ci: prove the relay-driven mesh lifecycle — discover, join, infer, deny — with real nodes - #3862
Conversation
152f526 to
c8239e7
Compare
|
cool ci. if the smoke is flaky on shared runners, maybe pin a longer join timeout in the workflow rather than retrying the whole job |
c8239e7 to
8c56830
Compare
59bef12 to
4281774
Compare
|
This is the right shape and it covers something nothing else does — relay-as-control-plane, with no out-of-band token. The membership/binding/admission half (phases 1–3, 6) is genuinely unique to Buzz and worth landing on its own merit. Also worth noting the stated draft gate is already met: Mesh Lifecycle has three green runs on Actions now. I've spent the last few days chasing two shared-compute bugs to root cause (#3909, and Mesh-LLM/mesh-llm#1126 / #1129), so I wanted to write down where this harness would and wouldn't have caught them — and a concrete, cheap way to extend it. Take or leave any of it. What it would not have caught todayI grepped the diff rather than going on impressions:
That isn't a criticism of scope — it's that both bugs live in the timing and agent layers, and this harness deliberately sits below them. The inference phase is Suggestion 1 — a stalled-node phase (this is the cheap, high-value one)The bug behind #3909 was that a node which is bound but HTTP-silent — loading weights, pulling layers, or mid-prefill — was read as dead and force-restarted the whole app. Measured on a live 26B node: The valuable insight for CI is that this is a timing bug, not a size bug. You don't need a big model or a GPU to reproduce it — you only need the node to stop answering HTTP while its port stays bound. The harness already spawns and holds each node as a // phase N: a busy/loading node must not be classified as dead
kill(serve_pid, SIGSTOP);
assert!(tcp_connect(api_addr).is_ok()); // port still bound -> alive
assert!(http_get(api_addr, "/v1/models").is_timeout()); // ... but silent
kill(serve_pid, SIGCONT);
assert!(http_get(api_addr, "/v1/models").is_ok()); // recovers
One honest caveat: If it's worth asserting the Buzz-side classification directly rather than just the wire behaviour, Suggestion 2 — a prose-only agent phaseThe second failure was subtler: a small model answers in prose and never calls the send tool, so the turn ends A deterministic stub is enough to pin the contract, and needs no model at all:
Reading relay state is the part I'd push hardest. During my own testing the agent's self-report and the harness logs both told me a reply had landed when it hadn't; only an independent authenticated read was trustworthy. On soakDeliberately not suggesting soak in CI. The restart loop needs hours to surface naturally, which is hopeless for a PR gate — and it's already covered in real life: I've had the desktop app sharing a 26B model for 2 days 4 hours continuous with zero restarts on the post-#3909 build (previously it bounced within minutes of a cold 26B share). That's the right place for soak. CI's job is the deterministic version of the same failure, which is what Suggestion 1 buys. Suggestion 3 — multi-node MoA, as a separate thingDifferent question, worth separating: does routing to the virtual My honest view is it doesn't belong in this workflow, or possibly in Buzz CI at all:
Better as either a mesh-llm-side benchmark ( Multi-hostEverything here is single-runner over loopback, which is right for CI. Real cross-host QUIC, real RTT and contention need actual separate machines and belong in a pre-release lab tier rather than a PR gate. Two things I'd carry over from doing that by hand: pin the model per node (an |
|
🤖 (from micn's agent) Good call — addressed in 4281774: the model-visibility windows are now tunable via |
|
