riemann-tools-fb 0.3.0
Sign up to get free protection for your applications and to get access to all the features.
- checksums.yaml +7 -0
- data/LICENSE +21 -0
- data/README.markdown +18 -0
- data/bin/riemann-bench +70 -0
- data/bin/riemann-diskstats +86 -0
- data/bin/riemann-fd +66 -0
- data/bin/riemann-health +270 -0
- data/bin/riemann-net +101 -0
- data/bin/riemann-proc +48 -0
- data/lib/riemann/tools.rb +132 -0
- metadata +86 -0
checksums.yaml
ADDED
@@ -0,0 +1,7 @@
|
|
1
|
+
---
|
2
|
+
SHA1:
|
3
|
+
metadata.gz: 25dce28ceece7fa46dec1e11f34c25cd635755da
|
4
|
+
data.tar.gz: d9a177ddf05988aa078e2af1e5cd715487791856
|
5
|
+
SHA512:
|
6
|
+
metadata.gz: 81e9be74305a81fc5befef9f73afd37e34146a3a9ce117f88f398322226175a6d1742d8cda90bf250689961fed69cd710b2337a1a2e404cc9df577fec7d9a1f7
|
7
|
+
data.tar.gz: b1c56956e9f9dce4c891a59fba51a5769c38ce327a16c320737f4eb64bca7a0ae637d747fa00c1b17a9b65e81f8893c08e23988ee731e1a0955cee99cb6cfec6
|
data/LICENSE
ADDED
@@ -0,0 +1,21 @@
|
|
1
|
+
The MIT License
|
2
|
+
|
3
|
+
Copyright (c) 2011 Kyle Kingsbury
|
4
|
+
|
5
|
+
Permission is hereby granted, free of charge, to any person obtaining a copy
|
6
|
+
of this software and associated documentation files (the "Software"), to deal
|
7
|
+
in the Software without restriction, including without limitation the rights
|
8
|
+
to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
|
9
|
+
copies of the Software, and to permit persons to whom the Software is
|
10
|
+
furnished to do so, subject to the following conditions:
|
11
|
+
|
12
|
+
The above copyright notice and this permission notice shall be included in
|
13
|
+
all copies or substantial portions of the Software.
|
14
|
+
|
15
|
+
THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
|
16
|
+
IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
|
17
|
+
FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
|
18
|
+
AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
|
19
|
+
LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
|
20
|
+
OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
|
21
|
+
THE SOFTWARE.
|
data/README.markdown
ADDED
@@ -0,0 +1,18 @@
|
|
1
|
+
Riemann Tools
|
2
|
+
=============
|
3
|
+
|
4
|
+
Tiny programs to submit events to Riemann.
|
5
|
+
|
6
|
+
Riemann-health, for example, submits events about the current CPU, load,
|
7
|
+
memory, and disk use. Bench submits randomly distributed metrics for load
|
8
|
+
testing. I've got a whole bunch of these internally for monitoring Redis, Riak,
|
9
|
+
queues, etc. Most have internal configuration dependencies, so it'll be a while
|
10
|
+
before I can extract them for re-use.
|
11
|
+
|
12
|
+
Get started
|
13
|
+
==========
|
14
|
+
|
15
|
+
``` bash
|
16
|
+
gem install riemann-tools
|
17
|
+
riemann-health --host my.riemann.server
|
18
|
+
```
|
data/bin/riemann-bench
ADDED
@@ -0,0 +1,70 @@
|
|
1
|
+
#!/usr/bin/env ruby
|
2
|
+
|
3
|
+
# Connects to a server (first arg) and populates it with a constant stream of
|
4
|
+
# events for testing.
|
5
|
+
|
6
|
+
require 'rubygems'
|
7
|
+
require 'riemann/client'
|
8
|
+
require 'pp'
|
9
|
+
|
10
|
+
class Riemann::Bench
|
11
|
+
attr_accessor :client, :hosts, :services, :states
|
12
|
+
def initialize
|
13
|
+
@hosts = [nil] + (0...10).map { |i| "host#{i}" }
|
14
|
+
@hosts = %w(a b c d e f g h i j)
|
15
|
+
@services = %w(test1 test2 test3 foo bar baz xyzzy attack cat treat)
|
16
|
+
@states = {}
|
17
|
+
@client = Riemann::Client.new(:host => (ARGV.first || 'localhost'))
|
18
|
+
end
|
19
|
+
|
20
|
+
def evolve(state)
|
21
|
+
m = state[:metric] + (rand - 0.5) * 0.1
|
22
|
+
m = [[0,m].max, 1].min
|
23
|
+
|
24
|
+
s = case m
|
25
|
+
when 0...0.75
|
26
|
+
'ok'
|
27
|
+
when 0.75...0.9
|
28
|
+
'warning'
|
29
|
+
when 0.9..1.0
|
30
|
+
'critical'
|
31
|
+
end
|
32
|
+
|
33
|
+
{
|
34
|
+
:metric => m,
|
35
|
+
:state => s,
|
36
|
+
:host => state[:host],
|
37
|
+
:service => state[:service],
|
38
|
+
:description => "at #{Time.now}"
|
39
|
+
}
|
40
|
+
end
|
41
|
+
|
42
|
+
def tick
|
43
|
+
# pp @states
|
44
|
+
hosts.product(services).each do |id|
|
45
|
+
client << (states[id] = evolve(states[id]))
|
46
|
+
end
|
47
|
+
end
|
48
|
+
|
49
|
+
def run
|
50
|
+
start
|
51
|
+
loop do
|
52
|
+
sleep 0.05
|
53
|
+
tick
|
54
|
+
end
|
55
|
+
end
|
56
|
+
|
57
|
+
def start
|
58
|
+
hosts.product(services).each do |host, service|
|
59
|
+
states[[host, service]] = {
|
60
|
+
:metric => 0.5,
|
61
|
+
:state => 'ok',
|
62
|
+
:description => "Starting up",
