Connection String with multiple hosts is not being parsed as expected

[chandelabhishek]

Hi There,

According to PG docs multiple hosts can be supplied in connection string.

for example:
If I supply connectionString as postgres://user:@localhost:6432,localhost:5432/db_name
it is being parsed as this:

{
    driver: 'postgres',
    user: 'user',
    password: '',
    database: ':6432,localhost/db_name',
    host: 'localhost',
    port: '5432'
}

above is throwing error. as the connection which gets constructed from this is wrong. This is the issue with pg-connection-string.

Also, I am curious, why do we need to parse a connection-string, why can't we directly pass it to libpq?

Will you accept a PR for this?

[wpatrik14]

I have just raised a similar issue to pg-native library: #
Duplicate of #brianc/node-pg-native#96

[beeing]

Probably will be easier to use a middleware like pgpool or pgbouncer for this.

[lukos]

The functionality that is important is that if I need to failover a database, I can do so without updating connection strings because the library will automatically choose a server that is the primary. I can't see that pgbouncer or pgpool have this functionality.

[boromisp]

You can achieve that with a generic TCP reverse proxy, like HAProxy. First result in ddg for dynamically changing servers in haproxy: https://stackoverflow.com/a/53294153

But you get more features from postgres-aware programs, like pgpool:
https://www.pgpool.net/docs/latest/en/html/runtime-config-failover.html

I've never set something like this up before, the last time I've done postgres HA, it was with virtual IPs, but that usually isn't a good option in cloud environments.

[chandelabhishek]

If this option is available, then it can do switchover by itself, as libpq has that capability. Then we don't need to maintain an extra moving part like ha-proxy or pg-pool.