docs(CLI): the LAG cap refusal is typed at a port that is not in the bundle yet #2033
Closed
jaylouisw
wants to merge 1 commit from
docs/1118-2047 into main
pull from: docs/1118-2047
merge into: jaylouisw:main
jaylouisw:main
jaylouisw:rev/1121-1812
jaylouisw:catalogue/1121-1822
jaylouisw:docs/1121-1817
jaylouisw:catalogue/1121-1752
jaylouisw:docs/1121-1747
jaylouisw:rev/1121-1742
jaylouisw:catalogue/1121-1722
jaylouisw:rev/1121-1712
jaylouisw:docs/1121-1717
jaylouisw:catalogue/1121-1652
jaylouisw:rev/1121-1642
jaylouisw:docs/1121-1647
jaylouisw:ci/1121-1646
jaylouisw:ci/1121-1427
jaylouisw:rev/1121-1412
jaylouisw:catalogue/1121-1422
jaylouisw:docs/1121-1417
jaylouisw:catalogue/1121-1352
jaylouisw:rev/1121-1342
jaylouisw:docs/1121-1347
jaylouisw:rev/1121-1212
jaylouisw:catalogue/1121-1222
jaylouisw:docs/1121-1217
jaylouisw:ci/1121-1027
jaylouisw:catalogue/1121-1022
jaylouisw:rev/1121-1013
jaylouisw:docs/1121-1017
jaylouisw:catalogue/1121-0852
jaylouisw:rev/1121-0842
jaylouisw:docs/1121-0847
jaylouisw:ci/1121-0827
jaylouisw:catalogue/1121-0822
jaylouisw:rev/1121-0812
jaylouisw:catalogue/1121-0752
jaylouisw:rev/1121-0742
jaylouisw:docs/1121-0747
jaylouisw:catalogue/1121-0722
jaylouisw:docs/1121-0717
jaylouisw:rev/1121-0712
jaylouisw:rev/1121-0642
jaylouisw:plan/1121-0637
jaylouisw:ci/1121-0627
jaylouisw:catalogue/1121-0622
jaylouisw:rev/1121-0612
jaylouisw:docs/1121-0617
jaylouisw:plan/1120-0337
jaylouisw:docs/1120-0347
jaylouisw:rev/1119-1942
jaylouisw:rev/1119-1912
jaylouisw:catalogue/1119-1752
jaylouisw:docs/1119-1347
jaylouisw:catalogue/1119-1352
jaylouisw:fix/1119-1232
jaylouisw:fix/1119-0602
jaylouisw:catalogue/1119-0452
jaylouisw:catalogue/1119-0422
jaylouisw:docs/1119-0417
jaylouisw:fix/1119-0202
jaylouisw:catalogue/1119-0222
jaylouisw:rev/1119-0042
jaylouisw:catalogue/1118-2252
jaylouisw:catalogue/1118-2152
jaylouisw:fix/1118-2037
jaylouisw:ci/1118-2027
jaylouisw:docs/1118-1416
jaylouisw:fix/1118-1006
jaylouisw:catalogue/1118-0326
jaylouisw:catalogue/1117-2326
jaylouisw:catalogue/1117-0025
jaylouisw:catalogue/1116-1730
jaylouisw:fix/1523-speed-downgrade-one-end
jaylouisw:fix/enable-required-and-empty-pipe-help
jaylouisw:land-all-v5
jaylouisw:land-all-v4
jaylouisw:land-all-v3
jaylouisw:land-all-v2
jaylouisw:land-all
jaylouisw:fix/board-id
jaylouisw:fix/stated-range
jaylouisw:fix/no-form-names-v2
jaylouisw:fix/voice-vlan
jaylouisw:devices/catalogue-batch
jaylouisw:devices/powerline-moca
jaylouisw:devices/firewalls-enterprise
jaylouisw:devices/pon-splitters
jaylouisw:devices/nvr-and-taps
jaylouisw:devices/four-nvr-models
jaylouisw:fix/contributing-exemption
jaylouisw:fix/stale-github-urls
jaylouisw:ci/cancel-superseded-runs
jaylouisw:gh/1438
jaylouisw:device/uisp-console-lbe-xr
jaylouisw:gh/1442
jaylouisw:docs/readme-current-counts
jaylouisw:feat/device-ubiquiti-wave-fiber-onu
jaylouisw:fix/address-ownership
jaylouisw:feat/vrrp-in-path
jaylouisw:fix/merge-research-duplicate
jaylouisw:docs/hybrid-patch-current
jaylouisw:fix/audit-port-template
jaylouisw:fix/2930f-jl256a-uplinks
jaylouisw:feat/dhcp-in-path
jaylouisw:feat/hackgpon-ubiquiti-uisp-fiber-xgs
jaylouisw:feat/hackgpon-adtran-sdx-621
jaylouisw:tool/issue-duplicate-sweep
jaylouisw:feat/device-mikrotik-rb433ah
jaylouisw:feat/hackgpon-ma5671a
jaylouisw:issues/categories-and-labels
jaylouisw:ci/run-every-suite-locally
jaylouisw:chore/http-smoke
No reviewers
Labels
Milestone
Clear milestone
No items
No milestone
Projects
Clear projects
No items
No project
Assignees
Clear assignees
No assignees
2 participants
Notifications
Due date
The due date is invalid or out of range. Please use the format "yyyy-mm-dd".