🤖 (from micn's agent) Thanks for the root-cause writeup — this is exactly the analysis we wanted. I ran your comment plus the harness internals past an independent expert review; summary of where we landed: On "would it have caught #3909 / the prose-reply drop": agreed, no. Both live in layers this harness deliberately sits below (desktop supervisor timing; ACP agent behavior). No pushback there. On Suggestion 1 (SIGSTOP stalled-node phase): we're not adding it to this PR, for a sharper reason than cost. The expert's key observation: before #3909, On Suggestion 2 (prose-only agent + independent relay read): agreed it's valuable, and it's a follow-up — partly because it first needs a product decision the test would encode: when an agent never tool-calls, is the contract auto-publish, an explicit failed outcome, or "reminder then send"? #3763's guard nags but can't compel; a stub test has to assert something specific. Your "read the event back with a different authenticated identity, correlate by e-tag" principle is the part we'll definitely keep — the expert independently flagged self-reports as untrustworthy for exactly the reason you hit. On Suggestion 3 (MoA) and soak: agreed on both — MoA lift belongs in mesh-llm benchmarks or a periodic lab job; a cheap One more gap the expert added that's worth recording for follow-ups: the harness reimplements the binding/verification logic, so publisher and verifier could drift together and stay self-consistent while the desktop breaks — the keep-in-sync comments are the only guard. Extracting the pure binding/discovery logic into a shared crate both sides link would close that properly, and is probably the highest-value follow-up of all of these. Scope of this PR stays as-is: the relay-as-control-plane lifecycle (phases 1–6). Follow-ups, roughly in value order: shared binding crate → prose-agent contract test → optional suspension/recovery phase → |
…inference, denial
Run the full Buzz shared-compute join story in CI with three real mesh-llm
node processes on one runner, using the Buzz relay as the control plane
(no hand-carried invite tokens):
1. membership — A and B added via buzz-admin (kind:13534 roster); C is not
2. advertise — members publish client-signed kind:30003 discovery notes
with owner + endpoint binding signatures (the exact payload
the desktop coordinator publishes)
3. trust — the serve node derives its admission allowlist from the
relay (statuses ∩ roster) and starts TrustPolicy::Allowlist
4. join — the client verifies bindings + membership and dials the
relay-discovered endpoint
5. infer — chat completion routes client → QUIC → serve node (CPU,
SmolLM2-135M)
6. deny — the stranger gets zero statuses from the membership-gated
relay AND is refused mesh admission with the leaked endpoint
Mirrors the shape mesh-llm's own CI uses for its two-node smokes (tiny CPU
model, one runner, real QUIC mesh), swapping mdns/token bootstrap for the
relay-driven lifecycle. start-relay-for-tests.sh gains opt-in NIP-43
membership gating env passthrough.
Signed-off-by: Michael Neale <michael.neale@gmail.com>
- Stranger relay denial now requires the relay's own NIP-42 membership
rejection ('not a relay member'); successful auth or any other connect
failure fails the run instead of passing as denial.
- Differential admission proof: after the stranger's attempt, the trusted
client must still route inference (VERIFY_AGAIN over stdin), so a dead
serve node can no longer masquerade as an admission denial. Stranger
child exit status is also checked.
- Allowlist gate now requires the exact expected {A, B} owner-id set
(MESH_EXPECTED_OWNERS carries ids, not a count).
- Hard deadlines everywhere: child stdout is pumped through a reader
thread (recv_timeout), wait_child kills on timeout; outer dial retries
reduced to 3 (mesh-llm already retries internally per attempt).
- Workflow: pin ubuntu-24.04, trigger paths + cache key include
Cargo.lock (mesh pin bumps roll the runtime cache), add buzz-admin/
test-client/ws-client paths, upload the harness log artifact too.
- Scope note in the module docs: this is an independent protocol harness
mirroring the desktop payloads byte-for-byte, not a link against
desktop discovery code.
