Files
patroni/patroni/postgresql/citus.py
T

420 lines
19 KiB
Python

import logging
import re
import time
from threading import Condition, Event, Thread
from urllib.parse import urlparse
from typing import Any, Dict, List, Optional, Union, Tuple, TYPE_CHECKING
from .connection import Connection
from ..dcs import CITUS_COORDINATOR_GROUP_ID, Cluster
from ..psycopg import connect, quote_ident
if TYPE_CHECKING: # pragma: no cover
from psycopg import Cursor
from psycopg2 import cursor
from . import Postgresql
CITUS_SLOT_NAME_RE = re.compile(r'^citus_shard_(move|split)_slot(_[1-9][0-9]*){2,3}$')
logger = logging.getLogger(__name__)
class PgDistNode(object):
"""Represents a single row in the `pg_dist_node` table"""
def __init__(self, group: int, host: str, port: int, event: str, nodeid: Optional[int] = None,
timeout: Optional[float] = None, cooldown: Optional[float] = None) -> None:
self.group = group
# A weird way of pausing client connections by adding the `-demoted` suffix to the hostname
self.host = host + ('-demoted' if event == 'before_demote' else '')
self.port = port
# Event that is trying to change or changed the given row.
# Possible values: before_demote, before_promote, after_promote.
self.event = event
self.nodeid = nodeid
# If transaction was started, we need to COMMIT/ROLLBACK before the deadline
self.timeout = timeout
self.cooldown = cooldown or 10000 # 10s by default
self.deadline: float = 0
# All changes in the pg_dist_node are serialized on the Patroni
# side by performing them from a thread. The thread, that is
# requested a change, sometimes needs to wait for a result.
# For example, we want to pause client connections before demoting
# the worker, and once it is done notify the calling thread.
self._event = Event()
def wait(self) -> None:
self._event.wait()
def wakeup(self) -> None:
self._event.set()
def __eq__(self, other: Any) -> bool:
return isinstance(other, PgDistNode) and self.event == other.event\
and self.host == other.host and self.port == other.port
def __ne__(self, other: Any) -> bool:
return not self == other
def __str__(self) -> str:
return ('PgDistNode(nodeid={0},group={1},host={2},port={3},event={4})'
.format(self.nodeid, self.group, self.host, self.port, self.event))
def __repr__(self) -> str:
return str(self)
class CitusHandler(Thread):
def __init__(self, postgresql: 'Postgresql', config: Optional[Dict[str, Union[str, int]]]) -> None:
super(CitusHandler, self).__init__()
self.daemon = True
self._postgresql = postgresql
self._config = config
self._connection = Connection()
self._pg_dist_node: Dict[int, PgDistNode] = {} # Cache of pg_dist_node: {groupid: PgDistNode()}
self._tasks: List[PgDistNode] = [] # Requests to change pg_dist_node, every task is a `PgDistNode`
self._in_flight: Optional[PgDistNode] = None # Reference to the `PgDistNode` being changed in a transaction
self._schedule_load_pg_dist_node = True # Flag that "pg_dist_node" should be queried from the database
self._condition = Condition() # protects _pg_dist_node, _tasks, _in_flight, and _schedule_load_pg_dist_node
self.schedule_cache_rebuild()
def is_enabled(self) -> bool:
return isinstance(self._config, dict)
def group(self) -> Optional[int]:
return int(self._config['group']) if isinstance(self._config, dict) else None
def is_coordinator(self) -> bool:
return self.is_enabled() and self.group() == CITUS_COORDINATOR_GROUP_ID
def is_worker(self) -> bool:
return self.is_enabled() and not self.is_coordinator()
def set_conn_kwargs(self, kwargs: Dict[str, Any]) -> None:
if isinstance(self._config, dict): # self.is_enabled():
kwargs.update({'dbname': self._config['database'],
'options': '-c statement_timeout=0 -c idle_in_transaction_session_timeout=0'})
self._connection.set_conn_kwargs(kwargs)
def schedule_cache_rebuild(self) -> None:
with self._condition:
self._schedule_load_pg_dist_node = True
def on_demote(self) -> None:
with self._condition:
self._pg_dist_node.clear()
self._tasks[:] = []
self._in_flight = None
def query(self, sql: str, *params: Any) -> Union['Cursor[Any]', 'cursor']:
try:
logger.debug('query(%s, %s)', sql, params)
cursor = self._connection.cursor()
cursor.execute(sql.encode('utf-8'), params or None)
return cursor
except Exception as e:
logger.error('Exception when executing query "%s", (%s): %r', sql, params, e)
self._connection.close()
with self._condition:
self._in_flight = None
self.schedule_cache_rebuild()
raise e
def load_pg_dist_node(self) -> bool:
"""Read from the `pg_dist_node` table and put it into the local cache"""
with self._condition:
if not self._schedule_load_pg_dist_node:
return True
self._schedule_load_pg_dist_node = False
try:
cursor = self.query("SELECT nodeid, groupid, nodename, nodeport, noderole"
" FROM pg_catalog.pg_dist_node WHERE noderole = 'primary'")
except Exception:
return False
with self._condition:
self._pg_dist_node = {r[1]: PgDistNode(r[1], r[2], r[3], 'after_promote', r[0]) for r in cursor}
return True
def sync_pg_dist_node(self, cluster: Cluster) -> None:
"""Maintain the `pg_dist_node` from the coordinator leader every heartbeat loop.
