep-Alive's to prevent
|
|
|
|
// failing checks due to the keepalive interval.
|
|
|
|
trans := *http.DefaultTransport.(*http.Transport)
|
|
|
|
trans.DisableKeepAlives = true
|
|
|
|
|
|
|
|
// Create the HTTP client.
|
|
|
|
c.httpClient = &http.Client{
|
|
|
|
Timeout: 10 * time.Second,
|
|
|
|
Transport: &trans,
|
|
|
|
}
|
|
|
|
|
2015-01-12 23:58:25 +00:00
|
|
|
// For long (>10s) interval checks the http timeout is 10s, otherwise the
|
|
|
|
// timeout is the interval. This means that a check *should* return
|
|
|
|
// before the next check begins.
|
2015-01-29 06:37:48 +00:00
|
|
|
if c.Timeout > 0 && c.Timeout < c.Interval {
|
2015-03-15 20:30:50 +00:00
|
|
|
c.httpClient.Timeout = c.Timeout
|
2015-01-29 06:37:48 +00:00
|
|
|
} else if c.Interval < 10*time.Second {
|
2015-03-15 20:30:50 +00:00
|
|
|
c.httpClient.Timeout = c.Interval
|
2015-01-12 23:58:25 +00:00
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2015-01-09 22:43:24 +00:00
|
|
|
c.stop = false
|
|
|
|
c.stopCh = make(chan struct{})
|
|
|
|
go c.run()
|
|
|
|
}
|
|
|
|
|
|
|
|
// Stop is used to stop an HTTP check.
|
|
|
|
func (c *CheckHTTP) 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 *CheckHTTP) run() {
|
|
|
|
// Get the randomized initial pause time
|
|
|
|
initialPauseTime := randomStagger(c.Interval)
|
|
|
|
c.Logger.Printf("[DEBUG] agent: pausing %v before first HTTP request of %s", initialPauseTime, c.HTTP)
|
|
|
|
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 HTTP check
|
|
|
|
func (c *CheckHTTP) check() {
|
2015-05-18 17:12:10 +00:00
|
|
|
req, err := http.NewRequest("GET", c.HTTP, nil)
|
|
|
|
if err != nil {
|
|
|
|
c.Logger.Printf("[WARN] agent: http request failed '%s': %s", c.HTTP, err)
|
|
|
|
c.Notify.UpdateCheck(c.CheckID, structs.HealthCritical, err.Error())
|
|
|
|
return
|
|
|
|
}
|
|
|
|
|
|
|
|
req.Header.Set("User-Agent", HttpUserAgent)
|
|
|
|
|
|
|
|
resp, err := c.httpClient.Do(req)
|
2015-01-09 22:43:24 +00:00
|
|
|
if err != nil {
|
|
|
|
c.Logger.Printf("[WARN] agent: http request failed '%s': %s", c.HTTP, err)
|
|
|
|
c.Notify.UpdateCheck(c.CheckID, structs.HealthCritical, err.Error())
|
|
|
|
return
|
|
|
|
}
|
2015-01-12 22:35:28 +00:00
|
|
|
defer resp.Body.Close()
|
2015-01-09 22:43:24 +00:00
|
|
|
|
2015-01-13 20:18:18 +00:00
|
|
|
// Format the response body
|
|
|
|
body, err := ioutil.ReadAll(resp.Body)
|
|
|
|
if err != nil {
|
|
|
|
c.Logger.Printf("[WARN] agent: check '%v': Get error while reading body: %s", c.CheckID, err)
|
|
|
|
body = []byte{}
|
|
|
|
}
|
|
|
|
result := fmt.Sprintf("HTTP GET %s: %s Output: %s", c.HTTP, resp.Status, body)
|
|
|
|
|
2015-01-12 21:58:57 +00:00
|
|
|
if resp.StatusCode >= 200 && resp.StatusCode <= 299 {
|
|
|
|
// PASSING (2xx)
|
2015-01-13 20:18:18 +00:00
|
|
|
c.Logger.Printf("[DEBUG] agent: check '%v' is passing", c.CheckID)
|
2015-01-09 22:43:24 +00:00
|
|
|
c.Notify.UpdateCheck(c.CheckID, structs.HealthPassing, result)
|
|
|
|
|
2015-01-12 21:58:57 +00:00
|
|
|
} else if resp.StatusCode == 429 {
|
|
|
|
// WARNING
|
|
|
|
// 429 Too Many Requests (RFC 6585)
|
|
|
|
// The user has sent too many requests in a given amount of time.
|
2015-01-13 20:18:18 +00:00
|
|
|
c.Logger.Printf("[WARN] agent: check '%v' is now warning", c.CheckID)
|
|
|
|
c.Notify.UpdateCheck(c.CheckID, structs.HealthWarning, result)
|
2015-01-09 22:43:24 +00:00
|
|
|
|
2015-01-12 21:58:57 +00:00
|
|
|
} else {
|
|
|
|
// CRITICAL
|
2015-01-13 20:18:18 +00:00
|
|
|
c.Logger.Printf("[WARN] agent: check '%v' is now critical", c.CheckID)
|
|
|
|
c.Notify.UpdateCheck(c.CheckID, structs.HealthCritical, result)
|
2015-01-09 22:43:24 +00:00
|
|
|
}
|
|
|
|
}
|