|
63
|
+
:host => host,
|
64
|
+
:service => service
|
65
|
+
}
|
66
|
+
end
|
67
|
+
end
|
68
|
+
end
|
69
|
+
|
70
|
+
Riemann::Bench.new.run
|
@@ -0,0 +1,86 @@
|
|
1
|
+
#!/usr/bin/env ruby
|
2
|
+
#
|
3
|
+
require 'rubygems'
|
4
|
+
require 'riemann/tools'
|
5
|
+
|
6
|
+
class Riemann::Tools::Diskstats
|
7
|
+
include Riemann::Tools
|
8
|
+
|
9
|
+
opt :devices, "Devices to monitor", :type => :strings, :default => nil
|
10
|
+
opt :ignore_devices, "Devices to ignore", :type => :strings, :default =>nil
|
11
|
+
|
12
|
+
def initialize
|
13
|
+
@old_state = nil
|
14
|
+
end
|
15
|
+
|
16
|
+
def state
|
17
|
+
f = File.read('/proc/diskstats')
|
18
|
+
state = f.split("\n").reject { |d| d =~ /(ram|loop)/ }.inject({}) do |s, line|
|
19
|
+
if line =~ /^(?:\s+\d+){2}\s+([\w\d\-]+) (.*)$/
|
20
|
+
dev = $1
|
21
|
+
|
22
|
+
['reads reqs',
|
23
|
+
'reads merged',
|
24
|
+
'reads sector',
|
25
|
+
'reads time',
|
26
|
+
'writes reqs',
|
27
|
+
'writes merged',
|
28
|
+
'writes sector',
|
29
|
+
'writes time',
|
30
|
+
'io reqs',
|
31
|
+
'io time',
|
32
|
+
'io weighted'
|
33
|
+
].map do |service|
|
34
|
+
"#{dev} #{service}"
|
35
|
+
end.zip(
|
36
|
+
$2.split(/\s+/).map { |str| str.to_i }
|
37
|
+
).each do |service, value|
|
38
|
+
s[service] = value
|
39
|
+
end
|
40
|
+
end
|
41
|
+
|
42
|
+
s
|
43
|
+
end
|
44
|
+
|
45
|
+
# Filter interfaces
|
46
|
+
if is = opts[:devices]
|
47
|
+
state = state.select do |service, value|
|
48
|
+
is.include? service.split(' ').first
|
49
|
+
end
|
50
|
+
end
|
51
|
+
|
52
|
+
if ign = opts[:ignore_devices]
|
53
|
+
state = state.reject do |service, value|
|
54
|
+
ign.include? service.split(' ').first
|
55
|
+
end
|
56
|
+
end
|
57
|
+
|
58
|
+
state
|
59
|
+
end
|
60
|
+
|
61
|
+
def tick
|
62
|
+
state = self.state
|
63
|
+
|
64
|
+
if @old_state
|
65
|
+
state.each do |service, metric|
|
66
|
+
delta = metric - @old_state[service]
|
67
|
+
|
68
|
+
report(
|
69
|
+
:service => "diskstats " + service,
|
70
|
+
:metric => (delta.to_f / opts[:interval]),
|
71
|
+
:state => "ok"
|
72
|
+
)
|
73
|
+
|
74
|
+
if service =~ /io time$/
|
75
|
+
report(:service => "diskstats " + service.gsub(/time/, 'util'),
|
76
|
+
:metric => (delta.to_f / (opts[:interval]*1000)),
|
77
|
+
:state => "ok")
|
78
|
+
end
|
79
|
+
end
|
80
|
+
end
|
81
|
+
|
82
|
+
@old_state = state
|
83
|
+
end
|
84
|
+
end
|
85
|
+
|
86
|
+
Riemann::Tools::Diskstats.run
|
data/bin/riemann-fd
ADDED
@@ -0,0 +1,66 @@
|
|
1
|
+
#!/usr/bin/env ruby
|
2
|
+
|
3
|
+
# Reports current file descriptor use to riemann.
|
4
|
+
# By default reports the total system fd usage, can also report usage of individual processes
|
5
|
+
|
6
|
+
require File.expand_path('../../lib/riemann/tools', __FILE__)
|
7
|
+
|
8
|
+
class Riemann::Tools::Health
|
9
|
+
include Riemann::Tools
|
10
|
+
|
11
|
+
opt :fd_sys_warning, "open file descriptor threshold for system", :default => 800
|
12
|
+
opt :fd_sys_critical, "open file descriptor critical threshold for system", :default => 900
|
13
|
+
opt :fd_proc_warning, "open file descriptor threshold for process", :default => 800
|
14
|
+
opt :fd_proc_critical, "open file descriptor critical threshold for process", :default => 900
|
15
|
+
opt :processes, "list of processes to measure fd usage in addition to system total", :type => :ints
|
16
|
+
|
17
|
+
def initialize
|
18
|
+
@limits = {
|
19
|
+
:fd => {:critical => opts[:fd_sys_critical], :warning => opts[:fd_sys_warning]},
|
20
|
+
:process => {:critical => opts[:fd_proc_critical], :warning => opts[:fd_proc_warning]},
|
21
|
+
}
|
22
|
+
ostype = `uname -s`.chomp.downcase
|
23
|
+
puts "WARNING: OS '#{ostype}' not explicitly supported. Falling back to Linux" unless ostype == "linux"
|
24
|
+
@fd = method :linux_fd
|
25
|
+
end
|
26
|
+
|
27
|
+
def alert(service, state, metric, description)
|
28
|
+
report(
|
29
|
+
:service => service.to_s,
|
30
|
+
:state => state.to_s,
|
31
|
+
:metric => metric.to_f,
|
32
|
+
:description => description
|
33
|
+
)
|
34
|
+
end
|
35
|
+
|
36
|
+
def linux_fd
|
37
|
+
sys_used = Integer(`lsof | wc -l`)
|
38
|
+
if sys_used > @limits[:fd][:critical]
|
39
|
+
alert "fd sys", :critical, sys_used, "system is using #{sys_used} fds"
|
40
|
+
elsif sys_used > @limits[:fd][:warning]
|
41
|
+
alert "fd sys", :warning, sys_used, "system is using #{sys_used} fds"
|
42
|
+
else
|
43
|
+
alert "fd sys", :ok, sys_used, "system is using #{sys_used} fds"
|
44
|
+
end
|
45
|
+
|
46
|
+
unless opts[:processes].nil?