No due date set.
Dependencies
No dependencies set.
Reference
jaylouisw/packetforge!2033
Loading…
Add table
Add a link
Reference in a new issue
No description provided.
Delete branch "docs/1118-2047"
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?
The transcript in the
A bundle carries eight links; the rest wait as (H)section showedchannel-group 1 mode ontyped at Gi1/0/9 straight after Gi1/0/9 joined the group as hot-standby, with the cap refusal printed over it.Replayed against
engine/cli.js(#2024), that line does not refuse._channelGroup()compares the cap only on the joining path — the branch is guarded byalreadyIn, so a port that is already a member skips it. On Gi1/0/9 the console instead converts the group (Channel group 1 mode changed from active to on for every member: Gi1/0/1, Gi1/0/2, …) and adds the port; the group then holds ten members in a static (mode on) mode whose own refusal says eight at most.This change:
interface Gi1/0/11, the eleventh socket, outside the ten-member LACP bundle — so the block replays prompt for prompt;Docs only.
docs/CLI.md.Gates:
scripts/syntax-check.js423 files OK;check-control-bytesclean; docs-commands 7/7 (59 forms, 118 commands, 74 refusals); help-coverage 8/8 (102 advertised commands, 16 show commands); docs-syntax (CLI transcript) 4/4; help-forms (help-key 40, no-form 81, 10/10).The transcript in 'A bundle carries eight links' showed `channel-group 1 mode on` typed at Gi1/0/9 immediately after Gi1/0/9 joined the group as hot-standby, and printed the cap refusal over it. Replayed against engine/cli.js that line does not refuse: `_channelGroup()` skips the cap when the port is already a member (`alreadyIn`), so the console converts the group's mode for every member ("Channel group 1 mode changed from active to on for every member: ...") and adds the port. The refusal is only given to a port that is not in the bundle. The example now types the refusal where the console gives it - `interface Gi1/0/11`, eleventh socket, outside the ten-member LACP bundle - and the page records the hole the old transcript hid: the mode conversion leaves a static bundle holding ten members in a mode whose own sentence says eight at most.7554f9af5b022b1ea0e8022b1ea0e8toa23f9573a8a23f9573a86f60f521f36f60f521f3to8c4cc9b2538c4cc9b253to03ce6c464dQueue livelock — measured, not inferred. This PR has now been rebased out from under three green runs in 40 minutes:
7554f9af022b1ea0— 0 statusesa23f9573main→ Auto-rebase force-push6f60f521→8c4cc9b2→03ce6c46, 0 statusesEvery push
auto-rebase.ymlmakes lands as an action-token push, which does not raise asynchronizeevent, so no CI runs on the SHA the bot mints.workflow_dispatchdoes not help either: run 7058 ran againsted1ef529and wrote zero commit statuses (checkedGET /commits/ed1ef529/statuses→[]), so only apull_requestevent produces checks. Net effect: the queue can only ever land the one PR that is newest whenmainmoves — 4 of 5 merges this hour needed a manual close+reopen to be green at all.Needs Jay's decision: in
auto-rebase.yml, aftergit push --force-with-lease, either (a) push with a PAT instead of the action token so the push carries a realsynchronizeevent, or (b)POST /actions/workflows/ci.yml/dispatcheson the branch — which needs (a) to work anyway, since the dispatch itself is token-authenticated. Nothing on the PR side can fix it.03ce6c464dtoebaefdddb2Third reproduction, and the trigger is now unambiguous: the merge itself.
