Can hasura-engine read-replicas all share a single metadata database?

[larsblumberg]

Context:

Without using Hasura Cloud to still achieve postgres replica load-balancing, it's possible to run a separate, second hasura-engine read-only instance which connects to a postgres read-replica instance. The main/master hasura-engine instance connects against postgres master for R/W ops while the hasura read-only instance connects against the postgres read-replica in read-only mode.

To have hasura-engine connect in read-only mode, one needs to strip off cron triggers and event triggers from the metadata. Notice that otherwise hasura will connect in read-write mode which will be rejected by the postgres read-replica.

Doing so you can load-balance graphql read-only operations such as live-subscriptions and queries against the hasura master and the single hasura read-replica.

Problem:

Above setup works well and now I'd like to increase the postgres read-replica to 2 read-only instances for client apps to load-balance against.

Questions:

Can hasura-engine read-only instances all share a single hasura metadata database? Or does each hasura instance needs its very own metadata database so that number of hasura read-only instances = number of metadata instances?

[adas98012]

As per the Hasura Docs:

Hasura Cloud and Hasura Enterprise can load balance queries and subscriptions across read replicas while sending all mutations and metadata API calls to the primary.

So, it will be a single metadata database (attached to primary) and will be shared across the Read Replicas.

[larsblumberg]

Thanks, that answers my question concerning all read-replicas being able to share the same metadata postgres instance.