Always convert num_inconsistencies to type int, add cli args, fix kazoo usage
This commit is contained in:
Björn Busse
parent
1d70df9e47
commit
5af5afe57e
@ -16,10 +16,6 @@
|
||||
# * Add an option to relay all metrics
|
||||
#
|
||||
# * Add hdfs/hbase binaries to container
|
||||
#
|
||||
# * Fix command line argument handling
|
||||
#
|
||||
# * Add loglevel and other missing args
|
||||
|
||||
from __future__ import absolute_import
|
||||
from __future__ import division
|
||||
@ -31,7 +27,7 @@ from bs4 import BeautifulSoup
|
||||
from flatten_json import flatten
|
||||
import io
|
||||
import json
|
||||
from kazoo import client as kz_client
|
||||
from kazoo.client import KazooClient
|
||||
import logging
|
||||
import os
|
||||
from prometheus_client import start_http_server, Summary
|
||||
@ -92,9 +88,10 @@ class zk():
|
||||
zk_client = ""
|
||||
|
||||
@classmethod
|
||||
def main(self, address, timeout=5):
|
||||
def main(self, address_list, use_tls, timeout=5):
|
||||
|
||||
zk_client = kz_client.KazooClient(address)
|
||||
addresses = ','.join(address_list)
|
||||
zk_client = KazooClient(addresses, use_ssl=use_tls, read_only=True)
|
||||
|
||||
try:
|
||||
zk_client.start(timeout)
|
||||
@ -301,8 +298,11 @@ class hbase_exporter():
|
||||
self.check_health()
|
||||
|
||||
|
||||
def check_health():
|
||||
if self.num_inconsistencies > 0:
|
||||
def check_health(self):
|
||||
if self.num_inconsistencies == -1:
|
||||
prom_hbase_healthy.set(0)
|
||||
return False
|
||||
elif self.num_inconsistencies > 0:
|
||||
prom_hbase_healthy.set(0)
|
||||
return False
|
||||
|
||||
@ -399,7 +399,7 @@ class hbase_exporter():
|
||||
|
||||
re_status = re.compile(r'^Status:\s*(.+?)\s*$')
|
||||
re_inconsistencies = re.compile(r'^\s*(\d+)\s+inconsistencies\s+detected\.?\s*$')
|
||||
num_inconsistencies = None
|
||||
self.num_inconsistencies = None
|
||||
hbck_status = None
|
||||
|
||||
logging.info("HBase: Running hbck consistency check")
|
||||
@ -409,13 +409,14 @@ class hbase_exporter():
|
||||
|
||||
if p.returncode != 0:
|
||||
logging.info("Failed to run hbck (%d)" % (p.returncode))
|
||||
self.num_inconsistencies = -1
|
||||
return False
|
||||
|
||||
for line in output:
|
||||
match = re_inconsistencies.match(line)
|
||||
|
||||
if match:
|
||||
num_inconsistencies = match.group(1)
|
||||
self.num_inconsistencies = match.group(1)
|
||||
logging.info('Number of inconsistencies: %s', hbck_status)
|
||||
continue
|
||||
|
||||
@ -428,18 +429,18 @@ class hbase_exporter():
|
||||
|
||||
if hbck_status is None:
|
||||
logging.info('Failed to find hbck status result')
|
||||
if num_inconsistencies is None:
|
||||
if self.num_inconsistencies is None:
|
||||
logging.info('Failed to find number of inconsistencies')
|
||||
self.num_inconsistencies = -1
|
||||
|
||||
if num_inconsistencies != None:
|
||||
num_inconsistencies = int(num_inconsistencies)
|
||||
if self.num_inconsistencies != None:
|
||||
self.num_inconsistencies = int(self.num_inconsistencies)
|
||||
|
||||
if not isinstance(num_inconsistencies, int):
|
||||
if not isinstance(self.num_inconsistencies, int):
|
||||
logging.info('Error: Non-integer detected for the number of inconsistencies')
|
||||
self.num_inconsistencies = -1
|
||||
return False
|
||||
|
||||
self.num_inconsistencies = num_inconsistencies