|
47
|
+
opts[:processes].each do |process|
|
48
|
+
used = Integer(`lsof -p #{process} | wc -l`)
|
49
|
+
name, pid = `ps axo comm,pid | grep -w #{process}`.split
|
50
|
+
if used > @limits[:process][:critical]
|
51
|
+
alert "fd #{name} #{process}", :critical, used, "process #{name} #{process} is using #{used} fds"
|
52
|
+
elsif used > @limits[:process][:warning]
|
53
|
+
alert "fd #{name} #{process}", :warning, used, "process #{name} #{process} is using #{used} fds"
|
54
|
+
else
|
55
|
+
alert "fd #{name} #{process}", :ok, used, "process #{name} #{process} is using #{used} fds"
|
56
|
+
end
|
57
|
+
end
|
58
|
+
end
|
59
|
+
end
|
60
|
+
|
61
|
+
def tick
|
62
|
+
@fd.call
|
63
|
+
end
|
64
|
+
end
|
65
|
+
|
66
|
+
Riemann::Tools::Health.run
|
data/bin/riemann-health
ADDED
@@ -0,0 +1,270 @@
|
|
1
|
+
#!/usr/bin/env ruby
|
2
|
+
|
3
|
+
# Reports current CPU, disk, load average, and memory use to riemann.
|
4
|
+
|
5
|
+
require File.expand_path('../../lib/riemann/tools', __FILE__)
|
6
|
+
|
7
|
+
class Riemann::Tools::Health
|
8
|
+
include Riemann::Tools
|
9
|
+
|
10
|
+
opt :cpu_warning, "CPU warning threshold (fraction of total jiffies)", :default => 0.9
|
11
|
+
opt :cpu_critical, "CPU critical threshold (fraction of total jiffies)", :default => 0.95
|
12
|
+
opt :disk_warning, "Disk warning threshold (fraction of space used)", :default => 0.9
|
13
|
+
opt :disk_critical, "Disk critical threshold (fraction of space used)", :default => 0.95
|
14
|
+
opt :load_warning, "Load warning threshold (load average / core)", :default => 3
|
15
|
+
opt :load_critical, "Load critical threshold (load average / core)", :default => 8
|
16
|
+
opt :memory_warning, "Memory warning threshold (fraction of RAM)", :default => 0.85
|
17
|
+
opt :memory_critical, "Memory critical threshold (fraction of RAM)", :default => 0.95
|
18
|
+
opt :checks, "A list of checks to run.", :type => :strings, :default => ['cpu', 'load', 'memory', 'disk']
|
19
|
+
|
20
|
+
def initialize
|
21
|
+
@limits = {
|
22
|
+
:cpu => {:critical => opts[:cpu_critical], :warning => opts[:cpu_warning]},
|
23
|
+
:disk => {:critical => opts[:disk_critical], :warning => opts[:disk_warning]},
|
24
|
+
:load => {:critical => opts[:load_critical], :warning => opts[:load_warning]},
|
25
|
+
:memory => {:critical => opts[:memory_critical], :warning => opts[:memory_warning]}
|
26
|
+
}
|
27
|
+
case (ostype = `uname -s`.chomp.downcase)
|
28
|
+
when 'darwin'
|
29
|
+
@cores = `sysctl -n hw.ncpu`.to_i
|
30
|
+
@cpu = method :darwin_cpu
|
31
|
+
@disk = method :disk
|
32
|
+
@load = method :darwin_load
|
33
|
+
@memory = method :darwin_memory
|
34
|
+
darwin_top
|
35
|
+
when 'freebsd'
|
36
|
+
@cores = `sysctl -n hw.ncpu`.to_i
|
37
|
+
@cpu = method :freebsd_cpu
|
38
|
+
@disk = method :disk
|
39
|
+
@load = method :freebsd_load
|
40
|
+
@memory = method :freebsd_memory
|
41
|
+
else
|
42
|
+
@cores = cores
|
43
|
+
puts "WARNING: OS '#{ostype}' not explicitly supported. Falling back to Linux" unless ostype == "linux"
|
44
|
+
