Make tutorials 50–53 safer to operate

[yunwei37]

What changed

• keep lesson 50 attached until signal/duration, emit READY, then detach before reporting a stable final counter snapshot;
• make lesson 51 dry-run by default and require an exact local+remote 4-tuple before destroying one socket;
• add device, inode, and file-type identity to lesson 52 slow-read events;
• make lesson 53 report qdisc conflicts and an explicit, manual crash-recovery command;
• strengthen topic-selection and tutorial-writing skills so practical operator workflows are checked before publication;
• preserve the original bilingual tutorial narratives and keep public examples on the normal operator path rather than showing test-harness output.

Why

The existing examples loaded successfully but had lifecycle, safety, or actionability gaps. These changes make the public workflows usable without turning the tutorials into test fixtures.

Validation

• clean host builds for lessons 50–53;
• documentation unit tests, generated-index check, relative-link check, and exact embedded-source checks;
• KVM integration tests on x86_64 Linux 7.2.0-rc4 for all four lessons, covering READY/signal cleanup, exact socket selection, VFS object identity, qdisc conflicts, pacing, and crash recovery.

Lesson 54 is intentionally excluded and remains in a separate PR for further revision.

[tgoodwin]

Independent review gate 1 — exact-head maintainer review

Reviewed final pushed head ac8008aaf8b04ea405fd11371bae93c753b27901 against origin/main.

Verdict: no blocking or actionable findings.

The review specifically rechecked the prior lesson 53 blocker: global struct_ops conflicts now print qdisc state across all interfaces and do not suggest deleting the requested interface, while target-interface root-qdisc conflicts retain the interface-specific diagnostic and recovery command. The cross-interface regression exercises a stale bpf_pacer owner on one interface and a launch attempt on another.

Also verified:

• changed-path scope is limited to tutorial skills and lessons 50–53; no lesson 54 path is in the diff;
• 50–53 safety/lifecycle behavior matches the requested improvements;
• English and Chinese complete kernel/BPF source blocks match the checked-in source exactly;
• long user-space loaders use focused excerpts under the final skill rule, while concise loaders remain complete;
• public tutorial paths contain no repository test commands, harness output, fixtures, or migration-history narration;
• both updated skills pass quick_validate.py; documentation tests (15/15), source-sync tests (13/13), link checks, git diff --check, host builds, and KVM functional runs passed.

Code-growth review: exact diff is 27 files, +1606/-620 (production/build +389/-134, docs/skills +913/-386, tests +304/-100). The size is justified by the requested operational and safety coverage across four existing tutorials plus bilingual source synchronization; no fixture bloat or unrelated abstraction was found.

[tgoodwin]

Independent review gate 2 — cross-agent review

A separate Kimi reviewer inspected final pushed head ac8008aaf8b04ea405fd11371bae93c753b27901 against origin/main in an externally enforced read-only workspace.

Verdict: approve; no blocking or actionable findings.

The review verified:

• exact scope: 27 files, +1606/-620, with no lesson 54 path;
• lessons 50 and 52 contain complete kernel/BPF source and permitted focused excerpts for their long user-space loaders; lessons 51 and 53 retain complete embedded loaders;
• src/53-egress-pacer/tests/test_egress_pacer.py explicitly verifies the stale bpf_pacer owner before the cross-interface conflict check;
• global and target-interface qdisc conflicts remain safely distinguished;
• the final EN/ZH prose substitutions remove repository-test narration from lessons 51 and 52;
• the updated tutorial-writing skills consistently encode complete kernel source, focused long-user-space excerpts, normal user workflows, and private test infrastructure;
• remaining upstream selftest links are references, not reader-facing repository test steps.

Code-growth verdict: production/build +389/-134, docs/skills +913/-386, tests +304/-100; growth is focused and proportionate, and fixtures are minimal. Scope-boundary verdict: all public behavior, documentation, and skill changes are authorized by the requested 50–53 safety/usability work; no shared architecture or unrelated API was added.