added extra metrics + no crash when scraped node has no disks

MANIFEST.in

@@ -0,0 +1,2 @@
+include prometheus-swift-exporter.service
+include post.sh

README.md

@@ -12,7 +12,7 @@ Install prometheus_client:
 pip install prometheus_client
 ```
 
-## Installation
+## Manual Installation
 
 ```
 # Copy example config in place, edit to your needs
@@ -42,6 +42,13 @@ Or to run interactively:
 
 Configuration options are documented in prometheus-swift-exporter.yaml shipped with this project
 
+# To build rpm
+
+```
+python setup.py  bdist_rpm --post-install post.sh
+```
+
+This installs the systemd service file but not the prometheus-swift-exporter.yaml
 # FAQ
 
 ## Why hardcode swift host list?

post.sh

@@ -0,0 +1,4 @@
+#!/bin/bash
+
+touch /etc/default/prometheus-swift-exporter
+/usr/bin/systemctl enable prometheus-swift-exporter.service

prometheus-swift-exporter

@@ -26,10 +26,12 @@ from os import path
 import traceback
 import urlparse
 import requests
+import json as simplejson
 from BaseHTTPServer import BaseHTTPRequestHandler
 from BaseHTTPServer import HTTPServer
 from SocketServer import ForkingMixIn
-from prometheus_client import CollectorRegistry, generate_latest, Gauge, CONTENT_TYPE_LATEST
+from prometheus_client import CollectorRegistry, generate_latest, \
+                              Gauge, CONTENT_TYPE_LATEST
 
 import logging
 import logging.handlers
@@ -40,88 +42,142 @@ log = logging.getLogger('poe-logger')
 class Swift():
     def __init__(self):
         self.registry = CollectorRegistry()
-        self.baseurl = 'http://{{}}:{}/recon/{{}}'.format(config.get('swift_port','6000'))
+        self.baseurl = 'http://{{}}:{}/recon/{{}}'.format(
+            config.get('swift_port', '9002'))
         self.swift_hosts = config.get('swift_hosts', [])
 
+    def gen_get(self, h, p):
+        r = requests.get(self.baseurl.format(h, 'diskusage'))
+        if r.status_code != 200:
+            raise requests.exceptions.RequestException
+        return r
+
     def gen_up_stats(self):
         labels = ['cloud', 'hostname']
         swift_up = Gauge('swift_host_up', 'Swift host reachability',
                          labels, registry=self.registry)
         for h in self.swift_hosts:
             try:
-                requests.get(self.baseurl.format(h, 'diskusage'))
+                r = self.gen_get(h, 'diskusage')
                 swift_up.labels(config['cloud'], h).set(1)
             except requests.exceptions.RequestException:
                 swift_up.labels(config['cloud'], h).set(0)
 
     def gen_disk_usage_stats(self):
         labels = ['cloud', 'hostname', 'device', 'type']
-        swift_disk = Gauge('swift_disk_usage_bytes', 'Swift disk usage in bytes',
-                           labels, registry=self.registry)
+        swift_disk = Gauge(
+            'swift_disk_usage_bytes',
+            'Swift disk usage in bytes',
+            labels,
+            registry=self.registry)
         for h in self.swift_hosts:
             try:
-                r = requests.get(self.baseurl.format(h, 'diskusage'))
-            except requests.exceptions.RequestException:
+                r = self.gen_get(h, 'diskusage')
+            except requests.exceptions.RequestException as ValueError:
                 continue
             for disk in r.json():
-                if not all([disk.get(i, False) for i in ['size', 'used', 'device']]):
+                if not all([disk.get(i, False)
+                            for i in ['size', 'used', 'device']]):
                     continue
-                swift_disk.labels(config['cloud'], h, disk['device'], 'size').set(int(disk['size']))
-                swift_disk.labels(config['cloud'], h, disk['device'], 'used').set(int(disk['used']))
+                swift_disk.labels(config['cloud'],
+                                  h, disk['device'],
+                                  'size').set(int(disk['size']))
+                swift_disk.labels(config['cloud'],
+                                  h, disk['device'],
+                                  'used').set(int(disk['used']))
 
     def gen_quarantine_stats(self):
         labels = ['cloud', 'hostname', 'ring']
-        swift_quarantine = Gauge('swift_quarantined_objects', 'Number of quarantined objects',
-                                 labels, registry=self.registry)
+        swift_quarantine = Gauge(
+            'swift_quarantined_objects',
+            'Number of quarantined objects',
+            labels,
+            registry=self.registry)
         for h in self.swift_hosts:
             try:
-                r = requests.get(self.baseurl.format(h, 'quarantined'))
+                r = self.gen_get(h, 'quarantined')
             except requests.exceptions.RequestException:
                 continue
             for ring in ['accounts', 'objects', 'containers']:
-                swift_quarantine.labels(config['cloud'], h, ring).set(r.json().get(ring))
+                if (isinstance(r.json(), dict)):
+                    swift_quarantine.labels(
+                        config['cloud'], h, ring).set(
+                        r.json().get(ring))
+
+    def gen_unmounted_stats(self):
+        labels = ['cloud', 'hostname', 'device']
+        swift_unmounted = Gauge('swift_umounted_disks', 'Disk is unmounted',
+                                labels, registry=self.registry)
+        for h in self.swift_hosts:
+            try:
+                r = self.gen_get(h, 'unmounted')
+            except requests.exceptions.RequestException:
+                continue
+            for disk in r.json():
+                if disk.get('mounted', True) is False:
+                    swift_unmounted.labels(config['cloud'],
+                                           h, disk['device']).set(1)
 
