fix(install): surface Higress gateway startup diagnostics - #984
fix(install): surface Higress gateway startup diagnostics#984RerankerGuo wants to merge 2 commits into
Conversation
📊 CI Metrics ReportSummary
By Role
Per-Test Breakdown
Trends✅ 4 test(s) improved (fewer LLM calls) Generated by HiClaw CI on 2026-07-13 15:50:03 UTC |
|
@shiyiyue1102 Hi! Quick note on the SHARD_A copaw/copaw and copaw/hermes failures: I downloaded the test artifacts from run 28634278368. In the copaw/hermes case, the So the failure mode is the Manager not creating bob (a test-prompt interpretation flake), not anything in the install diagnostics this PR adds. The change itself is shell diagnostics + a sysctl note in the FAQ. Could you either re-run CI or merge if you agree the flake is unrelated? Happy to dig further if you can point me at what assertion actually failed. |
83d2500 to
c2c7746
Compare
|
@shiyiyue1102 Hi! Re-rebased onto current
This is a different failure pattern from the earlier SHARD_A copaw/hermes flake. Test 14 has its own LLM-flake surface (3 workers, 4 phases, complex branch-name / file-path / report-name invariants), and there's no equivalent deterministic-bypass like #985 yet for it. The change in this PR is shell diagnostics for Higress Gateway startup + an English/Chinese FAQ note on sysctl checks — it doesn't touch Manager agent runtime, worker git ops, or test-14 setup. Could you re-run SHARD_B copaw/hermes, or merge this PR if you agree the test 14 failure is unrelated? Happy to dig into the test-14 artifacts (run 28926397804, job 85816940745) if it would help. Also: PR #985 (SHARD_A bob-setup deterministic fix) just turned green and was pinged this morning as priority. |
v1 over-prescribed by pre-creating the project room and instructing Manager not to call create-project.sh. That broke Manager's natural workflow — it lost its entry point for the rest of the coordination and ended up writing ad-hoc Matrix HTTP scripts (run 28948704338). copaw/hermes made Manager do 136 LLM calls + alice 54 calls + bob/charlie 0 (phases 2-4 never started). v2 only pre-creates the 3 workers (the actual flake source on SHARD_A). Manager still walks through create-project.sh + YOLO Step 1d (meta.json active transition) + 4-phase coordination as the workflow intends — this test measures that flow, not LLM timing on worker creation. Refs: agentscope-ai#984
7756e4d to
b0d4f81
Compare
b0d4f81 to
ae9dfa8
Compare
Pre-create the 3 workers (alice, bob, charlie) via hiclaw apply worker and the shared project via create-project.sh + immediate meta.json activation + MinIO sync, before sending the 4-phase task to Manager. Manager now focuses on phase coordination only — the worker + project setup steps that previously stalled on Manager's tool-guard approval prompt (jq | mv | cat on meta.json) are now in the test runner and never enter the LLM control loop. Mirrors the fix in agentscope-ai#985 for test-06-multi-worker.sh (bob creation flake). Both fixes isolate LLM-driven coordination from infrastructure setup so the test measures coordination quality, not LLM timing. Fixes the SHARD_B copaw/hermes test-14 failure seen on agentscope-ai#984 (run 28926397804, job 85816940745), where Manager hung for 30 minutes on a tool-guard approval prompt while trying to update meta.json status. Refs: agentscope-ai#984
v1 over-prescribed by pre-creating the project room and instructing Manager not to call create-project.sh. That broke Manager's natural workflow — it lost its entry point for the rest of the coordination and ended up writing ad-hoc Matrix HTTP scripts (run 28948704338). copaw/hermes made Manager do 136 LLM calls + alice 54 calls + bob/charlie 0 (phases 2-4 never started). v2 only pre-creates the 3 workers (the actual flake source on SHARD_A). Manager still walks through create-project.sh + YOLO Step 1d (meta.json active transition) + 4-phase coordination as the workflow intends — this test measures that flow, not LLM timing on worker creation. Refs: agentscope-ai#984
1 participant
Summary
/statusfs.inotify.max_user_instanceslimit from the ARM64 Envoy/V8 crash that can persist after sysctls are raisedRefs #888.
Verification
bash -n install/hiclaw-install.shgit diff --checkfatal log tail, and the Wasm compatibility warning
main(6f0c7da)pwshis not installed in this environmentCI note
Both refreshed runs passed every build and 9/10 integration matrix jobs. The
only failures were two different pre-existing
test-06-multi-workersetupraces in copaw/hermes SHARD_A:
This PR only changes installer diagnostics and FAQ text. #985 now removes both
Manager-conversation dependencies from Bob's deterministic test setup.
Temporary dependency
78e9385), whose full CI passed all 10 integration matrices.b0d4f81dpassed every build and all 10 integration matrices.