def update_nrpe_config():
update_nagios_pgpass()
nrpe = NRPE()
user = nagios_username()
port = postgresql.port()
nrpe.add_check(shortname="pgsql", description="Check pgsql", check_cmd="check_pgsql -P {} -l {}".format(port, user))
if reactive.is_state("postgresql.replication.is_master"):
# TODO: These should be calcualted from the backup schedule,
# which is difficult since that is specified in crontab format.
warn_age = 172800
crit_age = 194400
backups_log = helpers.backups_log_path()
nrpe.add_check(
shortname="pgsql_backups",
description="Check pgsql backups",
check_cmd=("check_file_age -w {} -c {} -f {}" "".format(warn_age, crit_age, backups_log)),
)
else:
# Standbys don't do backups. We still generate a check though,
# to ensure alerts get through to monitoring after a failover.
nrpe.add_check(
shortname="pgsql_backups",
description="Check pgsql backups",
check_cmd=r"check_dummy 0 standby_does_not_backup",
)
nrpe.write()
reactive.remove_state("postgresql.nagios.needs_update")
def test_port(self, pgconf_path):
# Pull the configured port from postgresql.conf.
with tempfile.NamedTemporaryFile("w") as pgconf:
pgconf.write("# Some rubbish\n")
pgconf.write(" Port = 1234 # Picked by pg_createcluster(1)\n")
pgconf.flush()
pgconf_path.return_value = pgconf.name
self.assertEqual(postgresql.port(), 1234)
with tempfile.NamedTemporaryFile("w") as pgconf:
pgconf.write("port='1235'\n")
pgconf.write("# Some rubbish\n")
pgconf.flush()
pgconf_path.return_value = pgconf.name
self.assertEqual(postgresql.port(), 1235)
with tempfile.NamedTemporaryFile("w") as pgconf:
pgconf_path.return_value = pgconf.name
self.assertEqual(postgresql.port(), 5432) # Fallback to default.
def test_port(self, pgconf_path):
# Pull the configured port from postgresql.conf.
with tempfile.NamedTemporaryFile('w') as pgconf:
pgconf.write('# Some rubbish\n')
pgconf.write(' Port = 1234 # Picked by pg_createcluster(1)\n')
pgconf.flush()
pgconf_path.return_value = pgconf.name
self.assertEqual(postgresql.port(), 1234)
with tempfile.NamedTemporaryFile('w') as pgconf:
pgconf.write("port='1235'\n")
pgconf.write('# Some rubbish\n')
pgconf.flush()
pgconf_path.return_value = pgconf.name
self.assertEqual(postgresql.port(), 1235)
with tempfile.NamedTemporaryFile('w') as pgconf:
pgconf_path.return_value = pgconf.name
self.assertEqual(postgresql.port(), 5432) # Fallback to default.
def db_relation_common(rel):
"""Publish unit specific relation details."""
local = rel.local
if "database" not in local:
return # Not yet ready.
# Version number, allowing clients to adjust or block if their
# expectations are not met.
local["version"] = postgresql.version()
# Calculate the state of this unit. 'standalone' will disappear
# in a future version of this interface, as this state was
# only needed to deal with race conditions now solved by
# Juju leadership. We check for is_primary() rather than
# the postgresql.replication.is_master reactive state to
# publish the correct state when we are using manual replication
# (there might be multiple independent masters, possibly useful for
# sharding, or perhaps this is a multi master BDR setup).
if postgresql.is_primary():
if reactive.helpers.is_state("postgresql.replication.has_peers"):
local["state"] = "master"
else:
local["state"] = "standalone"
else:
local["state"] = "hot standby"
# Host is the private ip address, but this might change and
# become the address of an attached proxy or alternative peer
# if this unit is in maintenance.
local["host"] = hookenv.unit_private_ip()
# Port will be 5432, unless the user has overridden it or
# something very weird happened when the packages where installed.
local["port"] = str(postgresql.port())
# The list of remote units on this relation granted access.
# This is to avoid the race condition where a new client unit
# joins an existing client relation and sees valid credentials,
# before we have had a chance to grant it access.
local["allowed-units"] = " ".join(unit for unit, relinfo in rel.items() if "private-address" in relinfo)
def db_relation_common(rel):
"""Publish unit specific relation details."""
local = rel.local
if "database" not in local:
return # Not yet ready.
