Skip to content
Merged
Show file tree
Hide file tree
Changes from all commits
Commits
File filter

Filter by extension

Filter by extension


Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
4 changes: 3 additions & 1 deletion AGENTS.md
Original file line number Diff line number Diff line change
Expand Up @@ -212,7 +212,9 @@ peers granted (groupnet's `consistency::lease`, tier T3). A write is **fenced**
its origin call: it publishes `IndexOp::Pending` and waits until every lease-holder has
applied it or lapsed (`WriteSync::fence`), and each peer answers the key from the origin
until the outcome event settles it (`index::pending`); a writer's next life hands
anything it left open to the origin reconciliation. A write ends when every
anything it left open to the origin reconciliation. A peer that joined or gapped after
an announcement answers every key from the origin until the floor each event carries
(the writer's oldest open write) passes the first event it saw. A write ends when every
lease-holder has either applied the outcome (one ack round) or had its lease lapse
in the writer's engine (bounded by `D`, on the straggler's own clock). That replaces the
old view-stability heuristic (`Stability`/`settled()`), which survives only for
Expand Down
32 changes: 16 additions & 16 deletions Cargo.lock

Some generated files are not rendered by default. Learn more about how customized files appear on GitHub.

2 changes: 1 addition & 1 deletion Cargo.toml
Original file line number Diff line number Diff line change
Expand Up @@ -24,7 +24,7 @@ blake3 = "1.8.5"
bytes = { version = "1", features = ["serde"] }
futures = "0.3"
getrandom = "0.4"
groupnet = { git = "https://github.com/napbat/groupnet", rev = "30407c0bb057a7f12bf1520397f7398b38f97103", default-features = false, features = ["runtime", "consistency-leases", "consistency-volatile-recovery", "consistency-volatile-bootstrap-bulk", "dns", "tcp", "udp"] }
groupnet = { git = "https://github.com/napbat/groupnet", rev = "e7c7b38cde46d338d54e4a7bb978dd1052e6fc9e", default-features = false, features = ["runtime", "consistency-leases", "consistency-volatile-recovery", "consistency-volatile-bootstrap-bulk", "dns", "tcp", "udp"] }
http = "1"
http-body-util = "0.1"
# Already under hyper-util/s3s; direct so the metrics endpoint can build its own responses.
Expand Down
16 changes: 12 additions & 4 deletions README.md
Original file line number Diff line number Diff line change
Expand Up @@ -108,10 +108,11 @@ gossiped advertisements refresh admitted neighbors' addresses; discovering anoth
member does not grant admission. The Helm chart configures every ordinal, including
later-starting peers. Groupnet starts one managed network asynchronously and closes
its links and resolver after the planned mutation drain and feed seal.
The ring is bounded by both
event count and encoded bytes: sustained long-key writes shorten its replay window and
surface the same explicit gap instead of growing one state entry past the transport
envelope and silently arresting peer application.
The ring is bounded by both event count (4096) and encoded bytes (one 60 kB gossip
frame), and the bytes bind first: a `strong` write is two events, about 230 bytes for a
44-byte key, so a peer about 260 such writes behind gets a gap. Longer keys shorten
that window and surface the same explicit gap instead of growing one state entry past
the transport envelope and silently arresting peer application.
After a `strong`/`strong-acks` write has a real all-readers acknowledgement (or every
straggler has lost serving authority), its safe feed prefix is retired while the newest
head remains advertised as an anchor. Healthy feeds therefore carry an anchor plus only
Expand Down Expand Up @@ -329,6 +330,13 @@ box is the long form):
answered **503** without calling the origin (`ack_timeouts`). A write the origin
refused abandons its fence. A node holding a peer's open fence offers no index image
to a joining node.
- **A node that never saw a fence answers every key from the origin until it is
settled.** A node that joins, or gaps, after a write was fenced cannot tell which key
it fences. Every feed event carries its writer's oldest open write, so the first
event such a node applies from that writer shows whether one predates what it saw;
until a later event shows none does, the node answers every read from the origin,
and offers no index image. If the writer crashes first, the node is waiting on it
as a lease granter anyway, and its next life's gap rebuilds the index.
- **A feed gap discards the gapped node's index.** The node missed writes it can no
longer order its rows against, so it keeps nothing for a peer-image install to carry
forward; the install or the origin rebuild repopulates it.
Expand Down
16 changes: 11 additions & 5 deletions docs/fleet-bootstrap.md
Original file line number Diff line number Diff line change
Expand Up @@ -148,10 +148,10 @@ LIST/HEAD callbacks. A row carries exact key, size when known,
storage class. Peer rows deliberately become **skeletal**: `Content-Type` and
user HEAD metadata are not fabricated; HEAD falls through to origin. The
donor cannot offer an image while any unresolved per-key uncertainty, or any
peer's pending `strong` write, exists. A follower carries its own unresolved
uncertainty and the pending writes it holds across the swap (see Follower
handoff); such reads keep going to the origin until exact HEAD reconciliation,
the writer's outcome, or origin rebuild.
peer's pending `strong` write, exists, nor while a peer may hold one it never
saw. A follower carries its own unresolved uncertainty and the pending writes it
holds across the swap (see Follower handoff); such reads keep going to the
origin until exact HEAD reconciliation, the writer's outcome, or origin rebuild.
Absence and LIST can be served locally only after a complete image, continuous
native handoff, independent frontier check, and the existing lease/domain
gate. Direct external S3 mutations still lack a fleet event and remain outside
Expand All @@ -164,6 +164,10 @@ unknown schema. Encoded bytes, decoded rows, native overlap, donor suffix,
and each physical reply clone use distinct shared `ByteAdmission` permits.
The decoder precharges conservative hash-table and tree-node capacity before
allocation. This bounds private image ownership, not allocator RSS exactly.
A follower hashes and decodes each chunk as it lands, so verification only
finishes both: Groupnet's worker refreshes the donor's claim between its
steps, and one long decode after the last chunk let that claim lapse on the
follower under a live donor.
The exact `DonorJournal::storage_bound` must fit the configured Suffix and
global caps before cloning the image. Tests price the configured bound and
measured live usage; a cap refusal falls back to origin without publishing a
Expand Down Expand Up @@ -327,7 +331,9 @@ token epoch. The fenced key stays origin-read, and a LIST whose prefix covers it
stays origin-served, until that key's own HEAD resolves it in the installed index
or a later definitive mutation supersedes it, as on a node that never
bootstrapped. No later fence can reuse a carried token. Pending peer writes cross
the swap unchanged and are settled by their writers' later events.
the swap unchanged and are settled by their writers' later events. So does what
the follower knows of each writer's feed: a write announced before the follower
first saw that feed keeps every key origin-read until the writer's floor passes it.

**No live effect is lost.** A native effect the barrier does not cover
misaligns the cuts and answers `NativePending`; the stage waits for a later
Expand Down
Loading
Loading