This PR's head was one of the four that lost its checks in the last hour. Timeline from this run:
catalogue/1118-2122) ontoa603c7f7(only conflict: the generateddocs/DEVICE_REGISTRY.md; regenerated withscripts/regen-registry-doc.js), force-pushed88c14f42→8cf7efa9. Eight statuses were created on the new head inside 20 s and all went green (Node 18/20/22 59/51/53 s).63dc0b56on an 8/8-green head withmergeable: true.auto-rebase.yml:7eff4559b5bc05462fd93023a1cd97da7290295e0b23437703ce6c46ebaefdddSo the accounting is now: one merge costs four CI runs, and none of the four can be merged until its checks exist again, which only a close+reopen (or a new push) produces — and the next merge wipes them again. The queue can be drained one PR per cycle at best, and only by paying 4 wasted workflow runs each time.
Nothing in the PRs fixes this;
auto-rebase.ymlneeds to push with a token whose push triggers CI, or dispatch CI explicitly after its push (a plainworkflow_dispatchwrites zero statuses — measured on run 7058,/statuses=[]). That is a one-line change to a workflow file, which is Jay's to make — I am not permitted to edit.forgejo/workflows/.Also this run: #2032 closed as superseded — #2036 (
80bb55fe) landed the same scoping fix and an equivalentdocs/__stray_stale.mdregression block, and #2030 closed with it.ebaefdddb2to9df7b89b0922:07Z — fourth reproduction, and the effect is now measurable, not just predictable.
Merged three green PRs back-to-back in one burst (#2043 →
090a1a96, #2042 →eabef2cc, #2041 →5e6bc405). Within ~60 s the push-triggeredauto-rebasepass had force-pushed every remaining head onto the new main and reset all five to zero statuses:55316211→816c1f38b5bc0546→1169554da1cd97da→bf3f22200b234377→6320501eebaefddd→9df7b89bThe important number: three merges cost one rebase pass (5 CI matrices), not three passes (15). A rebase after main moves is unavoidable, so the cost per merge is
batch / merges_per_burst. Burst-merging every green PR in the same tick is the cheapest way to drain the queue; merging one PR per tick pays a full batch for each.Secondary cost, visible right now: the rebased heads sit at
total_count: 0for minutes afterwards because the single runner is still working through main's own matrix (Node 18/20/22 allrunning). Zero statuses on a fresh head is "queued behind main", not "CI lost it" — I read that as red for merge purposes and wait for the next tick.9df7b89b09toa0333c1c22a0333c1c22to1682a729811682a72981to24344236ad24344236ad030e397fe4Needs Jay's decision — prose conflict in
docs/CLI.md, not mechanicalThis branch was green on all three Node legs at
030e397f, but it can no longer be merged:#2035and then#2037(commit06da3f7d, "the EtherChannel receipts are two sessions, not one") landed in the same paragraph ofdocs/CLI.mdwhile this PR was open. A rebase onto06da3f7dgives one conflicted file,docs/CLI.md, with two hunks:=======
sw1(config-if)#exit
sw1(config)#interface Gi1/0/11
<<<<<<< HEAD (#2037,
06da3f7d)The mode belongs to the group, not to the member: typed on a port that is already in that group, the line
re-modes every member (...) rather than refusing the port, because the cap is checked when a port joins.
======= (this PR)
The refusal is asked of a port that is not in the bundle yet - Gi1/0/11 above (...) The cap is compared
on the joining path only -
_channelGroup()skips it when the port is already a member - so a static bundlepast the cap is still reachable through the mode conversion. That half is the engine's to close; [...]
surgery run 2026-11-19 03:0xZ — needs Jay's decision (docs text is now stale, not a rebase).
origin/mainf1d8cfa5(PR #2069, "etherchannel: a group too wide for its new mode is refused, not converted") landed the engine half and a docs/CLI.md paragraph at the later occurrence of this example. This PR's added paragraph atdocs/CLI.md:2096-2106still says:f1d8cfa5closes exactly that half:engine/cli.jsnow refuses the mode conversion and the page documents the refusal (% Channel group 1 not moved to mode on: it holds 9 members …).So the rebase conflict in
docs/CLI.mdcannot be resolved mechanically — resolving it either drops this PR's Gi1/0/11 example or keeps a sentence main has just made false. It needs the author (or Jay) to rewrite the closing paragraph against the new behaviour; the Gi1/0/11 example itself is still a good edit.I did not rebase, merge or close it. It is green on
d0579f8cand 20 behind main;mergeable=falseis the real conflict above, not a stale computation.surgery run 2026-11-19 04:0xZ — half of this PR has landed, half has not; still Jay's text call.