We can't always rely on REST API calls from worker nodes in order
to maintain `pg_dist_node`, therefore at least once per heartbeat
loop we make sure that workes registered in `self._pg_dist_node`
cache are matching the cluster view from DCS by creating tasks
the same way as it is done from the REST API."""
if not self.is_coordinator():
return
with self._condition:
if not self.is_alive():
self.start()
self.add_task('after_promote', CITUS_COORDINATOR_GROUP_ID, self._postgresql.connection_string)
for group, worker in cluster.workers.items():
leader = worker.leader
if leader and leader.conn_url\
and leader.data.get('role') in ('master', 'primary') and leader.data.get('state') == 'running':
self.add_task('after_promote', group, leader.conn_url)
def find_task_by_group(self, group: int) -> Optional[int]:
for i, task in enumerate(self._tasks):
if task.group == group:
return i
def pick_task(self) -> Tuple[Optional[int], Optional[PgDistNode]]:
"""Returns the tuple(i, task), where `i` - is the task index in the self._tasks list
Tasks are picked by following priorities:
1. If there is already a transaction in progress, pick a task
that that will change already affected worker primary.
2. If the coordinator address should be changed - pick a task
with group=0 (coordinators are always in group 0).
3. Pick a task that is the oldest (first from the self._tasks)"""
with self._condition:
if self._in_flight:
i = self.find_task_by_group(self._in_flight.group)
else:
while True:
i = self.find_task_by_group(CITUS_COORDINATOR_GROUP_ID) # set_coordinator
if i is None and self._tasks:
i = 0
if i is None:
break
task = self._tasks[i]
if task == self._pg_dist_node.get(task.group):
self._tasks.pop(i) # nothing to do because cached version of pg_dist_node already matches
else:
break
task = self._tasks[i] if i is not None else None
# When tasks are added it could happen that self._pg_dist_node
# wasn't ready (self._schedule_load_pg_dist_node is False)
# and hence the nodeid wasn't filled.
if task and task.group in self._pg_dist_node:
task.nodeid = self._pg_dist_node[task.group].nodeid
return i, task
def update_node(self, task: PgDistNode) -> None:
if task.nodeid is not None:
self.query('SELECT pg_catalog.citus_update_node(%s, %s, %s, true, %s)',
task.nodeid, task.host, task.port, task.cooldown)
elif task.event != 'before_demote':
row = self.query("SELECT pg_catalog.citus_add_node(%s, %s, %s, 'primary', 'default')",
task.host, task.port, task.group).fetchone()
if row is not None:
task.nodeid = row[0]
def process_task(self, task: PgDistNode) -> bool:
"""Updates a single row in `pg_dist_node` table, optionally in a transaction.
The transaction is started if we do a demote of the worker node or before promoting the other worker if
there is no transaction in progress. And, the transaction is committed when the switchover/failover completed.
.. note:
The maximum lifetime of the transaction in progress is controlled outside of this method.
.. note:
Read access to `self._in_flight` isn't protected because we know it can't be changed outside of our thread.
:param task: reference to a :class:`PgDistNode` object that represents a row to be updated/created.
:returns: `True` if the row was succesfully created/updated or transaction in progress
was committed as an indicator that the `self._pg_dist_node` cache should be updated,
or, if the new transaction was opened, this method returns `False`.
"""
if task.event == 'after_promote':
# The after_promote may happen without previous before_demote and/or
# before_promore. In this case we just call self.update_node() method.
# If there is a transaction in progress, it could be that it already did
# required changes and we can simply COMMIT.
if not self._in_flight or self._in_flight.host != task.host or self._in_flight.port != task.port:
self.update_node(task)
if self._in_flight:
self.query('COMMIT')
return True
else: # before_demote, before_promote
if task.timeout:
task.deadline = time.time() + task.timeout
if not self._in_flight:
self.query('BEGIN')
self.update_node(task)
return False
def process_tasks(self) -> None:
while True:
# Read access to `_in_flight` isn't protected because we know it can't be changed outside of our thread.
if not self._in_flight and not self.load_pg_dist_node():
break
i, task = self.pick_task()
if not task or i is None:
break
try:
update_cache = self.process_task(task)
except Exception as e:
logger.error('Exception when working with pg_dist_node: %r', e)
update_cache = None
with self._condition:
if self._tasks:
if update_cache:
self._pg_dist_node[task.group] = task
if update_cache is False: # an indicator that process_tasks has started a transaction
self._in_flight = task
else:
self._in_flight = None
if id(self._tasks[i]) == id(task):
self._tasks.pop(i)
task.wakeup()
def run(self) -> None:
while True:
try:
with self._condition:
if self._schedule_load_pg_dist_node:
timeout = -1
elif self._in_flight:
timeout = self._in_flight.deadline - time.time() if self._tasks else None
else:
timeout = -1 if self._tasks else None
if timeout is None or timeout > 0:
self._condition.wait(timeout)
elif self._in_flight:
logger.warning('Rolling back transaction. Last known status: %s', self._in_flight)
self.query('ROLLBACK')
self._in_flight = None
self.process_tasks()
except Exception:
logger.exception('run')
def _add_task(self, task: PgDistNode) -> bool:
with self._condition:
i = self.find_task_by_group(task.group)
