[Nic-dorman]

tests/e2e/testnet.rs set TEST_CORE_CONNECTION_TIMEOUT_SECS = 2. It came from 857398c, which kept DHT leave waits short during teardown and, in the same commit, made shutdown concurrent.

But saorsa-core uses a node's connection_timeout for two things: the connection timeout, and the request timeout of every DHT operation (network.rs: request_timeout: config.connection_timeout). So the e2e nodes gave each peer 2 s to answer any DHT lookup, against 25 s in production.

ant-client's harness had the same trap at 5 s, and it caused its E2E macOS flake (V2-1852, ant-client #248). On loaded runners, witnessed close-group lookups dropped slow but live peers, and uploads failed with "found 6 peers, need 7". Both harnesses now use 10 s under the same constant name, with the same explanation, and each says to keep the two in step. shutdown() here already stops nodes concurrently, so the longer value doesn't add up across nodes.

Linear issue

Closes V2-1869

Risk tier

• T0 — docs / tooling / CI / pure UX-output. Repo CI only.
• T1 — client-only, no network-facing behavior change. CI + prod compat smoke.
• T2 — node/client logic with behavioral surface, no protocol/format/economics change. Dev testnet + ADR.
• T3 — protocol / storage format / payments / routing. T2 evidence + adversarial testing.

Compatibility

• Wire: none
• Storage: none
• API: none. E2E test harness only (tests/e2e/testnet.rs).

Semver impact

• breaking
• feature
• fix

Test evidence

This repo, dev2 (Linux, 4 cores, debug, -D warnings). The e2e target was run with --features test-utils -- --test-threads=1, as CI runs it, both binaries built from 369a2e2:

Timeout	Result	Time
2 s (main)	120 passed, 0 failed, 3 ignored	988.7 s
10 s (this PR)	120 passed, 0 failed, 3 ignored	987.4 s

The tests that stop nodes partway through (replication, chunk) are no slower.

Choosing 10 s (ant-client's flaky test, 14 nodes, CPU capped with systemd-run -p CPUQuota):

• At 1.5 cores and at 1 core, 5 s failed every run, and 10, 15, 20 and 25 s passed every run (3/3, or 2/2 for the higher values at 1 core).
• 10 s was the lowest value tried that held.
• Run time didn't depend on the value.

New dependency

none

ADR

n/a

Mitigation / rollback

Revert. If a test that waits out a dead peer turns out too slow, lower the constant, and lower ant-client's to match.

🤖 Generated with Claude Code

[MaidSafe-Hermes]

APPROVE — Reviewed WithAutonomi/ant-node#277 at 6c1df6e3882a029e8c996a29eef2690385b0a839 (base 369a2e28bcbfbaa0127034be0f8f11b10fbfb00b). The change is limited to the E2E test harness timeout; it does not alter production configuration. The timeout increase is reasonable for slow but live peers, and shutdown waits run concurrently. No material correctness or compatibility defect found.

Coverage: 2/2 seats reviewed — sol-1 (gpt-6.1-sol xhigh) and glm-1 (z-ai/glm-5.3); both seats agree (two model families). E2E results cited in the PR description are author-reported, not independently reproduced.

The failing Security Audit for inherited RUSTSEC-2026-0387 is a merge gate, not a blocker to this approval.