@cpu = method :linux_cpu
|
45
|
+
@disk = method :disk
|
46
|
+
@load = method :linux_load
|
47
|
+
@memory = method :linux_memory
|
48
|
+
end
|
49
|
+
|
50
|
+
opts[:checks].each do |check|
|
51
|
+
case check
|
52
|
+
when "disk"
|
53
|
+
@disk_enabled = true
|
54
|
+
when "load"
|
55
|
+
@load_enabled = true
|
56
|
+
when "cpu"
|
57
|
+
@cpu_enabled = true
|
58
|
+
when "memory"
|
59
|
+
@memory_enabled = true
|
60
|
+
end
|
61
|
+
end
|
62
|
+
end
|
63
|
+
|
64
|
+
def alert(service, state, metric, description)
|
65
|
+
report(
|
66
|
+
:service => service.to_s,
|
67
|
+
:state => state.to_s,
|
68
|
+
:metric => metric.to_f,
|
69
|
+
:description => description
|
70
|
+
)
|
71
|
+
end
|
72
|
+
|
73
|
+
def cores
|
74
|
+
i = 0;
|
75
|
+
File.read("/proc/cpuinfo").split(/\n\n/).inject({}) do |cores, p|
|
76
|
+
physical_id = p[/physical id\s+:\s+(\d+)/, 1]
|
77
|
+
core_id = p[/core id\s+:\s+(\d+)/, 1]
|
78
|
+
if physical_id and core_id
|
79
|
+
cores["#{physical_id}:#{core_id}"] = true
|
80
|
+
elsif physical_id
|
81
|
+
cores["#{physical_id}:"] = true
|
82
|
+
else
|
83
|
+
cores[i += 1] = true;
|
84
|
+
end
|
85
|
+
|
86
|
+
cores
|
87
|
+
end.size
|
88
|
+
end
|
89
|
+
|
90
|
+
def report_pct(service, fraction, report)
|
91
|
+
if fraction
|
92
|
+
if fraction > @limits[service][:critical]
|
93
|
+
alert service, :critical, fraction, "#{sprintf("%.2f", fraction * 100)}% #{report}"
|
94
|
+
elsif fraction > @limits[service][:warning]
|
95
|
+
alert service, :warning, fraction, "#{sprintf("%.2f", fraction * 100)}% #{report}"
|
96
|
+
else
|
97
|
+
alert service, :ok, fraction, "#{sprintf("%.2f", fraction * 100)}% #{report}"
|
98
|
+
end
|
99
|
+
end
|
100
|
+
end
|
101
|
+
|
102
|
+
def linux_cpu
|
103
|
+
new = File.read('/proc/stat')
|
104
|
+
unless new[/cpu\s+(\d+)\s+(\d+)\s+(\d+)\s+(\d+)/]
|
105
|
+
alert 'cpu', :unknown, nil, "/proc/stat doesn't include a CPU line"
|
106
|
+
return false
|
107
|
+
end
|
108
|
+
u2, n2, s2, i2 = [$1, $2, $3, $4].map { |e| e.to_i }
|
109
|
+
|
110
|
+
if @old_cpu
|
111
|
+
u1, n1, s1, i1 = @old_cpu
|
112
|
+
|
113
|
+
used = (u2+n2+s2) - (u1+n1+s1)
|
114
|
+
total = used + i2-i1
|
115
|
+
fraction = used.to_f / total
|
116
|
+
|
117
|
+
report_pct :cpu, fraction, "user+nice+sytem\n\n#{`ps -eo pcpu,pid,comm | sort -nrb -k1 | head -10`.chomp}"
|
118
|
+
end
|
119
|
+
|
120
|
+
@old_cpu = [u2, n2, s2, i2]
|
121
|
+
end
|
122
|
+
|
123
|
+
def linux_load
|
124
|
+
load = File.read('/proc/loadavg').split(/\s+/)[0].to_f / @cores
|
125
|
+
if load > @limits[:load][:critical]
|
126
|
+
alert "load", :critical, load, "1-minute load average/core is #{load}"
|
127
|
+
elsif load > @limits[:load][:warning]
|
128
|
+
alert "load", :warning, load, "1-minute load average/core is #{load}"
|
129
|
+
else
|
130
|
+
alert "load", :ok, load, "1-minute load average/core is #{load}"
|
131
|
+
end
|
132
|
+
end
|
133
|
+
|
134
|
+
def linux_memory
|
135
|
+
m = File.read('/proc/meminfo').split(/\n/).inject({}) { |info, line|
|
136
|
+
x = line.split(/:?\s+/)
