feat: add new jail config metrics

Add new metrics around basic jail configuration. The new metrics expose the
max retries, ban time, and find time for each jail.
Update project README with the new metrics.
This commit is contained in:
Hector 2021-10-14 20:52:25 +00:00
parent bb5c15de1b
commit 56730c8774
4 changed files with 95 additions and 0 deletions

View File

@ -157,6 +157,9 @@ Exposed metrics:
* `jail_banned_total` (per jail) - Total number of banned IPs since fail2ban startup (includes expired bans)
* `jail_failed_current` (per jail) - Number of current failures
* `jail_failed_total` (per jail) - Total number of failures since fail2ban startup
* `jail_config_ban_time` (per jail) - How long an IP is banned for in this jail (in seconds)
* `jail_config_find_time` (per jail) - How far back the filter will look for failures in this jail (in seconds)
* `jail_config_max_retry` (per jail) - The max number of failures allowed before banning an IP in this jail
* `version` - Version string of the exporter and fail2ban
**Sample**
@ -186,6 +189,18 @@ f2b_jail_failed_current{jail="sshd"} 6
# TYPE f2b_jail_failed_total gauge
f2b_jail_failed_total{jail="recidive"} 7
f2b_jail_failed_total{jail="sshd"} 125
# HELP f2b_config_jail_ban_time How long an IP is banned for in this jail (in seconds)
# TYPE f2b_config_jail_ban_time gauge
f2b_config_jail_ban_time{jail="recidive"} 604800
f2b_config_jail_ban_time{jail="sshd"} 600
# HELP f2b_config_jail_find_time How far back will the filter look for failures in this jail (in seconds)
# TYPE f2b_config_jail_find_time gauge
f2b_config_jail_find_time{jail="recidive"} 86400
f2b_config_jail_find_time{jail="sshd"} 600
# HELP f2b_config_jail_max_retries The number of failures allowed until the IP is banned by this jail
# TYPE f2b_config_jail_max_retries gauge
f2b_config_jail_max_retries{jail="recidive"} 5
f2b_config_jail_max_retries{jail="sshd"} 5
# HELP f2b_up Check if the fail2ban server is up
# TYPE f2b_up gauge
f2b_up 1

View File

@ -46,6 +46,21 @@ var (
"Total number of IPs banned by this jail (includes expired bans)",
[]string{"jail"}, nil,
)
metricJailBanTime = prometheus.NewDesc(
prometheus.BuildFQName(namespace, "config", "jail_ban_time"),
"How long an IP is banned for in this jail (in seconds)",
[]string{"jail"}, nil,
)
metricJailFindTime = prometheus.NewDesc(
prometheus.BuildFQName(namespace, "config", "jail_find_time"),
"How far back will the filter look for failures in this jail (in seconds)",
[]string{"jail"}, nil,
)
metricJailMaxRetry = prometheus.NewDesc(
prometheus.BuildFQName(namespace, "config", "jail_max_retries"),
"The number of failures allowed until the IP is banned by this jail",
[]string{"jail"}, nil,
)
metricVersionInfo = prometheus.NewDesc(
prometheus.BuildFQName(namespace, "", "version"),
"Version of the exporter and fail2ban server",
@ -98,6 +113,7 @@ func (c *Collector) collectJailMetrics(ch chan<- prometheus.Metric, s *socket.Fa
for i := range jails {
c.collectJailStatsMetric(ch, s, jails[i])
c.collectJailConfigMetrics(ch, s, jails[i])
}
}
@ -123,6 +139,36 @@ func (c *Collector) collectJailStatsMetric(ch chan<- prometheus.Metric, s *socke
)
}
func (c *Collector) collectJailConfigMetrics(ch chan<- prometheus.Metric, s *socket.Fail2BanSocket, jail string) {
banTime, err := s.GetJailBanTime(jail)
if err != nil {
c.socketRequestErrorCount++
log.Printf("failed to get ban time for jail %s: %v", jail, err)
} else {
ch <- prometheus.MustNewConstMetric(
metricJailBanTime, prometheus.GaugeValue, float64(banTime), jail,
)
}
findTime, err := s.GetJailFindTime(jail)
if err != nil {
c.socketRequestErrorCount++
log.Printf("failed to get find time for jail %s: %v", jail, err)
} else {
ch <- prometheus.MustNewConstMetric(
metricJailFindTime, prometheus.GaugeValue, float64(findTime), jail,
)
}
maxRetry, err := s.GetJailMaxRetries(jail)
if err != nil {
c.socketRequestErrorCount++
log.Printf("failed to get max retries for jail %s: %v", jail, err)
} else {
ch <- prometheus.MustNewConstMetric(
metricJailMaxRetry, prometheus.GaugeValue, float64(maxRetry), jail,
)
}
}
func (c *Collector) collectVersionMetric(ch chan<- prometheus.Metric, s *socket.Fail2BanSocket) {
var err error
var fail2banVersion = ""

View File

@ -118,6 +118,21 @@ func (s *Fail2BanSocket) GetJailStats(jail string) (JailStats, error) {
return stats, newBadFormatError(statusCommand, response)
}
func (s *Fail2BanSocket) GetJailBanTime(jail string) (int, error) {
command := fmt.Sprintf(banTimeCommandFmt, jail)
return s.sendSimpleIntCommand(command)
}
func (s *Fail2BanSocket) GetJailFindTime(jail string) (int, error) {
command := fmt.Sprintf(findTimeCommandFmt, jail)
return s.sendSimpleIntCommand(command)
}
func (s *Fail2BanSocket) GetJailMaxRetries(jail string) (int, error) {
command := fmt.Sprintf(maxRetriesCommandFmt, jail)
return s.sendSimpleIntCommand(command)
}
func (s *Fail2BanSocket) GetServerVersion() (string, error) {
response, err := s.sendCommand([]string{versionCommand})
if err != nil {
@ -132,6 +147,22 @@ func (s *Fail2BanSocket) GetServerVersion() (string, error) {
return "", newBadFormatError(versionCommand, response)
}
// sendSimpleIntCommand sends a command to the fail2ban socket and parses the response to extract an int.
// This command assumes that the response data is in the format of `(d, d)` where `d` is a number.
func (s *Fail2BanSocket) sendSimpleIntCommand(command string) (int, error) {
response, err := s.sendCommand(strings.Split(command, " "))
if err != nil {
return -1, err
}
if lvl1, ok := response.(*types.Tuple); ok {
if banTime, ok := lvl1.Get(1).(int); ok {
return banTime, nil
}
}
return -1, newBadFormatError(command, response)
}
func newBadFormatError(command string, data interface{}) error {
return fmt.Errorf("(%s) unexpected response format - cannot parse: %v", command, data)
}

View File

@ -12,6 +12,9 @@ const (
pingCommand = "ping"
statusCommand = "status"
versionCommand = "version"
banTimeCommandFmt = "get %s bantime"
findTimeCommandFmt = "get %s findtime"
maxRetriesCommandFmt = "get %s maxretry"
socketReadBufferSize = 1024
)