lldp: show lldp is the status view, neighbours stay on show lldp neighbors #1915
No reviewers
Labels
No milestone
No project
No assignees
1 participant
Notifications
Due date
No due date set.
Dependencies
No dependencies set.
Reference
jaylouisw/packetforge!1915
Loading…
Add table
Add a link
Reference in a new issue
No description provided.
Delete branch "rev/1117-2301"
Deleting a branch is permanent. Although the deleted branch may continue to exist for a short time before it actually gets removed, it CANNOT be undone in most cases. Continue?
show lldpanswered with the neighbour table, so the protocol's own status could not be asked for: a box with LLDP running and nothing heard read identically to one with LLDP off, and a technician typingshow lldpto check the protocol got a port list.What changed
show lldpprints the four-line global block, byte for byte from a Catalyst 9500 transcript in Cisco's field notice on unknown-protocol drops (doc 224586):Global LLDP Information:/Status: ACTIVE/ advertisements every 30 s / hold time 120 s / reinitialisation delay 2 s.engine/lldp.js(DEFAULT_TX_INTERVAL_S,DEFAULT_TTL_S,DEFAULT_REINIT_DELAY_S) so the status block and the TTL every LLDPDU carries cannot drift; the TTL is the interval x the hold multiplier 4.show lldp neighborsis unchanged (table,Total entries displayed: N); any other word is refused in one line and names it.ACTIVEstates this model, not a Cisco default:announceLldp()walks every device and there is no global switch, so% LLDP is not enabledwould be a claim about state the model cannot hold.docs/CLI.md: new table row, the block, its source, and the sentence naming the model's choice.What stays open
Whether a factory-default Cisco access switch runs LLDP before
lldp runis genuinely contradictory in public sources (default tables say disabled; field reports and the field-notice transcript say ACTIVE). That needs a per-devicelldp on|offplus a capability token, and it is not done.research/lldp-default-state.mdnow records what was implemented and what is left.Tests
node scripts/syntax-check.js401 files OK; fullnpm test328 suites, 11,410 assertions, 0 failed. Newtest/show-lldp-status.test.js(21 assertions) covers the block byte for byte, the two views differing, the empty-but-enabled case, the constant tie, and the refusal. Four suites that read the table through the bare command now ask for it by the word that prints it.show lldpis the status view, and neighbours stayed onshow lldp neighborseaf80bc421e39618f5e9Not merged, and it is not the code. The head run (
e39618f5, run 5724) is red on one leg only:test/link-parts.test.js:78 passed, 1 failed, and the failing line isharness failed: Error: socket hang up— an HTTP client losing its socket mid-suite, not an assertion about behaviour.11403 assertions passed, 1 failed.I did not take "flake" on trust. In a scratch worktree at
e39618f5I ran the three suites the failure touches, on this machine:Evidence that the runner, not this branch, is the variable: main's own push run for
cf8cb020three minutes earlier (run 5719, the same window) failed 4 assertions intest/server-bind.test.json Node 22 — thePF_HOST=0.0.0.0case, wherefreePort()hands out a port another suite can take before the server binds — while Node 18 and 20 passed there. Two consecutive runs, two different single-test failures, disjoint Node versions, both passing when re-run. That is a load-sensitive runner, not a regression.So: no rebase, no source change. The branch is clean against main — 9 files, all LLDP (
engine/lldp.js,engine/cli.js,docs/CLI.md,research/lldp-default-state.md, five tests), and it is 1 commit behindee1e91a7, which touched onlydocs/DEVICE_REGISTRY.md,research/open-device-requests.jsonandtest/closed-issues-audit.test.js— no overlap.What it needs is one more green run on this head or a rebased one. Jay's rule is that a non-conflicting green PR automerges, and I do not merge a head whose status is not success, so I have left it armed rather than merging over a socket hang up. If the next push to main rebases it, the fresh run should merge it on its own.
Two things worth Jay's attention, since this is the second flake in three minutes:
test/server-bind.test.jshas a real port race (freePort()closes the socket before the child process binds, so a parallel suite can take it), andtest/link-parts.test.jshas no retry on a dropped socket. Both are cheap to make deterministic and neither is a behavioural question. — surgery loope39618f5e976b7237a55