|
||||
|
||||
|
||||
@staticmethod
|
||||
def hbaseui_parse_table(table):
|
||||
@ -478,8 +479,9 @@ if __name__ == '__main__':
|
||||
|
||||
parser = argparse.ArgumentParser( description="")
|
||||
parser.add_argument('--hbase-master', dest='hbase_master', action='append', help="HBase master address, can be specified multiple times", type=str, default=hbase_master_default_address)
|
||||
parser.add_argument('--hdfs-namenodes', dest='hdfs_namenode', action='append', help="HDFS namenode address, can be specified multiple times", type=str, default=hdfs_namenode_default_address)
|
||||
parser.add_argument('--zookeeper-server-address', dest='zookeeper_server', action='append', help="ZooKeeper server address, can be specified multiple times", type=str)
|
||||
parser.add_argument('--hdfs-namenode', dest='hdfs_namenode', action='append', help="HDFS namenode address, can be specified multiple times", type=str, default=hdfs_namenode_default_address)
|
||||
parser.add_argument('--zookeeper-server-address', dest='zk_server', action='append', help="ZooKeeper server address, can be specified multiple times", type=str)
|
||||
parser.add_argument('--zookeeper-use-tls', dest='zk_use_tls', help="Use TLS when connecting to ZooKeeper", type=bool, default=False)
|
||||
parser.add_argument('--prometheus-exporter-port', dest='prom_http_port', help="Listen port for Prometheus export", type=int, default=9010)
|
||||
parser.add_argument('--logfile', dest='logfile', help="Path to optional logfile", type=str)
|
||||
parser.add_argument('--loglevel', dest='loglevel', help="Loglevel, default: INFO", type=str, default='INFO')
|
||||
@ -488,16 +490,17 @@ if __name__ == '__main__':
|
||||
prom_http_port = args.prom_http_port
|
||||
logfile = args.logfile
|
||||
loglevel = args.loglevel
|
||||
zookeeper_server = args.zookeeper_server
|
||||
zk_server = args.zk_server
|
||||
zk_use_tls = args.zk_use_tls
|
||||
hbase_master = args.hbase_master
|
||||
hdfs_namenodes = args.hdfs_namenode
|
||||
del locals()['args']
|
||||
|
||||
nzookeeper_server = len(zookeeper_server)
|
||||
prom_zookeeper_num.set(nzookeeper_server)
|
||||
nzk_server = len(zk_server)
|
||||
prom_zookeeper_num.set(nzk_server)
|
||||
|
||||
# Optional File Logging
|
||||
if 'logfile' is not None:
|
||||
if logfile:
|
||||
tlog = logfile.rsplit('/', 1)
|
||||
logpath = tlog[0]
|
||||
logfile = tlog[1]
|
||||
@ -545,16 +548,8 @@ if __name__ == '__main__':
|
||||
|
||||
# Try to connect to one of the known servers
|
||||
while not r:
|
||||
for zk_address in zookeeper_server:
|
||||
nzk += 1
|
||||
logging.info("ZooKeeper: Trying to connect to "
|
||||
+ zk_address
|
||||
+ ' (' + str(nzk) + '/' + str(nzookeeper_server) + ')')
|
||||
|
||||
r = zk.main(zk_address)
|
||||
if r:
|
||||
break
|
||||
time.sleep(zk_reconnect_interval_s)
|
||||
r = zk.main(zk_server, zk_use_tls)
|
||||
time.sleep(zk_reconnect_interval_s)
|
||||
|
||||
if cluster_is_kerberized:
|
||||
znode_hbase = "/hbase"
|
||||
@ -578,6 +573,8 @@ if __name__ == '__main__':
|
||||
|
||||
jmx_query().main(hdfs_namenodes)
|
||||
hbase_exporter().main(hbase_master)
|
||||
#prom_zookeeper_num_live.set(nzookeeper_live)
|
||||
#prom_zookeeper_num_dead.set(nzk_server - nzookeeper_live)
|
||||
|
||||
nruns += 1
|
||||
|
||||
|
||||
Loading…
Reference in New Issue
Block a user