     def gen_replication_stats(self):
         labels = ['cloud', 'hostname', 'ring', 'type']
-        swift_repl = Gauge('swift_replication_stats', 'Swift replication stats', labels, registry=self.registry)
+        swift_repl = Gauge(
+            'swift_replication_stats',
+            'Swift replication stats',
+            labels,
+            registry=self.registry)
         labels = ['cloud', 'hostname', 'ring']
-        swift_repl_duration = Gauge('swift_replication_duration_seconds', 'Swift replication duration in seconds',
-                                    labels, registry=self.registry)
+        swift_repl_duration = Gauge(
+            'swift_replication_duration_seconds',
+            'Swift replication duration in seconds',
+            labels,
+            registry=self.registry)
         for h in self.swift_hosts:
             metrics = ['attempted', 'diff', 'diff_capped', 'empty',
                        'failure', 'hashmatch', 'no_change', 'remote_merge',
                        'remove', 'rsync', 'success', 'ts_repl']
             # Object replication is special
             try:
-                r = requests.get(self.baseurl.format(h, 'replication/object'))
+                r = self.gen_get(h, 'replication/object')
             except requests.exceptions.RequestException:
                 continue
             try:
-                swift_repl_duration.labels(config['cloud'], h, 'object').set(r.json()['object_replication_time'])
+                swift_repl_duration.labels(
+                    config['cloud'], h, 'object').set(
+                    r.json()['object_replication_time'])
             except TypeError:
-                print(traceback.format_exc())
+                continue
 
             for ring in ['account', 'container']:
                 try:
-                    r = requests.get(self.baseurl.format(h, 'replication/' + ring))
+                    r = self.gen_get(h, 'replication/' + ring)
                 except requests.exceptions.RequestException:
                     continue
                 try:
-                    swift_repl_duration.labels(config['cloud'], h, ring).set(r.json()['replication_time'])
+                    if (r.json()['replication_time']):
+                        swift_repl_duration.labels(
+                            config['cloud'], h, ring).set(
+                            r.json()['replication_time'])
                 except TypeError:
                     print(traceback.format_exc())
 
-                for metric in metrics:
-                    try:
-                        swift_repl.labels(config['cloud'], h, ring, metric).set(r.json()['replication_stats'][metric])
-                    except TypeError:
-                        print(traceback.format_exc())
+                if r.json().get('replication_stats'):
+                    for metric in metrics:
+                        try:
+                            swift_repl.labels(
+                                config['cloud'], h, ring, metric).set(
+                                r.json()['replication_stats'][metric])
+                        except TypeError:
+                            print("M", metric)
+                            print("C", config['cloud'], h, ring, metric)
+                            print("R", r.json()['replication_stats'])
+                            print(traceback.format_exc())
 
     def get_stats(self):
         self.gen_up_stats()
         self.gen_disk_usage_stats()
         self.gen_quarantine_stats()
         self.gen_replication_stats()
+        self.gen_unmounted_stats()
         return generate_latest(self.registry)
 
 
@@ -144,7 +200,7 @@ class OpenstackSwiftExporterHandler(BaseHTTPRequestHandler):
                 self.send_header('Content-Type', CONTENT_TYPE_LATEST)
                 self.end_headers()
                 self.wfile.write(output)
-            except:
+            except BaseException:
                 self.send_response(500)
                 self.end_headers()
                 self.wfile.write(traceback.format_exc())
@@ -168,15 +224,16 @@ def handler(*args, **kwargs):
 
 
 if __name__ == '__main__':
-    parser = argparse.ArgumentParser(usage=__doc__,
-                                     description='Prometheus OpenStack Swift exporter',
-                                     formatter_class=argparse.RawTextHelpFormatter)
+    parser = argparse.ArgumentParser(
+        usage=__doc__,
+        description='Prometheus OpenStack Swift exporter',
+        formatter_class=argparse.RawTextHelpFormatter)
     parser.add_argument('config_file', nargs='?',
                         help='Configuration file path',
                         default='/etc/prometheus/prometheus-swift-exporter.yaml',
                         type=argparse.FileType('r'))
     args = parser.parse_args()
-    log.setLevel(logging.DEBUG)
+    # log.setLevel(logging.DEBUG)
     for logsock in ('/dev/log', '/var/run/syslog'):
         if path.exists(logsock):
             log.addHandler(logging.handlers.SysLogHandler(address=logsock))

prometheus-swift-exporter.service

@@ -4,7 +4,7 @@ After=network.target
 
 [Service]
 EnvironmentFile=/etc/default/prometheus-swift-exporter
-ExecStart=/opt/prometheus-swift-exporter/prometheus-swift-exporter /etc/prometheus/prometheus-swift-exporter.yaml
+ExecStart=/usr/bin/prometheus-swift-exporter
 KillMode=process
 
 [Install]

setup.py

@@ -12,10 +12,11 @@ def read(fname):
     description="Exposes high level OpenStack Swift metrics to Prometheus.",
     license="GPLv3",
     keywords=["prometheus", "openstack", "swift", "exporter"],
-    url="https://github.com/ilanddev/prometheus-swift-exporter",
+    url="https://github.com/vorsprung/prometheus-swift-exporter",
     scripts=["prometheus-swift-exporter"],
+    data_files=[("/etc/systemd/system",["prometheus-swift-exporter.service"])],
     install_requires=["prometheus_client"],
-    long_description=read('README.md'),
+    long_description="Exposes high level OpenStack Swift metrics to Prometheus.",
     classifiers=[
         "Development Status :: 4 - Beta",
         "Topic :: System :: Networking :: Monitoring",