Phase 5 design decision: distinct-tag active connector switch (advanced mode, design-only) - #24
Merged
Merged
Conversation
…ed mode) Replaces the design skeleton with the committed scope per the owner's directive, reviewed to consensus by Grok 4.5 (4 rounds) and GPT-5.6-sol (4 rounds), both final verdicts PROCEED: 1. Same-tag native HA stays the default and recommended model; the monitor stays read-only, permanently. 2. Apply mode is a separate ADVANCED mode: the distinct-tag active connector switch — operator-invoked, one-shot, auditable forced selection of which connector pool (tag) serves the AI domain set, via plan/apply-plan. 3. For online-but-bad forced selection: wrong egress IP, degraded provider path, provider quota, manual evacuation — cases native oldest-first cannot see. No health-driven watcher in v1.3. 4. Composes with (never replaces) Tailscale-native HA: native HA keeps operating inside each pool; the switch moves the domain set between pools. 5. Pre-implementation gate: the operator docs (deployment models, migration both directions, three-layer rollback, warnings) must be MERGED before implementation work opens. Mechanism decisions recorded: a connector-scoped policy_tool.py planning subcommand (working name connector-plan) with allowlisted --switch-to (exact one-element replace, --expected-from compare-and-swap for the scripted path, verbatim from-array in the manifest) and --declare (tagOwners + autoApprovers + DNS grant only) operations — the ordinary full-merge plan path stays byte-identical; planner-side snapshot preconditions (TOCTOU- closed); pool-pair configuration invariants + drift refusal + dual-tag warnings; own connector-switch-state.json (schema_version 1, decoupling the multi-fallback schema-v2 bump); advisory cooldown; fail-closed readback with no automatic second write; migration paths made executable (declare-first; retag reclassified as maintenance-window with remaining-node precondition). Roadmap: flagship item rewritten to the decided scope + acceptance; the multi-fallback item's coordination sentence now records the decoupling. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
F-e-u-e-r
force-pushed
the
design/phase5-connector-switch
branch
from
July 15, 2026 19:28
fa3358d to
245a10f
Compare
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
What
Docs-only design-decision PR (the revised Phase 5 flagship scope). Replaces the
docs/design/connector-failover-apply.mdskeleton with the committed design and updates the two v1.3 Roadmap items.The owner's five binding points, all encoded: (1) same-tag native HA stays default/recommended; (2) apply mode is a separate advanced mode — the distinct-tag active connector switch; (3) scoped to online-but-bad forced selection (wrong egress IP / bad provider path / quota / manual evacuation); (4) not a replacement for Tailscale-native HA (they compose; no auto-watcher in v1.3); (5) explicit docs, migration, rollback, and warnings are gated BEFORE implementation.
Review
Tri-model consensus reached before this PR was opened: Grok 4.5 (high) — 4 rounds, final PROCEED; GPT-5.6-sol (max) — 4 rounds, final PROCEED. 13 design findings raised across rounds (2 Blocker/Major-class mechanism gaps, TOCTOU, migration executability, drift semantics, etc.), every one incorporated and closure-confirmed by both reviewers.
Verification
Docs-only:
check_docs_links(41 files OK), README parity OK, docs tests OK. No behavior change; implementation is gated per the design's pre-implementation gate (operator-docs PR merges before any implementation PR opens).🤖 Generated with Claude Code