Patroni taking too long for failover

[BenchmarkingBuffalo]

Please, answer some short questions which should help us to understand your problem / question better?

• Which image of the operator are you using? registry.opensource.zalan.do/acid/postgres-operator:latest
• Where do you run it - cloud or metal? Bare Metal K8s
• Are you running Postgres Operator in production? no
• Type of issue? Question Hi there, I am trying out the postgres-operator to deploy a HA-postgres-cluster on kubernetes. I am using this manifest:

  apiVersion: "acid.zalan.do/v1"
  kind: postgresql
  metadata:
  name: acid-minimal-cluster
  namespace: default
  spec:
  teamId: "acid"
  volume:
  size: 1Gi
  numberOfInstances: 2
  users:
  postgres:  # database owner
  - superuser
  - createdb
  foo_user: []  # role for application foo
  databases:
  postgres: postgres  # dbname: owner
  preparedDatabases:
  bar: {}
  postgresql:
  version: "12"
  enableMasterLoadBalancer: true
  allowedSourceRanges:  # load balancers' source ranges for both master and replica services
  - 0.0.0.0/24
  patroni:
  ttl: 2
  loop_wait: 1
  retry_timeout: 0
  master_start_timeout: 0
  synchronous_mode: false
  synchronous_mode_strict: false
  maximum_lag_on_failover: 33554432

  I want to check, what happens, when I disconnect the network from the worker node the master is running on. What I saw, was this: Logs from former standby:

  
  2020-09-22 13:32:58,868 INFO: does not have lock
  2020-09-22 13:32:58,871 INFO: no action.  i am a secondary and i am following a leader
  2020-09-22 13:32:59,873 INFO: Lock owner: acid-minimal-cluster-0; I am acid-minimal-cluster-1
  2020-09-22 13:32:59,873 INFO: does not have lock
  2020-09-22 13:32:59,879 INFO: no action.  i am a secondary and i am following a leader
  2020-09-22 13:33:00,862 INFO: Lock owner: acid-minimal-cluster-0; I am acid-minimal-cluster-1
  2020-09-22 13:33:00,862 INFO: does not have lock
  2020-09-22 13:33:00,864 INFO: no action.  i am a secondary and i am following a leader
  2020-09-22 13:33:58,493 ERROR: Can not fetch local timeline and lsn from replication connection
  Traceback (most recent call last):
  File "/usr/local/lib/python3.6/dist-packages/patroni/postgresql/__init__.py", line 735, in get_replica_timeline
  with self.get_replication_connection_cursor(**self.config.local_replication_address) as cur:
  File "/usr/lib/python3.6/contextlib.py", line 81, in __enter__
  return next(self.gen)
  File "/usr/local/lib/python3.6/dist-packages/patroni/postgresql/__init__.py", line 730, in get_replication_connection_cursor
  with get_connection_cursor(**conn_kwargs) as cur:
  File "/usr/lib/python3.6/contextlib.py", line 81, in __enter__
  return next(self.gen)
  File "/usr/local/lib/python3.6/dist-packages/patroni/postgresql/connection.py", line 43, in get_connection_cursor
  with psycopg2.connect(**kwargs) as conn:
  File "/usr/lib/python3/dist-packages/psycopg2/__init__.py", line 127, in connect
  conn = _connect(dsn, connection_factory=connection_factory, **kwasync)
  psycopg2.OperationalError: timeout expired
  could not connect to server: Network is unreachable
      Is the server running on host "localhost" (::1) and accepting
      TCP/IP connections on port 5432?

2020-09-22 13:34:23,786 INFO: Lock owner: acid-minimal-cluster-0; I am acid-minimal-cluster-1 2020-09-22 13:34:23,786 INFO: does not have lock 2020-09-22 13:34:23,860 INFO: no action. i am a secondary and i am following a leader 2020-09-22 13:34:23,862 WARNING: Loop time exceeded, rescheduling immediately. 2020-09-22 13:34:25,386 WARNING: Request failed to acid-minimal-cluster-0: GET http://10.36.0.1:8008/patroni (HTTPConnectionPool(host='10.36.0.1', port=8008): Max retries exceeded with url: /patroni (Caused by NewConnectionError('<urllib3.connection.HTTPConnection object at 0x7fe1b8fe3828>: Failed to establish a new connection: [Errno 113] No route to host',))) 2020-09-22 13:34:25,520 WARNING: Could not activate Linux watchdog device: "Can't open watchdog device: [Errno 2] No such file or directory: '/dev/watchdog'" 2020-09-22 13:34:25,582 INFO: promoted self to leader by acquiring session lock 2020-09-22 13:34:25,584 WARNING: Loop time exceeded, rescheduling immediately. 2020-09-22 13:34:25,584 INFO: Lock owner: acid-minimal-cluster-1; I am acid-minimal-cluster-1 2020-09-22 13:34:25,634 INFO: updated leader lock during promote server promoting 2020-09-22 13:34:25,671 INFO: cleared rewind state after becoming the leader

As you can see, I disconnected the network at 13:33:01 and there were no more logs for almost a minute. 
Then a timeout was reached (I dont know how I can change the timeout to a shorter time).
After 25 more seconds, the node started to promote itself.
Is there a way I can reduce this amount of time? 
What I basically want is the former standby to promote itself to the master as soon as the master does not renew his lock. 

[CyberDem0n]

After 25 more seconds, the node started to promote itself. Is there a way I can reduce this amount of time? What I basically want is the former standby to promote itself to the master as soon as the master does not renew his lock.

The promotion happens when the leader doesn't update the lock for 30 seconds (ttl). It is possible to reduce ttl, but I would hardly advise you not to do that. What if it is a temporary network glitch that will resolve itself soon?

[BenchmarkingBuffalo]

Hi, I changed the ttl to two seconds (see the manifest above), so that can’t be it or am I wrong?

[CyberDem0n]

Please, don't do it! No one could beat laws of physics!

If one changes ttl, loop_wait and retry_timeout also should be adjusted. There is a formula which must hold: loop_wait + 2*retry_timeout <= ttl: https://patroni.readthedocs.io/en/latest/SETTINGS.html

Besides that, there are some hardcoded timeouts, it is absolutely unsafe to go below 20 sec!

[BenchmarkingBuffalo]

Ok, thank you for your reply. I had already adjusted the other values as well, but I did not know about the hardcoded timeouts.