Replication broken in dual stack clusters
General
Escalation
General
Escalation
Description
Environment
K3s v1.23.6 as a dual stack cluster with Calico.
Smart Checklist
Activity
Show:
Ivan Pylypenko June 13, 2022 at 10:15 AM
Hi
At the moment operator and all underlaying routines expects single stack cluster network only (patroni should be provided with exact address to listen on).
We may consider adding dual stack in the future if that is feasible. Up to @Sergey Pronin
Details
Details
Assignee
Unassigned
UnassignedReporter
DerEnderKeks
DerEnderKeks Labels
Components
Affects versions
Priority
HighSmart Checklist
Open Smart Checklist
Smart Checklist
Open Smart Checklist
Created May 31, 2022 at 3:30 PM
Updated July 15, 2024 at 3:33 PM
In a dual stack cluster the replication fails with the following error:
2022-05-31 15:07:59,371 ERROR: Exception when working with master via replication connection Traceback (most recent call last): File "/usr/lib/python3.6/site-packages/patroni/postgresql/rewind.py", line 173, in _check_timeline_and_lsn with self._postgresql.get_replication_connection_cursor(**leader.conn_kwargs()) as cur: File "/usr/lib64/python3.6/contextlib.py", line 81, in __enter__ return next(self.gen) File "/usr/lib/python3.6/site-packages/patroni/postgresql/__init__.py", line 816, in get_replication_connection_cursor with get_connection_cursor(**conn_kwargs) as cur: File "/usr/lib64/python3.6/contextlib.py", line 81, in __enter__ return next(self.gen) File "/usr/lib/python3.6/site-packages/patroni/postgresql/connection.py", line 43, in get_connection_cursor conn = psycopg2.connect(**kwargs) File "/usr/lib64/python3.6/site-packages/psycopg2/__init__.py", line 127, in connect conn = _connect(dsn, connection_factory=connection_factory, **kwasync) psycopg2.OperationalError: could not translate host name "2003:xxxx:xxxx::xxxx 10.x.x.x" to address: Name or service not knownI noticed that the
PATRONI_POSTGRESQL_CONNECT_ADDRESSandPATRONI_RESTAPI_CONNECT_ADDRESSenv var are both set to a nonsens value of both pod IPs separated by a space. This seems to be caused [here|https://github.com/percona/percona-docker/blob/1db7d7840b87beb3c0adaebde36d1fc15d4f1603/postgresql-containers/bin/postgres-ha/bootstrap/pre-bootstrap.sh#L124.]This completely prevents usage of the replicas feature and maybe even more in dual stack clusters.
A possible simple solution would probably be to split the string returned by
hostname -iand just use the first IP.