|
137
|
+
# Assume kB...
|
138
|
+
info[x[0]] = x[1].to_i
|
139
|
+
info
|
140
|
+
}
|
141
|
+
|
142
|
+
free = m['MemFree'].to_i + m['Buffers'].to_i + m['Cached'].to_i
|
143
|
+
total = m['MemTotal'].to_i
|
144
|
+
fraction = 1 - (free.to_f / total)
|
145
|
+
|
146
|
+
report_pct :memory, fraction, "used\n\n#{`ps -eo pmem,pid,comm | sort -nrb -k1 | head -10`.chomp}"
|
147
|
+
end
|
148
|
+
|
149
|
+
def freebsd_cpu
|
150
|
+
u2, n2, s2, t2, i2 = `sysctl -n kern.cp_time 2>/dev/null`.split.map{ |e| e.to_i } #FreeBSD has 5 cpu stats
|
151
|
+
|
152
|
+
if @old_cpu
|
153
|
+
u1, n1, s1, t1, i1 = @old_cpu
|
154
|
+
|
155
|
+
used = (u2+n2+s2+t2) - (u1+n1+s1+t1)
|
156
|
+
total = used + i2-i1
|
157
|
+
fraction = used.to_f / total
|
158
|
+
|
159
|
+
report_pct :cpu, fraction, "user+nice+sytem+interrupt\n\n#{`ps -axo pcpu,pid,comm | sort -nrb -k1 | head -10`.chomp}"
|
160
|
+
end
|
161
|
+
|
162
|
+
@old_cpu = [u2, n2, s2, t2, i2]
|
163
|
+
end
|
164
|
+
|
165
|
+
def freebsd_load
|
166
|
+
m = `uptime`.split(':')[-1].chomp.gsub(/\s+/,'').split(',')
|
167
|
+
load = m[0].to_f / @cores
|
168
|
+
if load > @limits[:load][:critical]
|
169
|
+
alert "load", :critical, load, "1-minute load average/core is #{load}"
|
170
|
+
elsif load > @limits[:load][:warning]
|
171
|
+
alert "load", :warning, load, "1-minute load average/core is #{load}"
|
172
|
+
else
|
173
|
+
alert "load", :ok, load, "1-minute load average/core is #{load}"
|
174
|
+
end
|
175
|
+
end
|
176
|
+
|
177
|
+
def freebsd_memory
|
178
|
+
meminfo = `sysctl -n vm.stats.vm.v_page_count vm.stats.vm.v_wire_count vm.stats.vm.v_active_count 2>/dev/null`.chomp.split
|
179
|
+
fraction = (meminfo[1].to_f + meminfo[2].to_f) / meminfo[0].to_f
|
180
|
+
|
181
|
+
report_pct :memory, fraction, "used\n\n#{`ps -axo pmem,pid,comm | sort -nrb -k1 | head -10`.chomp}"
|
182
|
+
end
|
183
|
+
|
184
|
+
def darwin_top
|
185
|
+
raw = `top -l 1 | grep -i "^\\(cpu\\|physmem\\|load\\)"`.chomp
|
186
|
+
@topdata = {:stamp => Time.now.to_i }
|
187
|
+
raw.each_line do |ln|
|
188
|
+
if ln.match(/Load Avg: [0-9.]+, [0-9.]+, ([0-9.])+/i)
|
189
|
+
@topdata[:load] = $1.to_f
|
190
|
+
elsif ln.match(/CPU usage: [0-9.]+% user, [0-9.]+% sys, ([0-9.]+)% idle/i)
|
191
|
+
@topdata[:cpu] = 1 - ($1.to_f / 100)
|
192
|
+
elsif mdat = ln.match(/PhysMem: ([0-9]+)([BKMGT]) wired, ([0-9]+)([BKMGT]) active, ([0-9]+)([BKMGT]) inactive, ([0-9]+)([BKMGT]) used, ([0-9]+)([BKMGT]) free/i)
|
193
|
+
wired = mdat[1].to_i * (1024 ** "BKMGT".index(mdat[2]))
|
194
|
+
active = mdat[3].to_i * (1024 ** "BKMGT".index(mdat[4]))
|
195
|
+
inactive = mdat[5].to_i * (1024 ** "BKMGT".index(mdat[6]))
|
196
|
+
used = mdat[7].to_i * (1024 ** "BKMGT".index(mdat[8]))
|
197
|
+
free = mdat[9].to_i * (1024 ** "BKMGT".index(mdat[10]))
|
198
|
+
@topdata[:memory] = (wired + active + used).to_f / (wired + active + used + inactive + free)
|
199
|
+
end
|
200
|
+
end
|
201
|
+
end
|
202
|
+
|
203
|
+
def darwin_cpu
|
204
|
+
darwin_top unless (Time.now.to_i - @topdata[:stamp]) < opts[:interval]
|
205
|
+
unless @topdata[:cpu]
|
206
|
+
alert 'cpu', :unknown, nil, "unable to get CPU stats from top"
|
207
|
+
return false
|
208
|
+
end
|
209
|
+
report_pct :cpu, @topdata[:cpu], "usage\n\n#{`ps -eo pcpu,pid,comm | sort -nrb -k1 | head -10`.chomp}"
|
210
|
+
end
|
211
|
+
|
212
|
+
def darwin_load
|
213
|
+
darwin_top unless (Time.now.to_i - @topdata[:stamp]) < opts[:interval]
|
214
|
+
unless @topdata[:load]
|
215
|
+
alert 'load', :unknown, nil, "unable to get load ave from top"
|
216
|
+
return false
|
217
|
+
end
|
218
|
+
metric = @topdata[:load] / @cores
|
219
|
+
if metric > @limits[:load][:critical]
|
220
|
+
alert "load", :critical, metric, "1-minute load average per core is #{metric}"
|
221
|
+
elsif metric > @limits[:load][:warning]
|
222
|
+
alert "load", :warning, metric, "1-minute load average per core is #{metric}"
|
223
|
+
else
|
224
|
+
alert "load", :ok, metric, "1-minute load average per core is #{metric}"
|
225
|
+
end
|
226
|
+
end
|
227
|
+
|
228
|
+
def darwin_memory
|
229
|
+
darwin_top unless (Time.now.to_i - @topdata[:stamp]) < opts[:interval]
|
230
|
+
unless @topdata[:memory]
|
231
|
+
alert 'memory', :unknown, nil, "unable to get memory data from top"
|
232
|
+
return false
|
233
|
+
end
|
234
|
+
report_pct :memory, @topdata[:memory], "usage\n\n#{`ps -eo pmem,pid,comm | sort -nrb -k1 | head -10`.chomp}"
|
235
|
+
end
|
236
|
+
|
237
|
+
def disk
|
238
|
+
`df -P`.split(/\n/).each do |r|
|
239
|
+
f = r.split(/\s+/)
|
240
|
+
next unless f[0] =~ /^\//
|
241
|
+
next if f[0] == 'Filesystem'
|
242
|
+
x = f[4].to_f/100
|
243
|
+
|
244
|
+
if x > @limits[:disk][:critical]
|
245
|
+
alert "disk #{f[5]}", :critical, x, "#{f[4]} used"
|
246
|
+
elsif x > @limits[:disk][:warning]
|
247
|
+
alert "disk #{f[5]}", :warning, x, "#{f[4]} used"
|
248
|
+
else
|
249
|
+
alert "disk #{f[5]}", :ok, x, "#{f[4]} used"
