I believe that this is because it's not always sending feedback for replication messages, and only does so when there's a change to a table for a Meilisync index.
If instead we always send feedback, then it should allow Postgres to free WAL segments on idle databases.
My previous PR partially fixed the Postgres disk usage issue https://github.com/long2ice/meilisync/issues/84
It was pointed out that for idle databases the disk use was still growing.
I believe that this is because it's not always sending feedback for replication messages, and only does so when there's a change to a table for a Meilisync index. If instead we always send feedback, then it should allow Postgres to free WAL segments on idle databases.