# The `PgDistNode.timeout` == None is an indicator that it was scheduled from the sync_pg_dist_node().
if task.timeout is None:
# We don't want to override the already existing task created from REST API.
if i is not None and self._tasks[i].timeout is not None:
return False
# There is a little race condition with tasks created from REST API - the call made "before" the member
# key is updated in DCS. Therefore it is possible that :func:`sync_pg_dist_node` will try to create a
# task based on the outdated values of "state"/"role". To solve it we introduce an artificial timeout.
# Only when the timeout is reached new tasks could be scheduled from sync_pg_dist_node()
if self._in_flight and self._in_flight.group == task.group and self._in_flight.timeout is not None\
and self._in_flight.deadline > time.time():
return False
# Override already existing task for the same worker group
if i is not None:
if task != self._tasks[i]:
logger.debug('Overriding existing task: %s != %s', self._tasks[i], task)
self._tasks[i] = task
self._condition.notify()
return True
# Add the task to the list if Worker node state is different from the cached `pg_dist_node`
elif self._schedule_load_pg_dist_node or task != self._pg_dist_node.get(task.group)\
or self._in_flight and task.group == self._in_flight.group:
logger.debug('Adding the new task: %s', task)
self._tasks.append(task)
self._condition.notify()
return True
return False
def add_task(self, event: str, group: int, conn_url: str,
timeout: Optional[float] = None, cooldown: Optional[float] = None) -> Optional[PgDistNode]:
try:
r = urlparse(conn_url)
except Exception as e:
return logger.error('Failed to parse connection url %s: %r', conn_url, e)
host = r.hostname
if host:
port = r.port or 5432
task = PgDistNode(group, host, port, event, timeout=timeout, cooldown=cooldown)
return task if self._add_task(task) else None
def handle_event(self, cluster: Cluster, event: Dict[str, Any]) -> None:
if not self.is_alive():
return
worker = cluster.workers.get(event['group'])
if not (worker and worker.leader and worker.leader.name == event['leader'] and worker.leader.conn_url):
return
task = self.add_task(event['type'], event['group'],
worker.leader.conn_url,
event['timeout'], event['cooldown'] * 1000)
if task and event['type'] == 'before_demote':
task.wait()
def bootstrap(self) -> None:
if not isinstance(self._config, dict): # self.is_enabled()
return
conn_kwargs = self._postgresql.config.local_connect_kwargs
conn_kwargs['options'] = '-c synchronous_commit=local -c statement_timeout=0'
if self._config['database'] != self._postgresql.database:
conn = connect(**conn_kwargs)
try:
with conn.cursor() as cur:
cur.execute('CREATE DATABASE {0}'.format(
quote_ident(self._config['database'], conn)).encode('utf-8'))
finally:
conn.close()
conn_kwargs['dbname'] = self._config['database']
conn = connect(**conn_kwargs)
try:
with conn.cursor() as cur:
cur.execute('CREATE EXTENSION citus')
superuser = self._postgresql.config.superuser
params = {k: superuser[k] for k in ('password', 'sslcert', 'sslkey') if k in superuser}
if params:
cur.execute("INSERT INTO pg_catalog.pg_dist_authinfo VALUES"
"(0, pg_catalog.current_user(), %s)",
(self._postgresql.config.format_dsn(params),))
if self.is_coordinator():
r = urlparse(self._postgresql.connection_string)
cur.execute("SELECT pg_catalog.citus_set_coordinator_host(%s, %s, 'primary', 'default')",
(r.hostname, r.port or 5432))
finally:
conn.close()
def adjust_postgres_gucs(self, parameters: Dict[str, Any]) -> None:
if not self.is_enabled():
return
# citus extension must be on the first place in shared_preload_libraries
shared_preload_libraries = list(filter(
lambda el: el and el != 'citus',
[p.strip() for p in parameters.get('shared_preload_libraries', '').split(',')]))
parameters['shared_preload_libraries'] = ','.join(['citus'] + shared_preload_libraries)
# if not explicitly set Citus overrides max_prepared_transactions to max_connections*2
if parameters['max_prepared_transactions'] == 0:
parameters['max_prepared_transactions'] = parameters['max_connections'] * 2
# Resharding in Citus implemented using logical replication
parameters['wal_level'] = 'logical'
def ignore_replication_slot(self, slot: Dict[str, str]) -> bool:
if isinstance(self._config, dict) and self._postgresql.is_leader() and\
slot['type'] == 'logical' and slot['database'] == self._config['database']:
m = CITUS_SLOT_NAME_RE.match(slot['name'])
return bool(m and {'move': 'pgoutput', 'split': 'citus'}.get(m.group(1)) == slot['plugin'])
return False