Signed-off-by: Michael Neale <michael.neale@gmail.com>
…dows - zizmor ref-version-mismatch: actions/checkout comment now names the exact tag (v6.0.3) matching the pinned SHA. - Per review suggestion: model-visibility windows are tunable via MESH_CLIENT_WINDOW_SECS / MESH_STRANGER_WINDOW_SECS so slow shared runners can pin longer windows instead of re-running the whole job. Signed-off-by: Michael Neale <michael.neale@gmail.com>
8c1b7b4 to
5ad9ac5
Compare
* origin/main: (26 commits) fix(desktop): skip native notifications outside app bundles (#5004) ci: prove the relay-driven mesh lifecycle — discover, join, infer, deny — with real nodes (#3862) fix(desktop): virtualize channel member lists (#4991) fix(desktop): enforce owner-only access in internal builds (#4053) test(desktop): match attachment button label (#4993) fix(acp): pace observer telemetry at 1/s with per-channel batch envelopes (#4917) fix(desktop): enable the content security policy (#4614) fix(mobile): merge relay recounts with locally seen thread replies (#4633) fix(desktop): enable message editing in Inbox (#2198) relay: fuzz WebSocket 1012 restart-close timing on graceful drain (BUZZ_DRAIN_JITTER_MS) (#4542) fix(desktop): outline the selected community (#4969) fix(desktop): clamp thread panel to channel surface (#4965) style(messages): increase username contrast (#4948) fix(desktop): rename generic attachment action from 'Attach image' to 'Attach file' (#2381) (#4304) fix(reactions): support max-length custom emoji (#3833) feat(desktop): allow leaving your final community (#3621) fix(buzz-agent): recover from context-window 400s instead of sticking (#4946) docs(persona-pack): fix stale desktop import instructions (#4500) fix(desktop): route macos notification clicks (#4799) feat(mobile): sync themes per community (#3767) ... Co-authored-by: Eva <011987e296fd5006292d2f930b574be47c7801048d1983c46c425d3c95f0cffd@buzz.block.builderlab.xyz> Signed-off-by: Eva <011987e296fd5006292d2f930b574be47c7801048d1983c46c425d3c95f0cffd@buzz.block.builderlab.xyz>
…-agents-nav * origin/main: Refine agent runtime controls (#5026) test(desktop): await thread scroll anchor (#3174) Improve desktop mobile pairing flow (#5024) feat(desktop): show selected community in rail (#5000) fix(desktop): stop rate-limited reconnect backfill from tearing down the authenticated socket (#4990) fix(desktop): skip native notifications outside app bundles (#5004) ci: prove the relay-driven mesh lifecycle — discover, join, infer, deny — with real nodes (#3862) fix(desktop): virtualize channel member lists (#4991) fix(desktop): enforce owner-only access in internal builds (#4053) test(desktop): match attachment button label (#4993) fix(acp): pace observer telemetry at 1/s with per-channel batch envelopes (#4917) fix(desktop): enable the content security policy (#4614) Signed-off-by: Duncan <dcfd242e557282d7a1e2cf2e6877522682f1e5c6156dc92ca7d90eaedd3b0f95@buzz.block.builderlab.xyz>