# Version number, allowing clients to adjust or block if their
# expectations are not met.
local["version"] = postgresql.version()
# Calculate the state of this unit. 'standalone' will disappear
# in a future version of this interface, as this state was
# only needed to deal with race conditions now solved by
# Juju leadership. We check for is_primary() rather than
# the postgresql.replication.is_master reactive state to
# publish the correct state when we are using manual replication
# (there might be multiple independent masters, possibly useful for
# sharding, or perhaps this is a multi master BDR setup).
if postgresql.is_primary():
if reactive.helpers.is_state("postgresql.replication.has_peers"):
local["state"] = "master"
else:
local["state"] = "standalone"
else:
local["state"] = "hot standby"
# Host is the private ip address, but this might change and
# become the address of an attached proxy or alternative peer
# if this unit is in maintenance.
local["host"] = ingress_address(local.relname, local.relid)
# Port will be 5432, unless the user has overridden it or
# something very weird happened when the packages where installed.
local["port"] = str(postgresql.port())
# The list of remote units on this relation granted access.
# This is to avoid the race condition where a new client unit
# joins an existing client relation and sees valid credentials,
# before we have had a chance to grant it access.
local["allowed-units"] = " ".join(unit for unit, relinfo in rel.items() if len(incoming_addresses(relinfo)) > 0)
# The list of IP address ranges on this relation granted access.
# This will replace allowed-units, which does not work with cross
# model ralations due to the anonymization of the external client.
local["allowed-subnets"] = ",".join(
sorted({r: True for r in chain(*[incoming_addresses(relinfo) for relinfo in rel.values()])}.keys())
)
# v2 protocol. Publish connection strings for this unit and its peers.
# Clients should use these connection strings in favour of the old
# host, port, database settings. A single proxy unit can thus
# publish several end points to clients.
master = replication.get_master()
if replication.is_master():
master_relinfo = local
else:
master_relinfo = rel.peers.get(master)
local["master"] = relinfo_to_cs(master_relinfo)
if rel.peers:
all_relinfo = rel.peers.values()
all_relinfo = list(rel.peers.values()) if rel.peers else []
all_relinfo.append(rel.local)
standbys = filter(
None,
[relinfo_to_cs(relinfo) for relinfo in all_relinfo if relinfo.unit != master],
)
local["standbys"] = "\n".join(sorted(standbys)) or None
def update_nrpe_config():
update_nagios_pgpass()
nrpe = NRPE()
user = nagios_username()
port = postgresql.port()
nrpe.add_check(
shortname="pgsql",
description="Check pgsql",
check_cmd="check_pgsql -P {} -l {}".format(port, user),
)
# copy the check script which will run cronned as postgres user
with open("scripts/find_latest_ready_wal.py") as fh:
check_script = fh.read()
check_script_path = "{}/{}".format(helpers.scripts_dir(), "find_latest_ready_wal.py")
helpers.write(check_script_path, check_script, mode=0o755)
# create an (empty) file with appropriate permissions for the above
check_output_path = "/var/lib/nagios/postgres-wal-max-age.txt"
if not os.path.exists(check_output_path):
helpers.write(check_output_path, b"0\n", mode=0o644, user="******", group="postgres")
# retrieve the threshold values from the charm config
config = hookenv.config()
check_warn_threshold = config["wal_archive_warn_threshold"] or 0
check_crit_threshold = config["wal_archive_crit_threshold"] or 0
check_cron_path = "/etc/cron.d/postgres-wal-archive-check"
if check_warn_threshold and check_crit_threshold:
# create the cron job to run the above
check_cron = "*/2 * * * * postgres {}".format(check_script_path)
helpers.write(check_cron_path, check_cron, mode=0o644)
# copy the nagios plugin which will check the cronned output
with open("scripts/check_latest_ready_wal.py") as fh:
check_script = fh.read()
check_script_path = "{}/{}".format("/usr/local/lib/nagios/plugins", "check_latest_ready_wal.py")
helpers.write(check_script_path, check_script, mode=0o755)
# write the nagios check definition
nrpe.add_check(
shortname="pgsql_stale_wal",
description="Check for stale WAL backups",
check_cmd="{} {} {}".format(check_script_path, check_warn_threshold, check_crit_threshold),
)
if reactive.is_state("postgresql.replication.is_master"):
# TODO: These should be calculated from the backup schedule,
# which is difficult since that is specified in crontab format.
warn_age = 172800
crit_age = 194400
backups_log = helpers.backups_log_path()
nrpe.add_check(
shortname="pgsql_backups",
description="Check pgsql backups",
check_cmd=("check_file_age -w {} -c {} -f {}" "".format(warn_age, crit_age, backups_log)),
)
else:
# Standbys don't do backups. We still generate a check though,
# to ensure alerts get through to monitoring after a failover.
nrpe.add_check(
shortname="pgsql_backups",
description="Check pgsql backups",
check_cmd=r"check_dummy 0 standby_does_not_backup",
)
nrpe.write()
reactive.remove_state("postgresql.nagios.needs_update")
def publish_replication_details():
peer = helpers.get_peer_relation()
if peer is not None:
peer.local["host"] = hookenv.unit_private_ip()
peer.local["port"] = str(postgresql.port())
peer.local["allowed-units"] = " ".join(sorted(peer.keys()))
