v0.17.0 — engine 0.17.1

[0.17.0] — 2026-08-25

Engine 0.17.1 (thread-aware retrieval, reachable event time, convergent
replication; 0.17.1 restores import yantrikdb on Python 3.10).

Changed

• Pinned engine moves from git tag v0.15.4 to v0.17.1. The only engine API
  change that reaches the server is recall() gaining event_after /
  event_before (valid-time filter, engine #149 phase 2); both handler call
  sites and the integration/compat fixtures pass None — no behaviour
  change. Forwarding the HTTP since / until window into those parameters
  is a separate change (in-progress branch).

Verified

• All server test targets against engine 0.17.1: unit 931, commit_replay 350,
  compat 6, cpu_isolation 4, crash_recovery 2, crdt_convergence 2,
  http_integration 13, wire_format_v1_0 356, protocol 15.

Known CI flake (recorded, follow-up open)

Two windows-latest runs on #98 each failed one YRP harness test — run 32908996823 yrp::chaos_test::torn_state_boots_quarantined_then_rejoins_via_grant (restarted node reported status: ok as a follower instead of quarantined), run 32909824141 yrp::driver::tests::read_barrier_ok_on_leader_retry_on_follower (10 s election deadline on a ~6× slower runner). macOS/Ubuntu passed, the re-run passed, and 20×/10× sequential local runs on Windows passed. Root cause is harness nondeterminism (co-started identical election tickers; quarantine assertion without a partition-in-effect precondition), not the engine bump. Follow-up: deterministic/staggered test elections and a partition precondition in the chaos test.

What's Changed

• chore: republish protocol and witness as 0.1.1 — Apache-2.0 on crates.io by @spranab in #96
• docs: the README's contradiction example does not reproduce (#95) by @spranab in #97
• deps: engine v0.15.4 → v0.17.1; release v0.17.0 by @spranab in #98

Full Changelog: v0.16.2...v0.17.0