[Conjure Java Runtime] Part 19: Retrying is Futile... Sometimes.

[jeremyk-91]

Goals (and why):

• Fix a bug in [Conjure Java Runtime] Part 18: Internode Support #4366: a FastFailoverProxy should not be created when creating a single node proxy, as that's futile.

Implementation Description (bullets):

• Remove FastFailoverProxy creation from the createProxy (no failover) codepath.
• Rename semantics of configuration in TestProxyUtils to make it obvious that the default params are with retrying enabled.

Testing (What was existing testing like? What have you done to improve it?):

• Existing tests should work, and should be faster.

Concerns (what feedback would you like?):

• In production things shouldn't be affected, as the single node clients created e.g. timelock's PaxosLearner etc. aren't gated behind an AwaitingLeadershipProxy. Does this make sense / do we need to perform more drastic action?

Where should we start reviewing?: ConjureJavaRuntimeTargetFactory

Priority (whenever / two weeks / yesterday): yesterday 🔥Until this merges, all builds will probably be ten minutes slower than they should be.

[changelog-app]

Generate changelog in changelog/@unreleased

Type

• Feature
• Improvement
• Fix
• Break
• Deprecation
• Manual task
• Migration

Description

We no longer create a FastFailoverProxy when creating a single node proxy. Previously, if this was created when talking to an individual TimeLock node that was not the leader (or a service that would otherwise return 308s), we would spin for about 10 seconds before returning. Users of failover proxies to TimeLock (e.g. if using the timestamp or lock services from a TransactionManager) are unaffected.

Check the box to generate changelog(s)

• Generate changelog entry

[felixdesouza]

Interesting!

[felixdesouza]

strange indent

[jeremyk-91]

Fixed. Not sure what happened here

[svc-autorelease]

Released 0.171.1