| Index: trunk/debs/ganglios/debian/changelog |
| — | — | @@ -1,3 +1,9 @@ |
| | 2 | +ganglios (1.1) stable; urgency=low |
| | 3 | + |
| | 4 | + * changed listXMLSources to read the list from /etc/ganglia/gmetad.conf |
| | 5 | + |
| | 6 | + -- Ben Hartshorne <bhartshorne@wikimedia.org> Mon, 3 Oct 2011 16:35:00 -0700 |
| | 7 | + |
| 2 | 8 | ganglios (1.0-1) stable; urgency=low |
| 3 | 9 | |
| 4 | 10 | * Initial release |
| Index: trunk/debs/ganglios/src/ganglia_parser |
| — | — | @@ -102,39 +102,20 @@ |
| 103 | 103 | |
| 104 | 104 | def listXMLSources(): |
| 105 | 105 | """ returns a list of hosts to fetch ganglia stats from (the gmond |
| 106 | | - collector nodes). In a small network, this will likely be a single host. |
| 107 | | - You could simply hardcode the list if there is no coherent naming scheme. |
| 108 | | - In our network, the hosts are named 'nannybot1', 'nannybot2', etc. This |
| 109 | | - function polls nannybots of increasing number until it finds 3 that don't |
| 110 | | - respond and considers itself done.""" |
| 111 | | - i = 0 |
| 112 | | - missed = 0 |
| | 106 | + collector nodes). Pulls this list from the gmetad.conf. This will only |
| | 107 | + work when the ganglia_parser script is run on the same host as the ganglia |
| | 108 | + web UI and aggregator host.""" |
| 113 | 109 | nannybots = [] |
| 114 | 110 | logger.info("Retrieving list of nannybots...") |
| 115 | | - while True: |
| 116 | | - try: |
| 117 | | - nannybot_addr = socket.gethostbyname('nannybot%d.lindenlab.com' % i) |
| 118 | | - s = socket.socket(socket.AF_INET, socket.SOCK_STREAM) |
| 119 | | - s.settimeout(15.0) |
| 120 | | - s.connect((nannybot_addr, 8649)) |
| 121 | | - s.close() |
| 122 | | - nannybots.append('nannybot%d.lindenlab.com' % i) |
| 123 | | - logger.info('Checking nannybot%d.lindenlab.com: [OK]' % i) |
| 124 | | - missed = 0 |
| 125 | | - except (socket.gaierror): |
| 126 | | - logger.info('Checking nannybot%d.lindenlab.com: [FAILED]' % i) |
| 127 | | - missed += 1 |
| 128 | | - if missed > 3: |
| 129 | | - break |
| 130 | | - except (socket.error, socket.timeout), e: |
| 131 | | - logger.warning('Checking nannybot%d.lindenlab.com: [FAILED] with error %s' % (i, e)) |
| 132 | | - # a connection error (rather than non-existent) should not count towards |
| 133 | | - # the three missing that mark the end of the nannybots. |
| 134 | | - except Exception, e: |
| 135 | | - logger.critical('Caught unexpected exception while checking nannybot%d: %s' % (i, e)) |
| 136 | | - logger.critical('EXITING...') |
| 137 | | - raise e |
| 138 | | - i += 1 |
| | 111 | + # so long as ganglios is running on the same host as the ganglia web ui, it |
| | 112 | + # can use ganglia's gmetad.conf to get the list of sources. |
| | 113 | + gmetadconf = open('/etc/ganglia/gmetad.conf') |
| | 114 | + datasourcere = re.compile('^data_source "(?P<name>[^"]*)" (?P<hostlist>.*)') |
| | 115 | + for line in gmetadconf.readlines(): |
| | 116 | + match = datasourcere.match(line) |
| | 117 | + if match: |
| | 118 | + for host in match.group(2).split(): |
| | 119 | + nannybots.append(host) |
| 139 | 120 | |
| 140 | 121 | return nannybots |
| 141 | 122 | |