…n-surface * origin/main: (44 commits) fix(buzz-agent): classify read timeouts distinctly in LLM error messages (#4959) Refine agent runtime controls (#5026) test(desktop): await thread scroll anchor (#3174) Improve desktop mobile pairing flow (#5024) feat(desktop): show selected community in rail (#5000) fix(desktop): stop rate-limited reconnect backfill from tearing down the authenticated socket (#4990) fix(desktop): skip native notifications outside app bundles (#5004) ci: prove the relay-driven mesh lifecycle — discover, join, infer, deny — with real nodes (#3862) fix(desktop): virtualize channel member lists (#4991) fix(desktop): enforce owner-only access in internal builds (#4053) test(desktop): match attachment button label (#4993) fix(acp): pace observer telemetry at 1/s with per-channel batch envelopes (#4917) fix(desktop): enable the content security policy (#4614) fix(mobile): merge relay recounts with locally seen thread replies (#4633) fix(desktop): enable message editing in Inbox (#2198) relay: fuzz WebSocket 1012 restart-close timing on graceful drain (BUZZ_DRAIN_JITTER_MS) (#4542) fix(desktop): outline the selected community (#4969) fix(desktop): clamp thread panel to channel surface (#4965) style(messages): increase username contrast (#4948) fix(desktop): rename generic attachment action from 'Attach image' to 'Attach file' (#2381) (#4304) ... # Conflicts: # desktop/src-tauri/src/commands/mod.rs
Absorbs the relay half that shipped separately in #5133 (squash commit ad92335): the kind:30179 ingest acceptance hunk in crates/buzz-relay/src/handlers/ingest.rs was byte-identical on both sides, so this merge removes all relay-side changes from this PR's diff. #4999 now carries only the desktop + buzz-core codec half. No rebase, no force-push — history preserved per operator instruction. Co-authored-by: Tyler Longwell <tlongwell@block.xyz> Signed-off-by: Tyler Longwell <tlongwell@block.xyz> * origin/main: fix(bench): mention the orchestrator by pubkey when posting the task (#5136) feat(relay): accept kind:30179 private managed-agent events at ingest (#5133) fix(media): require authenticated reads (#4610) fix(desktop): preserve authoritative agent avatars (#4984) fix(desktop): next/back navigation during key creation onboarding (#4978) Alert community owners and admins when a new key joins (#4900) fix(desktop): prevent sidebar prefs from reverting on stale-localStorage boot (#5086) chore(hooks): run desktop typecheck in pre-push (#5110) feat(identity): recover desktop identity from a signed-in phone (#4845) fix(buzz-agent): classify read timeouts distinctly in LLM error messages (#4959) Refine agent runtime controls (#5026) test(desktop): await thread scroll anchor (#3174) Improve desktop mobile pairing flow (#5024) feat(desktop): show selected community in rail (#5000) fix(desktop): stop rate-limited reconnect backfill from tearing down the authenticated socket (#4990) fix(desktop): skip native notifications outside app bundles (#5004) ci: prove the relay-driven mesh lifecycle — discover, join, infer, deny — with real nodes (#3862) fix(desktop): virtualize channel member lists (#4991) Signed-off-by: Eva <011987e296fd5006292d2f930b574be47c7801048d1983c46c425d3c95f0cffd@buzz.block.builderlab.xyz>
…igration * origin/main: fix(bench): mention the orchestrator by pubkey when posting the task (#5136) feat(relay): accept kind:30179 private managed-agent events at ingest (#5133) fix(media): require authenticated reads (#4610) fix(desktop): preserve authoritative agent avatars (#4984) fix(desktop): next/back navigation during key creation onboarding (#4978) Alert community owners and admins when a new key joins (#4900) fix(desktop): prevent sidebar prefs from reverting on stale-localStorage boot (#5086) chore(hooks): run desktop typecheck in pre-push (#5110) feat(identity): recover desktop identity from a signed-in phone (#4845) fix(buzz-agent): classify read timeouts distinctly in LLM error messages (#4959) Refine agent