Repository navigation
feat(coord): NBD slots are a placement dimension; chart nbdsMax for two devices per guest (ADR 0112 addendum, phase 4) - #1607
Merged
Conversation
|
The latest Buf updates on your PR. Results from workflow CI / buf (pull_request).
|
nikhilunni
force-pushed
the
feat/swap-nbd-slots
branch
from
October 7, 2026 18:00
5718cc4 to
21e72c6
Compare
nikhilunni
added this pull request to stack #1609
October 7, 2026 18:12
nikhilunni
force-pushed
the
feat/swap-nbd-slots
branch
from
October 7, 2026 19:00
21e72c6 to
f329913
Compare
nikhilunni
force-pushed
the
feat/swap-nbd-slots
branch
from
October 7, 2026 19:23
f329913 to
1ad50cd
Compare
nikhilunni
force-pushed
the
feat/swap-nbd-slots
branch
from
October 7, 2026 19:50
1ad50cd to
13fe92a
Compare
nikhilunni
force-pushed
the
feat/swap-nbd-slots
branch
from
October 7, 2026 20:16
13fe92a to
16c12c1
Compare
nikhilunni
force-pushed
the
feat/swap-nbd-slots
branch
from
October 7, 2026 20:19
16c12c1 to
590a6a3
Compare
…wo devices per guest (ADR 0112 addendum, phase 4) A swap-enabled guest takes two NBD slots. Placement now counts them so exhaustion is `NoFit`, not a wait inside the host allocator. - The host heartbeat reports `nbd_slots_total`, `nbd_slots_in_use` (claimed, parked and quarantined; never warm or validation claims) and the attached device count per sandbox. Migration 0125 adds the host columns and `sessions.nbd_slot_need`, backfilled from the image config or a swap-bearing snapshot. - Every placer that lands a chunked disk passes the slot gate: create, queued create, resume, affinity, pinned host, capture placement and teleport admission. The ledger reserves only the remaining need for a sandbox the heartbeat has not yet reported, so a device is never counted twice. A host with NBD off is exempt (file fallback). No-fit details report `nbd_slots`; queue fit classes include the need. - Fleet views expose both counters (proto fields 35 and 36). - Chart defaults: `nbdsMax` and `warmSlots` 128. A change needs a node recreate because the kernel reads the module parameter at load. Conformance scenarios cover host placement with a partial attach, teleport reservation on both sides, and capture placement and reassignment, on both stores. Pure placement tests cover every tier and the no-fit reason. Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01UHxL6gj4o8EvxYpgtwEWaM
nikhilunni
force-pushed
the
feat/swap-nbd-slots
branch
from
October 7, 2026 20:20
590a6a3 to
5a9513b
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.
Summary
Phase 4 of the chunked-swap design (ADR 0112 addendum): NBD slots become a placement dimension, and the chart defaults size the slot pool for two devices per swap-enabled guest. Stacked on #1605 (phase 3). This is the last code phase; the dev-VM measurements of swap-in latency and capture upload cost are an operator step before the fleet roll.
What changes
HostUtilizationgainsnbd_slots_total,nbd_slots_in_use(claimed, parked and quarantined slots; warm and validation claims excluded) and the attached device count per sandbox. The allocator keeps an atomic claimed counter, incremented at every slot construction and decremented in its drop.sessions.nbd_slot_need(one root slot plus one for swap). Existing rows are backfilled from the image config they were created under or from a snapshot that carries a swap manifest; a live guest is counted by its host's sample either way. The reservation SQL uses the same session, capture-job and teleport arms as the memory ledger and subtracts devices the heartbeat already reports, so nothing is counted twice.nbd_slots_total == 0is exempt (file fallback). No-fit details reportnbd_slots; queue fit classes include the need.HostView(proto fields 35, 36; TypeScript bindings regenerated).nbdsMaxandwarmSlotsdefault to 128, with the node-recreate note in all three values files.Posture
Resume and teleport pickers read the ledger without a transaction, as the memory ledger does today; the host allocator remains the final arbiter. Recorded in the ADR addendum.
Tests
nbd_slots; queue classes split on need; allocator counts across acquire, quarantine, reclaim and drop.Validation
just checkgreen.helm lintpasses;just gen-protoreproduces the bindings.🤖 Generated with Claude Code
https://claude.ai/code/session_01UHxL6gj4o8EvxYpgtwEWaM