Allow JDBC connectors to retry on transient failures

[dominikzalewski]

Description

RetryingConnectionFactory now has a pluggable JDBC-connector specific retry strategy. This PR applies this change only to Oracle.

Release notes

( ) This is not user-visible or is docs only, and no release notes are required.
( ) Release notes are required. Please propose a release note for me.
(X) Release notes are required, with the following suggested text:
All JDBC-based connectors are now always retrying (5 times for 30 seconds with a backoff) if opening a connection fails. By default this happens only if SQLTransientException is encountered. For backward compatibility, for Oracle this is overridden to SQLRecoverableException.

[hashhar]

some comments

right now the refactors look larger than the actual change and I can't be sure what is refactor vs what is required for the functionality

[hashhar]

looks like pre-exisiting logic that we only use RCF with non-pooled connections.
I don't know if it was intentional or not, so either we should add RCF to pooled connection factory as well or add a comment explaining why we only use RCF with non-pooled connections.

[dominikzalewski]

Sorry, but this ask seems out of scope? This has only been revealed because I slightly refactor the existing code and how it is more evident what the logic is. The scope has been frozen before I started working. If there's a good reason that we should change it, then we will. But we need a good reason.

Acceptance criteria

• no new regressions
• unit tests for new framework that it actually retries
• it is possible to implement retry for any connector in future that fails sometimes to open a connection
• no custom retrying in any jdbc connector, all connectors use new framework

Let me know if I misunderstood AC laid out by @kokosing.

[dominikzalewski]

right now the refactors look larger than the actual change and I can't be sure what is refactor vs what is required for the functionality

I have broken down the change into a few commits, so that it is clear where the refactor ends and where adding functionality starts. If this is still not helpful, let's have a call.

[electrum]

The commit titles are too long. Please see https://cbea.ms/git-commit/

[dominikzalewski]

The commit titles are too long. Please see https://cbea.ms/git-commit/

Shortened commit messages and formatted the extended description to 72 chars column.

[kokosing]

How about PR title:

Allow JDBC connectors to retry on transient failures

Please provide PR description too and fill RN (release notes) form.

[dominikzalewski]

I have also forgot to mention to you before that there is also io.trino.plugin.jdbc.LazyConnectionFactory :) Just wanted to make sure this puzzle is not too simple for you and you remember about all puzzles.

Thanks. I spent some time analyzing this, as your previous hint from slack was actually tracking the JdbcClient chain, rather than ConnectionFactory. But I figured it out by myself and it made me so proud :)

Simplified the chain with Lazy and Stats. There's a separate commit.

[kokosing]

Just one comment.

[kokosing]

minor comments

I would like @hashhar to take a look too.

[hashhar]

The impl looks very good. I have no comments about it.

The only comment I have is about inspecting how the Guice map looks after the 2nd commit. From what I can tell we seem to have introduced duplication.

[hashhar]

For references here's the actual Guice map before and after change for both JdbcClient and ConnectionFactory

BEFORE:

DriverConnectionFactory (OracleClientModule#connectionFactory == @ForBaseJdbc)
RetryingConnectionFactory (RetryingConnectionFactory#ctor)
StatisticsAwareConnectionFactory (StatisticsAwareConnectionFactory#ctor)
LazyConnectionFactory (LazyConnectionFactory#ctor)
ReusableConnectionFactory (ResuableConnectionFactoryModule - conditional, if QueryConfig#isReuseEnabled is true)

------------------------------------------------------------

OracleClient (OracleClientModule#setup == @ForBaseJdbc)
ForwardingJdbcClient (JdbcDiagnosticModule#createJdbcClientWithStats)
StatisticsAwareJdbcClient (JdbcDiagnosticModule#createJdbcClientWithStats == @StatsCollecting)
CachingJdbcClient (CachingJdbcClient#ctor)

AFTER:

DriverConnectionFactory (OracleClientModule#connectionFactory == @ForBaseJdbc)
StatisticsAwareConnectionFactory (StatisticsAwareConnectionFactory#ctor)
RetryingConnectionFactory (RetryingConnectionFactory#ctor)
LazyConnectionFactory (LazyConnectionFactory#ctor)
ReusableConnectionFactory (ResuableConnectionFactoryModule - conditional, if QueryConfig#isReuseEnabled is true)

No change in JdbcClient tree after the change.

comments have been addressed

[hashhar]

@dominikzalewski can you please fill in the release notes template in the PR description since this would be a user visible change (even if not directly configurable).

[mosabua]

Fyi .. please check out the final release notes entry we added for future references and as an example.