|
250
|
+
end
|
251
|
+
end
|
252
|
+
end
|
253
|
+
|
254
|
+
def tick
|
255
|
+
if @cpu_enabled
|
256
|
+
@cpu.call
|
257
|
+
end
|
258
|
+
if @memory_enabled
|
259
|
+
@memory.call
|
260
|
+
end
|
261
|
+
if @disk_enabled
|
262
|
+
@disk.call
|
263
|
+
end
|
264
|
+
if @load_enabled
|
265
|
+
@load.call
|
266
|
+
end
|
267
|
+
end
|
268
|
+
end
|
269
|
+
|
270
|
+
Riemann::Tools::Health.run
|
data/bin/riemann-net
ADDED
@@ -0,0 +1,101 @@
|
|
1
|
+
#!/usr/bin/env ruby
|
2
|
+
|
3
|
+
# Gathers net statistics and submits them to Riemann.
|
4
|
+
|
5
|
+
require File.expand_path('../../lib/riemann/tools', __FILE__)
|
6
|
+
|
7
|
+
class Riemann::Tools::Net
|
8
|
+
include Riemann::Tools
|
9
|
+
|
10
|
+
opt :interfaces, "Interfaces to monitor", :type => :strings, :default => nil
|
11
|
+
opt :ignore_interfaces, "Interfaces to ignore", :type => :strings, :default =>['lo']
|
12
|
+
|
13
|
+
def initialize
|
14
|
+
@old_state = nil
|
15
|
+
@interfaces = opts[:interfaces].map(&:dup) if opts[:interfaces]
|
16
|
+
@ignore_interfaces = opts[:ignore_interfaces].map(&:dup)
|
17
|
+
end
|
18
|
+
|
19
|
+
def state
|
20
|
+
f = File.read('/proc/net/dev')
|
21
|
+
state = f.split("\n").inject({}) do |s, line|
|
22
|
+
if line =~ /\s*(\w+?):\s*([\s\d]+)\s*/
|
23
|
+
iface = $1
|
24
|
+
|
25
|
+
['rx bytes',
|
26
|
+
'rx packets',
|
27
|
+
'rx errs',
|
28
|
+
'rx drop',
|
29
|
+
'rx fifo',
|
30
|
+
'rx frame',
|
31
|
+
'rx compressed',
|
32
|
+
'rx multicast',
|
33
|
+
'tx bytes',
|
34
|
+
'tx packets',
|
35
|
+
'tx errs',
|
36
|
+
'tx drops',
|
37
|
+
'tx fifo',
|
38
|
+
'tx colls',
|
39
|
+
'tx carrier',
|
40
|
+
'tx compressed'].map do |service|
|
41
|
+
"#{iface} #{service}"
|
42
|
+
end.zip(
|
43
|
+
$2.split(/\s+/).map { |str| str.to_i }
|
44
|
+
).each do |service, value|
|
45
|
+
s[service] = value
|
46
|
+
end
|
47
|
+
end
|
48
|
+
|
49
|
+
s
|
50
|
+
end
|
51
|
+
|
52
|
+
# Filter interfaces
|
53
|
+
if is = @interfaces
|
54
|
+
state = state.select do |service, value|
|
55
|
+
is.include? service.split(' ').first
|
56
|
+
end
|
57
|
+
end
|
58
|
+
|
59
|
+
state = state.reject do |service, value|
|
60
|
+
@ignore_interfaces.include? service.split(' ').first
|
61
|
+
end
|
62
|
+
|
63
|
+
state
|
64
|
+
end
|
65
|
+
|
66
|
+
def tick
|
67
|
+
state = self.state
|
68
|
+
|
69
|
+
if @old_state
|
70
|
+
state.each do |service, metric|
|
71
|
+
delta = metric - @old_state[service]
|
72
|
+
svc_state = case service
|
73
|
+
when /drop$/
|
74
|
+
if metric > 0
|
75
|
+
'warning'
|
76
|
+
else
|
77
|
+
'ok'
|
78
|
+
end
|
79
|
+
when /errs$/
|
80
|
+
if metric > 0
|
81
|
+
'warning'
|
82
|
+
else
|
83
|
+
'ok'
|
84
|
+
end
|
85
|
+
else
|
86
|
+
'ok'
|
87
|
+
end
|
88
|
+
|
89
|
+
report(
|
90
|
+
:service => service.dup,
|
91
|
+
:metric => (delta.to_f / opts[:interval]),
|
92
|
+
:state => svc_state
|
93
|
+
)
