prisma-risk / tsoracle / 26349056193

95%

Build Type

push

github

Committed by web-flow

Commit Message

fix(client): stop clearing leader cache on unactionable NOT_LEADER (#236) (#268)

The retry loop's `HintRejected` arm called `clear_leader()` on every
NOT_LEADER it could not redirect. But that outcome is emitted for three
benign cases: a FAILED_PRECONDITION with no leader-hint trailer, a
malformed trailer, or a hinted endpoint dropped by the TLS-downgrade
guard. None are evidence the cached leader is wrong — only that the
contacted peer could not redirect us. Under request coalescing, clearing
the cache on each flap stampedes every in-flight caller back onto a cold
worklist against all configured nodes.

Leave the cache intact in that arm; the status is still preserved as the
surfaced error and the worklist advances. With its only production caller
gone, `ChannelPool::clear_leader` (and its unit test) is removed as dead
code. The behavioral guarantee is pinned by a new test that drives the
real `issue_rpc` loop against a loopback follower returning NOT_LEADER
with no trailer and asserts the cached leader survives.

Updates the client-discovery docs and a now-stale integration-test
comment that described the old clear-on-NOT_LEADER behavior.

Coverage Stats

34 of 43 new or added lines in 1 file covered. (79.07%)

6 existing lines in 2 files now uncovered.

10386 of 10967 relevant lines covered (94.7%)

542792.15 hits per line