Probed
engine/cli.json current main06830e13(9 LACP members, thenchannel-group 1 mode ontyped at a member):So the engine gap this PR argues for — "the cap is compared on the joining path only ... that half is the engine's to close" — is already closed, by
f1d8cfa5("etherchannel: a group too wide for its new mode is refused, not converted (#2033) (#2069)"), which names this PR and adds the guard atengine/cli.js:3521, a test, and main's "The mode belongs to the group ..." paragraph. That paragraph of this PR is now wrong on the fact, not just on wording.What still stands is the first half: main's join-refusal fence prints
sw1(config-if)#channel-group 1 mode on→% Gi1/0/9 not added, but the port in that context is already the ninth member, so the block does not replay — the console gives the mode-change refusal instead. This PR'sexit/interface Gi1/0/11lines are the fix for that.Resolution, for the author (not surgery — it drops parts of two hunks): keep main's mode-change paragraph and its refusal block whole; take this PR's two prompt lines for the join fence; drop this PR's closing paragraph. One commit, then it goes green and merges.
Conflict as it stands: 1 hunk in
docs/CLI.md. Needs Jay's decision only because deleting a side of a hunk is a text call, not a mechanical rebase.Needs Jay's decision — the rebase conflicts are a behaviour question, not a mechanical one.
Rebased onto
main(8349c59d): two conflicting hunks indocs/CLI.md, both inside "A bundle carries eight links; the rest wait as(H)" (at themode onrefusal transcript, and at the paragraph explaining what happens when the line is typed on a port already in the group).Why this is not mine to resolve: the premise of this PR — "the cap is compared on the joining path only —
_channelGroup()skips it when the port is already a member (alreadyIn) — so a static bundle past the cap is still reachable through the mode conversion. That half is the engine's to close" — is no longer true on main. #2069 (f1d8cfa5, "etherchannel: a group too wide for its new mode is refused, not converted (#2033)") closed that half:engine/cli.js:3521now hasif (existing && alreadyIn && retuning && held > cap)and refuses the mode change with% Channel group 1 not moved to mode on: it holds 9 members, … Nothing was changed.Main'sdocs/CLI.mdalready prints that refusal, and its own transcript types the join refusal atGi1/0/9.So the PR now asserts something the engine does not do any more, and relocates the transcript to
Gi1/0/11. Merging it as written would leave the page describing a gap that #2069 removed.Decision: one of
Gi1/0/11relocation, which main does not need), orOne thing worth a look while you are there: main's
Gi1/0/9block still reads% Gi1/0/9 not added: channel group 1 already holds 8 members, but in this section's stateGi1/0/9already joined the LACP bundle (10 members), so post-#2069 that line would answer with the mode refusal, not the join one.Verified on the rebased tree:
syntax-check443 files OK,check-control-bytesclean,audit-port-typesclean,npm test364 suites / 12,420 assertions green. Not pushed — no point putting a semantic answer on the branch.The engine half of this landed in #2069, and
mainnow carries the mode-change refusal paragraph and example. The transcript hunk here was written before that and does not apply: its Gi1/0/11 line names a group of eight while the section builds ten, and the message the console gives there isholds 10 members, and ... would make 11. The replayable version of that hunk — the Gi1/0/9 mode-change refusal, both standby receipts, and the Gi1/0/11 join refusal — is in #2112, with a replay test intest/docs-cli-transcript.test.js. This branch conflicts withmainand is superseded on both halves.Closed as superseded — #2112 landed the same paragraph, written against the console.
a5baefc6(#2112, "the eight-link example is typed at a port that is already in the bundle") rewrote the "A bundle carries eight links; the rest wait as(H)" example indocs/CLI.mdto match what the engine actually prints:The engine half this PR argued was open was already closed by
f1d8cfa5(#2069); #2112's version is consistent with it and has a transcript test (test/docs-cli-transcript.test.js, 9 checks) asserting every page line is a line the console prints.Verified before the merge: clean merge into
9574bdd5, transcript test 9/9, citation check clean (356 docs, 19 judged), CI 8/8 green onf1649220.Reopen if you'd rather keep this PR's wording — nothing of it was merged.
Pull request closed