|
94
|
+
end
|
95
|
+
end
|
96
|
+
|
97
|
+
@old_state = state
|
98
|
+
end
|
99
|
+
end
|
100
|
+
|
101
|
+
Riemann::Tools::Net.run
|
data/bin/riemann-proc
ADDED
@@ -0,0 +1,48 @@
|
|
1
|
+
#!/usr/bin/env ruby
|
2
|
+
|
3
|
+
# Reports running process count to riemann.
|
4
|
+
|
5
|
+
require File.expand_path('../../lib/riemann/tools', __FILE__)
|
6
|
+
|
7
|
+
class Riemann::Tools::Proc
|
8
|
+
include Riemann::Tools
|
9
|
+
|
10
|
+
opt :proc_regex, "regular expression that matches the process to be monitored", type: :string
|
11
|
+
opt :proc_min_critical, "running process count minimum", :default => 1
|
12
|
+
opt :proc_max_critical, "running process count maximum", :default => 1
|
13
|
+
|
14
|
+
def initialize
|
15
|
+
@limits = { :critical => { :min => opts[:proc_min_critical], :max => opts[:proc_max_critical] } }
|
16
|
+
|
17
|
+
abort "FATAL: specify a process regular expression, see --help for usage" unless opts[:proc_regex]
|
18
|
+
|
19
|
+
ostype = `uname -s`.chomp.downcase
|
20
|
+
puts "WARNING: OS '#{ostype}' not explicitly supported. Falling back to Linux" unless ostype == "linux"
|
21
|
+
@check = method :linux_proc
|
22
|
+
end
|
23
|
+
|
24
|
+
def alert(service, state, metric, description)
|
25
|
+
report(
|
26
|
+
:service => service.to_s,
|
27
|
+
:state => state.to_s,
|
28
|
+
:metric => metric.to_f,
|
29
|
+
:description => description
|
30
|
+
)
|
31
|
+
end
|
32
|
+
|
33
|
+
def linux_proc
|
34
|
+
process = opts[:proc_regex]
|
35
|
+
running = Integer(`ps axo args | grep #{process} | grep -v grep | grep -v riemann-proc | wc -l`)
|
36
|
+
if running > @limits[:critical][:max] or running < @limits[:critical][:min]
|
37
|
+
alert "proc #{process}", :critical, running, "process #{process} is running #{running} instances"
|
38
|
+
else
|
39
|
+
alert "proc #{process}", :ok, running, "process #{process} is running #{running} instances"
|
40
|
+
end
|
41
|
+
end
|
42
|
+
|
43
|
+
def tick
|
44
|
+
@check.call
|
45
|
+
end
|
46
|
+
end
|
47
|
+
|
48
|
+
Riemann::Tools::Proc.run
|
@@ -0,0 +1,132 @@
|
|
1
|
+
module Riemann
|
2
|
+
module Tools
|
3
|
+
require 'rubygems'
|
4
|
+
require 'trollop'
|
5
|
+
require 'riemann/client'
|
6
|
+
require 'timeout'
|
7
|
+
|
8
|
+
def self.included(base)
|
9
|
+
base.instance_eval do
|
10
|
+
def run
|
11
|
+
new.run
|
12
|
+
end
|
13
|
+
|
14
|
+
def opt(*a)
|
15
|
+
a.unshift :opt
|
16
|
+
@opts ||= []
|
17
|
+
@opts << a
|
18
|
+
end
|
19
|
+
|
20
|
+
def options
|
21
|
+
p = Trollop::Parser.new
|
22
|
+
@opts.each do |o|
|
23
|
+
p.send *o
|
24
|
+
end
|
25
|
+
Trollop::with_standard_exception_handling(p) do
|
26
|
+
p.parse ARGV
|
27
|
+
end
|
28
|
+
end
|
29
|
+
|
30
|
+
opt :host, "Riemann host", :default => '127.0.0.1'
|
31
|
+
opt :port, "Riemann port", :default => 5555
|
32
|
+
opt :event_host, "Event hostname", :type => String
|
33
|
+
opt :interval, "Seconds between updates", :default => 5
|
34
|
+
opt :tag, "Tag to add to events", :type => String, :multi => true
|
35
|
+
opt :ttl, "TTL for events", :type => Integer
|
36
|
+
opt :attribute, "Attribute to add to the event", :type => String, :multi => true
|
37
|
+
opt :timeout, "Timeout (in seconds) when waiting for acknowledgements", :default => 30
|
38
|
+
opt :tcp, "Use TCP transport instead of UDP (improves reliability, slight overhead.", :default => true
|
39
|
+
opt :daemon, "Run in background", :default => false
|
40
|
+
opt :logfile, "logfile path", :type => String, :default => '/tmp/riemann-tools.log'
|
41
|
+
opt :pidfile, "pidfile path", :type => String, :default => '/tmp/riemann-tools.pid'
|
42
|
+
end
|
43
|
+
end
|
44
|
+
|
45
|
+
# Returns parsed options (cached) from command line.
|
46
|
+
def options
|
47
|
+
@options ||= self.class.options
|
48
|
+
end
|
49
|
+
alias :opts :options
|
50
|
+
|
51
|
+
def attributes
|
52
|
+
@attributes ||= Hash[options[:attribute].map do |attr|
|
53
|
+
k,v = attr.split(/=/)
|
54
|
+
if k and v
|
55
|
+
[k,v]
|
56
|
+
end
|
57
|
+
end]
|
58
|
+
end
|
59
|
+
|
60
|
+
def report(event)
|
61
|
+
if options[:tag]
|
62
|
+
# Work around a bug with beefcake which can't take frozen strings.
|
63
|
+
event[:tags] = options[:tag].map(&:dup)
|
64
|
+
end
|
65
|
+
|
66
|
+
event[:ttl] ||= (options[:ttl] || (options[:interval] * 2))
|
67
|
+
|
68
|
+
if options[:event_host]
|
69
|
+
event[:host] = options[:event_host].dup
|
70
|
+
end
|
71
|
+
|
72
|
+
event = event.merge(attributes)
|
73
|
+
|
74
|
+
begin
|
75
|
+
Timeout::timeout(options[:timeout]) do
|
76
|
+
riemann << event
|
77
|
+
end
|
78
|
+
rescue Timeout::Error
|
79
|
+
riemann.connect
|
80
|
+
end
|
81
|
+
end
|
82
|
+
|
83
|
+
def new_riemann_client
|
84
|
+
r = Riemann::Client.new(
|
85
|
+
:host => options[:host],
|
86
|
+
:port => options[:port]
|
87
|
+
)
|
88
|
+
if options[:tcp]
|
89
|
+
r.tcp
|
90
|
+
else
|
91
|
+
r
|
92
|
+
end
|
93
|
+
end
|
94
|
+
|
95
|
+
def riemann
|
96
|
+
@riemann ||= new_riemann_client
|
97
|
+
end
|
98
|
+
alias :r :riemann
|
99
|
+
|
100
|
+
def run
|
101
|
+
daemonize if options[:daemon]
|
102
|
+
t0 = Time.now
|
103
|
+
loop do
|
104
|
+
begin
|
105
|
+
tick
|
106
|
+
rescue => e
|
107
|
+
$stderr.puts "#{e.class} #{e}\n#{e.backtrace.join "\n"}"
