High availability cluster functionality

[kressi]

It looks like #203 has been abandoned. Nevertheless, I could use HA cluster functionality. I am aware of #250 which got closed again. However, almost another year has passed without any visible activity. So, I am hoping this MR will be accepted.

• All commits from Failover functionality #203 are included, which fixes Support for Postgresql multi-host failover functionality #120
• Merge conflicts with main are resolved
• Change hostRecheckTime from int to Duration

I got stuck and would appreciate some guidance with following issues brought up by @mp911de.

• Simplify ClientFactory and ClientSupplier into a single interface (Failover functionality #203 (review))
• Simple and pluggable client/connect model (Failover functionality #203 (comment))

[mp911de]

This looks pretty exciting. I will have to take a deeper look in the next days.

[mp911de]

From a planning perspective, while I'd love to include the PR into our 0.9 GA release, I think it doesn't work out with our timing to ship 0.9 GA today. We will have to postpone it to one of the maintenance releases.

[mp911de]

I started looking today at the pull request and figured that I'm not able to rebase the PR. Can you please squash all commits into a single one and rebase the PR onto main, please?

[harishvashistha]

@kressi Can you please support to move these changes to official master branch of this library? We are eagerly waiting for your changes in official release.

[harishvashistha]

@mp911de @kressi can you help pushing this feature to official release of r2dbc library.

[kressi]

@harishvashistha, @mp911de, I will start working on it

[harishvashistha]

@kressi thanks for handling this on priority. It is a nice gesture. @mp911de please help us to take this work into official release.

[mp911de]

Thanks a lot @kressi, I plan to wrap up the cluster support this week. I expect that the next release will be a milestone of 1.0 to allow folks to try out this feature before we ship it with a GA release.

[harishvashistha]

@mp911de that would be great. Waiting eagerly on this.

[mp911de]

Not sure whether High availability cluster functionality is a bit misleading as the title of this change since what the change introduces is the detection of roles from a set of hosts and picking one host to connect to that works.

In any case, I like the outcome here and likely in the future, we could maintain active connections along with an improved host selector that applies on a per-transaction/per query basis so that e.g. read-only transactions might run directly on a replica instead of the master node.

[mp911de]

I polished up the code a bit and refactored the code to align with ConnectionFunction and ConnectionStrategy concepts where ConnectionFunction is a low-level connect-to-host facility and the strategy can make use of connection functions to connect to a particular target server.

I still need to work on the tests to get an idea whether all of this is correct but things look pretty decent now.

[mp911de]

Thanks, everyone for your contribution. That's merged and polished now.

[harishvashistha]

@mp911de Thanks, great news. which version of r2dbc-postgressql jar has this feature? or soon you're going to make something available?

[mp911de]

This feature will be shipped with 1.0 M1. You can verify the functionality already with the version 1.0.0.BUILD-SNAPSHOT and report back any issues. That way, we can find out whether things are working for you before we ship a milestone release.

[alexander-mikhailov]

@mp911de @kressi @harishvashistha
Is there a way to see in logs that plugin does work, like a line indicating a host was changed or similar?

One more question is:
Does the plugin support io.r2dbc.pool.ConnectionPool factory?

val connectionFactoryOptionsBuilder = ConnectionFactoryOptions.builder()
            .option(ConnectionFactoryOptions.HOST, "host1,host2")
            .option(ConnectionFactoryOptions.USER, "user")
            .option(ConnectionFactoryOptions.PASSWORD, "pwd")
            .option(ConnectionFactoryOptions.LOCK_WAIT_TIMEOUT, properties.lockTimeout)
            .option(ConnectionFactoryOptions.STATEMENT_TIMEOUT, properties.statementTimeout)
            .option(HOST_RECHECK_TIME, Duration.ofMillis(500))
            .option(ConnectionFactoryOptions.PROTOCOL, FAILOVER_PROTOCOL)
            .option(ConnectionFactoryOptions.DATABASE, "db")
        
val config = PostgresqlConnectionFactoryProvider.builder(connectionFactoryOptionsBuilder.build()).build()
    
val postgresqlConnectionFactory = PostgresqlConnectionFactory(config)
            
val cpc = ConnectionPoolConfiguration.builder(postgresqlConnectionFactory).build()

return ConnectionPool(cpc)