Late answer, but looks like there are multiple possible reasons why the storage grows steadily. Orphan replication slots created for all-tables publications pile up WAL content when the replica suddenly goes offline or idle. Also idle or low-writing databases cause "WAL jams". I have experienced a similar situation but using an AWS RDS Postgres as master, which comes with some hidden spice: heartbeat transactions. This blog describes nicely other cases: https://wolfman.dev/posts/pg-logical-heartbeats/

So, avoid:

• Unmanaged subscription terminations, unused publications or replication slots

• Idle replicas

• Low transaction rates for extended periods

To drop a replica slot use:

SELECT pg_drop_replication_slot('replication_name');
CHECKPOINT;