|
108
|
+
end
|
109
|
+
|
110
|
+
# Sleep.
|
111
|
+
sleep(options[:interval] - ((Time.now - t0) % options[:interval]))
|
112
|
+
end
|
113
|
+
end
|
114
|
+
|
115
|
+
def daemonize
|
116
|
+
exit if fork
|
117
|
+
Process.setsid
|
118
|
+
$0 = self.class.name.downcase.gsub('::','_')
|
119
|
+
$stdout.reopen(opts[:logfile], 'w')
|
120
|
+
$stdout.sync = true
|
121
|
+
$stderr.reopen($stdout)
|
122
|
+
exit if fork
|
123
|
+
f = File.new(opts[:pidfile], 'w')
|
124
|
+
f.write Process.pid
|
125
|
+
f.close
|
126
|
+
end
|
127
|
+
|
128
|
+
def tick
|
129
|
+
end
|
130
|
+
|
131
|
+
end
|
132
|
+
end
|
metadata
ADDED
@@ -0,0 +1,86 @@
|
|
1
|
+
--- !ruby/object:Gem::Specification
|
2
|
+
name: riemann-tools-fb
|
3
|
+
version: !ruby/object:Gem::Version
|
4
|
+
version: 0.3.0
|
5
|
+
platform: ruby
|
6
|
+
authors:
|
7
|
+
- Fede Borgnia
|
8
|
+
autorequire:
|
9
|
+
bindir: bin
|
10
|
+
cert_chain: []
|
11
|
+
date: 2015-01-21 00:00:00.000000000 Z
|
12
|
+
dependencies:
|
13
|
+
- !ruby/object:Gem::Dependency
|
14
|
+
name: riemann-client
|
15
|
+
requirement: !ruby/object:Gem::Requirement
|
16
|
+
requirements:
|
17
|
+
- - '>='
|
18
|
+
- !ruby/object:Gem::Version
|
19
|
+
version: 0.2.3
|
20
|
+
type: :runtime
|
21
|
+
prerelease: false
|
22
|
+
version_requirements: !ruby/object:Gem::Requirement
|
23
|
+
requirements:
|
24
|
+
- - '>='
|
25
|
+
- !ruby/object:Gem::Version
|
26
|
+
version: 0.2.3
|
27
|
+
- !ruby/object:Gem::Dependency
|
28
|
+
name: trollop
|
29
|
+
requirement: !ruby/object:Gem::Requirement
|
30
|
+
requirements:
|
31
|
+
- - '>='
|
32
|
+
- !ruby/object:Gem::Version
|
33
|
+
version: 1.16.2
|
34
|
+
type: :runtime
|
35
|
+
prerelease: false
|
36
|
+
version_requirements: !ruby/object:Gem::Requirement
|
37
|
+
requirements:
|
38
|
+
- - '>='
|
39
|
+
- !ruby/object:Gem::Version
|
40
|
+
version: 1.16.2
|
41
|
+
description: Utilities which submit events to Riemann.
|
42
|
+
email: fborgnia@gmail.com
|
43
|
+
executables:
|
44
|
+
- riemann-bench
|
45
|
+
- riemann-diskstats
|
46
|
+
- riemann-proc
|
47
|
+
- riemann-health
|
48
|
+
- riemann-fd
|
49
|
+
- riemann-net
|
50
|
+
extensions: []
|
51
|
+
extra_rdoc_files: []
|
52
|
+
files:
|
53
|
+
- lib/riemann/tools.rb
|
54
|
+
- bin/riemann-bench
|
55
|
+
- bin/riemann-diskstats
|
56
|
+
- bin/riemann-fd
|
57
|
+
- bin/riemann-health
|
58
|
+
- bin/riemann-net
|
59
|
+
- bin/riemann-proc
|
60
|
+
- LICENSE
|
61
|
+
- README.markdown
|
62
|
+
homepage: https://github.com/fborgnia/riemann-tools
|
63
|
+
licenses:
|
64
|
+
- MIT
|
65
|
+
metadata: {}
|
66
|
+
post_install_message:
|
67
|
+
rdoc_options: []
|
68
|
+
require_paths:
|
69
|
+
- lib
|
70
|
+
required_ruby_version: !ruby/object:Gem::Requirement
|
71
|
+
requirements:
|
72
|
+
- - '>='
|
73
|
+
- !ruby/object:Gem::Version
|
74
|
+
version: 1.9.3
|
75
|
+
required_rubygems_version: !ruby/object:Gem::Requirement
|
76
|
+
requirements:
|
77
|
+
- - '>='
|
78
|
+
- !ruby/object:Gem::Version
|
79
|
+
version: '0'
|
80
|
+
requirements: []
|
81
|
+
rubyforge_project: riemann-tools
|
82
|
+
rubygems_version: 2.0.3
|
83
|
+
signing_key:
|
84
|
+
specification_version: 4
|
85
|
+
summary: Utilities which submit events to Riemann.
|
86
|
+
test_files: []
|