Session and stream itests: timing-dependent and always-true assertions #81

New issue

Open

opened 2026-10-12 11:35:16 +00:00 by claude · 0 comments

claude commented

2026-10-12 11:35:16 +00:00

Member

QueryProcessIntegrationTest "each message arrives while the CLI is still running" fails under load — failed in ./mill __.itest and once in kyo.itest (2026-10-01, with the 3072-line flood test and real-CLI E2E in the same run); passed 3/3 alone and 2/2 in later full kyo.itest runs; direct ClaudeCodeStreamingTest, ProcessManagerTest timeout cleanup and effectful lifecycle failed in the same parallel run and pass alone
Chat and interrupt itests decide "the turn is in flight" with fixed sleeps (between = 300.millis, a 2 s stand-in delay); a loaded host can reorder them — kyo/itest/.../ChatIntegrationTest.scala, SessionIntegrationTest.scala "send returns before the result lands" (02-kyo-session review)
E2E interrupt asserts stillQueued.forall(_.nonEmpty), true for an empty list; ported as in zio — kyo/itest/.../SessionE2ETest.scala, zio/itest/.../SessionE2ETest.scala (02-kyo-session review)
effectful SessionIntegrationTest "full single-turn session lifecycle with mock CLI" gets the wrong message count under a parallel ./mill __.itest (effectful/itest/.../SessionIntegrationTest.scala:47, failed in 2 of 3 parallel runs on 2026-10-01); passes alone; not touched by this branch

Found closing kyo-experiment/02-kyo-session (code review and gate runs, 2026-10-01). Four follow-ups, one cause: tests whose result depends on scheduling under load, and one assertion that cannot fail. - `QueryProcessIntegrationTest` "each message arrives while the CLI is still running" fails under load — failed in `./mill __.itest` and once in `kyo.itest` (2026-10-01, with the 3072-line flood test and real-CLI E2E in the same run); passed 3/3 alone and 2/2 in later full `kyo.itest` runs; `direct` `ClaudeCodeStreamingTest`, `ProcessManagerTest` timeout cleanup and `effectful` lifecycle failed in the same parallel run and pass alone - Chat and interrupt itests decide "the turn is in flight" with fixed sleeps (`between = 300.millis`, a 2 s stand-in delay); a loaded host can reorder them — `kyo/itest/.../ChatIntegrationTest.scala`, `SessionIntegrationTest.scala` "send returns before the result lands" (02-kyo-session review) - E2E interrupt asserts `stillQueued.forall(_.nonEmpty)`, true for an empty list; ported as in `zio` — `kyo/itest/.../SessionE2ETest.scala`, `zio/itest/.../SessionE2ETest.scala` (02-kyo-session review) - `effectful` `SessionIntegrationTest` "full single-turn session lifecycle with mock CLI" gets the wrong message count under a parallel `./mill __.itest` (`effectful/itest/.../SessionIntegrationTest.scala:47`, failed in 2 of 3 parallel runs on 2026-10-01); passes alone; not touched by this branch

claude referenced this issue

2026-10-13 09:35:28 +00:00

kyo-experiment: a Kyo session with real interrupt, and the SDK verdict (slice 02-kyo-session) #88