diff --git a/docs/CURRENT-STATE.md b/docs/CURRENT-STATE.md index 98b9e4e..419e38e 100644 --- a/docs/CURRENT-STATE.md +++ b/docs/CURRENT-STATE.md @@ -104,8 +104,17 @@ > D-101's "PXE v4-first" and D-139's "provider dual because internet=v4" are overstated/conflated. > CONFIRMED still-hard: juju LP#1723240 (container VIPs, D-141 holds) + external v6 internet. NEW: Ceph > one-family-per-daemon constrains DEC-24 v6 replication. CHALLENGES prior framings -> surfaced for -> operator reconsideration, NOT ruled (GA-R5). Note: +> operator reconsideration. Note: > `docs/audit/container-elim-pass/pass6-ipv6-feasibility-note-20260810.md`. +> **RULED 2026-08-10 (GA-R5) -> D-101 AMENDMENT:** on the fresh evidence, metal-admin node commissioning +> for the 10.13 redeploy **TARGETS IPv6** (UEFI + DHCPv6 + HTTP boot), operator utterance "Adopt v6 +> (gated on live test)" -- GATED on a one-node canary (DHCPv6->UEFI-HTTP-boot->READY captured) with v4 +> fallback before any fleet commit (hard rule 2). Amends D-101's "PXE v4-first" (an unmeasured judgement); +> metal-admin plane stays dual-stack (D-139 unchanged); moots the D-101-wording DOCFIX. provider-public +> needed NO reconsideration (D-101 already rules it dual-stack + native v6 GUA). Ruling: +> `docs/design-decisions.md` D-101 AMENDMENT 2026-08-10. Owed at redeploy: UEFI node-VM loader + +> DHCPv6/HTTP-boot build + the canary gate. Still-owed DOCFIX: the stale OVN "No" row in +> `charm-ip-family-compatibility.md`. > > **dc0 checkpoint scope (operator 2026-08-08): "activate + smoke-test"** -- networks + Octavia (1 test > LB) + Designate (1 test zone) + wrap gates (cloud-assert BOM, controller backup, verify-live diff --git a/docs/design-decisions.md b/docs/design-decisions.md index df6f924..441fdaa 100644 --- a/docs/design-decisions.md +++ b/docs/design-decisions.md @@ -2742,6 +2742,49 @@ --- +## D-101 -- AMENDMENT (2026-08-10): metal-admin commissioning TARGETS IPv6 (gated on a live canary), reconsidering the "PXE is v4-first" clause + +**Status: RULED 2026-08-10 (GA-R5).** Reconsiders D-101's `IPv4 retained: MAAS PXE / provisioning +(PXE is v4-first ...)` clause on FRESH capability evidence -- the object was NOT to follow the prior +ruling but to test whether IPv6 is actually feasible on this link (operator direction 2026-08-10). +The prior clause was an unmeasured JUDGEMENT; the fresh research (pass6, upstream docs/source/bug +trackers, deployed MAAS 3.7.2) found: DHCPv6 commissioning has NO MAAS-side blocker at 3.7.2 (the two +historical blockers LP#1718209/LP#1939034 closed 2017 / MAAS 3.1); MAAS rack<->region over IPv6 is +FEASIBLE-NOW (source prefers `AF_INET6`; LP#2020142 fixed in 3.5.0); the ONLY hard blocker is +legacy-BIOS PXE (Intel Boot Agent / PXE-2.1, IPv4-only firmware) -- which UEFI HTTP boot avoids, and +the VR1 node VMs' loader is ours to set. Record: `docs/audit/container-elim-pass/pass6-w1-maas-ipv6- +feasibility.md` + `pass6-ipv6-feasibility-note-20260810.md`. + +**Question as presented.** "Given the fresh capability evidence, should the 10.13 redeploy adopt IPv6 +metal-admin commissioning (UEFI + DHCPv6 + HTTP boot), or keep v4 commissioning? Adopt = target it, +gated on a live canary test with v4 fallback, not commit the fleet blind." **Operator selection, exact +utterance: "Adopt v6 (gated on live test)".** + +**RULED.** For the 10.13 redeploy, metal-admin node commissioning **TARGETS IPv6** (UEFI-boot VMs + +DHCPv6 + HTTP boot). This AMENDS D-101's "PXE is v4-first" clause -- v4-first was a build choice and an +unmeasured judgement, not a capability ceiling. The metal-admin PLANE stays dual-stack (D-101 07-09 +amendment / D-139 unchanged); this ruling concerns the COMMISSIONING/boot-path family only. + +**GATE (hard -- no fleet commit on an inferred value, hard rule 2).** IPv6 commissioning is proven on +**ONE canary node** on this stack (DHCPv6 lease -> UEFI HTTP boot -> enlist -> commission -> READY, +captured) BEFORE the full fleet is committed; **v4 commissioning is the fallback** if the canary fails. +The ungated full-fleet option was explicitly NOT chosen. + +**Build implications (owed at redeploy design time; logged, not built):** VR1 node-VM definitions set +a UEFI (OVMF) loader; the metal-admin commissioning range gains DHCPv6 + the MAAS HTTP-boot path; the +canary test + its captured evidence is a new gate. Legacy-BIOS nodes are OUT (would reintroduce the +one real firmware blocker). Interacts with D-144's owed metal-admin/power work (the flat rebuild +already redefines these node VMs). + +**Reconciliation.** AMENDS D-101 (the PXE-v4-first clause only; the IPv6-primary principle and the +family matrix are UNCHANGED -- this ADVANCES "v6 wherever possible"). CONSISTENT-WITH D-139 (metal-admin +plane stays dual-stack; only the boot-path family shifts). Supersedes the pass5 "PXE stays v4 by ruling" +framing and moots the standalone "soften D-101 PXE wording" DOCFIX (this amendment IS that correction, +now with a ruling behind it). **Roosevelt-delta:** the rehearsal proves the v6-commissioning path a +bare-metal Roosevelt build would want (UEFI HTTP boot over v6); the legacy-BIOS caveat is moot on +modern server firmware. provider-public needed NO reconsideration -- D-101 already rules it dual-stack +with native v6 GUA ext_net; the v4 there is retained for external v4-internet reachability, unchanged. + ## D-102: IPv6 tenant addressing and MTU sub-policy (VR1) **Status:** MERGED INTO D-101 (2026-07-09, operator ruling). Content below is retained for numbering/history only, per this repo's append-only decision-record discipline -- D-101 is now the current, authoritative text for this sub-policy. Do not amend the body below; amend D-101 instead.