prisma-risk / tsoracle / 26462202659

95%

push

github

feat: GetCurrentMaxSafe RPC (#493)

* feat(proto): add GetCurrentMaxSafe RPC

External coordinators that fence reads on a durable safe-point can now query
the leader directly instead of reaching into the consensus state machine.

The handler in tsoracle-server is wired with a default-response stub so the
new trait method compiles across every TsoService impl in the workspace;
subsequent commits expose the underlying allocator + serving-core state and
swap the stub for the real read.

* feat(core): expose committed_high_water accessor on Allocator

* feat(server): serve GetCurrentMaxSafe from ServingCore

ServingCore now owns window_ahead and a current_max_safe_physical_ms()
accessor that derives the safe-point from the allocator's committed
high-water; the handler reads it together with the current leader epoch
(zero when not leader, matching the proto's cold-start sentinel).

* feat(client): add get_current_max_safe

Single-shot RPC targeting the cached leader (falling back to the first
configured endpoint when no leader is known yet). Returns a typed
MaxSafe carrying the physical-ms safe-point and the leader epoch. The
proto contract makes follower responses non-erroring (they return 0),
so a simple one-pass dial is enough for the first cut; pollers needing
freshness target the leader.

* test(server): integration test for GetCurrentMaxSafe roundtrip

* chore(server): register get_current_max_safe test binary with required-features

47 of 92 new or added lines in 7 files covered. (51.09%)

13296 of 13974 relevant lines covered (95.15%)

465584.33 hits per line