runtime controls (#5026) test(desktop): await thread scroll anchor (#3174) Improve desktop mobile pairing flow (#5024) feat(desktop): show selected community in rail (#5000) fix(desktop): stop rate-limited reconnect backfill from tearing down the authenticated socket (#4990) fix(desktop): skip native notifications outside app bundles (#5004) ci: prove the relay-driven mesh lifecycle — discover, join, infer, deny — with real nodes (#3862) Signed-off-by: Brad Seiler <seiler@squareup.com> # Conflicts: # deploy/charts/buzz/templates/deployment.yaml # deploy/charts/buzz/tests/render_test.yaml # deploy/charts/buzz/values.schema.json # deploy/charts/buzz/values.yaml
## Buzz Relay release v0.2.1 ### Changes since relay-v0.2.0: - fix(sdk): preserve self-mention p tags in message and forum event builders ([#4975](#4975)) ([`78c87ae20e`](78c87ae)) - feat(desktop): adding rich link previews to messages ([#3818](#3818)) ([`1922d49cb2`](1922d49)) - feat(relay): accept kind:30179 private managed-agent events at ingest ([#5133](#5133)) ([`ad923353a2`](ad92335)) - fix(media): require authenticated reads ([#4610](#4610)) ([`769ac70b74`](769ac70)) - feat(identity): recover desktop identity from a signed-in phone ([#4845](#4845)) ([`6eb65919f1`](6eb6591)) - ci: prove the relay-driven mesh lifecycle — discover, join, infer, deny — with real nodes ([#3862](#3862)) ([`38bf642fcf`](38bf642)) - relay: fuzz WebSocket 1012 restart-close timing on graceful drain (BUZZ_DRAIN_JITTER_MS) ([#4542](#4542)) ([`e14fff74d0`](e14fff7)) - fix(reactions): support max-length custom emoji ([#3833](#3833)) ([`2ea9385015`](2ea9385)) - fix(channels): restrict private-channel invitations ([#4612](#4612)) ([`efe1893dd3`](efe1893)) - fix(workflow): bind trigger author to the signed event ([#4607](#4607)) ([`885bed35ee`](885bed3)) - fix(git): revoke access for banned relay members ([#4608](#4608)) ([`997b8caaa4`](997b8ca)) - Define private managed agent wire protocol ([#4593](#4593)) ([`067c085f37`](067c085)) - perf(relay): index channel-id lookups and skip trace-only reads ([#4647](#4647)) ([`bc9e6528a7`](bc9e652)) - Polish mobile inbox and media flows ([#4512](#4512)) ([`feccf4eabc`](feccf4e)) - fix(git): allow deleting the default branch ([#4297](#4297)) ([`fc598f5f8d`](fc598f5)) - feat(projects): add buzz projects CLI commands (NIP-MP kind:30621) ([#4020](#4020)) ([`b7bb15122e`](b7bb151)) - perf(relay): serve relay-membership checks from the read replica ([#4124](#4124)) ([`ac4fa13b8e`](ac4fa13)) - fix(relay): allow open relays to set their NIP-11 workspace icon (kind:9033) ([#3998](#3998)) ([`5765fc74b7`](5765fc7)) - feat(relay): accept kind:30621 multi-repo projects at ingest ([#3171](#3171)) ([`cb9701cd30`](cb9701c)) - feat(relay): raise hosted community limit to five ([#3829](#3829)) ([`10d5a26414`](10d5a26)) - fix(relay): align NIP-11 max_limit with REQ ceiling ([#3635](#3635)) ([`23f0c26b1c`](23f0c26)) - feat(relay): gate kind 30178 team-catalog reads behind the shared tag ([#3358](#3358)) ([`114d40d9d3`](114d40d)) - fix(db): isolate usage metrics advisory-lock test on scratch DB ([#3670](#3670)) ([`dba97eecd9`](dba97ee)) - perf(presence): reduce heartbeat frequency ([#3783](#3783)) ([`bf139e8d0b`](bf139e8)) - feat(mesh): upgrade embedded mesh to v0.74 and harden shared compute (split 1/2 of #3467) ([#3741](#3741)) ([`4933672eb4`](4933672)) - feat(replica): portable heartbeat-token fence with snapshot-local reader routing ([#3268](#3268)) ([`63496cc1d4`](63496cc)) - fix(git): channel binding tooling + author remediation for unbound repos ([#3626](#3626)) ([`788b3c002b`](788b3c0)) - feat: configure S3 URL addressing style ([#3400](#3400)) ([`7012d86d52`](7012d86)) - feat(tracing): correlate trace IDs in relay logs ([#3608](#3608)) ([`005b5b819a`](005b5b8)) - fix(relay): avoid subscription lock inversion ([#3413](#3413)) ([`22be8bb351`](22be8bb)) - feat(cli): add users set-status command for NIP-38 profile status ([#3253](#3253)) ([`60158fce3e`](60158fc)) - feat(relay): make Postgres pool size configurable, default 50 ([#3191](#3191)) ([`2ce2d71cc3`](2ce2d71)) - feat(tracing): add datastore tracing plumbing ([#2760](#2760)) ([`e94b9aeda0`](e94b9ae)) - feat(invites): add use-limited invite links ([#3141](#3141)) ([`d500c2d5cf`](d500c2d)) - feat(admin): show reported message content in report detail ([#3149](#3149)) ([`f069a85503`](f069a85)) - resolve findings ([#3150](#3150)) ([`9b0f744804`](9b0f744)) - Revert "fix(cli,relay): resolve agents by verified owner" ([#3168](#3168)) ([`a041e2d21e`](a041e2d)) - fix(cli,relay): resolve agents by verified owner ([#2615](#2615)) ([`c3084b36d9`](c3084b3)) - fix(security): enforce durable community ban on NIP-43 relay-admin kinds 9030-9033 ([#3128](#3128)) ([`e2e0079101`](e2e0079)) - fix(security): authorize kind:9000 role changes in both directions ([#3017](#3017)) ([`00ecf2cac7`](00ecf2c)) - feat(desktop): handle project work from Inbox ([#3117](#3117)) ([`c5c4f390b6`](c5c4f39)) - feat(relay): make per-owner community limit configurable via BUZZ_MAX_COMMUNITIES_PER_OWNER ([#2599](#2599)) ([`2a051a404d`](2a051a4)) - feat(relay): add author-only-unless-shared read gate for kind 30175 ([#2768](#2768)) ([`ab3af82871`](ab3af82)) - fix(core): block IPv6 transition SSRF targets ([#2801](#2801)) ([`c26bf5945d`](c26bf59)) - fix(workflow): bypass system proxies for webhooks ([#2800](#2800)) ([`60a171b19e`](60a171b)) - fix(audit): hash created_at at the precision Postgres stores ([#2638](#2638)) ([`264a56a226`](264a56a)) - feat(desktop): make pull request reviews actionable ([#2510](#2510)) ([`9081ab0ec9`](9081ab0)) - fix(relay): decompress gzip-encoded git smart-HTTP request bodies ([#2670](#2670)) ([`5ca36e7b91`](5ca36e7)) - fix(sharing): preserve agent/team snapshot tEXt chunks through media sanitization ([#2438](#2438)) ([`b096b0a15a`](b096b0a)) - fix(relay): send 1012 restart close to all clients on graceful drain ([#2575](#2575)) ([`1911c69aa2`](1911c69)) - fix(media): sanitize animated image uploads ([#2524](#2524)) ([`8f8f5fa5a4`](8f8f5fa)) - fix(channels): strip leading hash prefixes from names ([#2250](#2250)) ([`d0ab3fdb05`](d0ab3fd)) - feat(relay): make Redis pool size configurable, default 16 ([#2521](#2521)) ([`bcc3e13069`](bcc3e13)) - feat(desktop+acp): spawn a harness per (agent, community) pair at GUI startup — warm sockets, lazy LLM pool ([#2122](#2122)) ([`61cc738ee8`](61cc738)) - feat(media): add S3-truth per-community storage sweep ([#2044](#2044)) ([`bd37a4d584`](bd37a4d)) - feat(relay): log NIP-98 pubkey attribution on HTTP bridge requests ([#2206](#2206)) ([`7e34bee62c`](7e34bee)) - Revert "feat(relay): inventory unreachable Git objects" ([#2275](#2275)) ([`0fb820f9bf`](0fb820f)) - feat(relay): inventory unreachable Git objects ([#2264](#2264)) ([`3afc9dae15`](3afc9da)) - relay: add author_type label to buzz_events_stored_total ([#2243](#2243)) ([`b9f54c43fe`](b9f54c4)) - fix(git): make project branch workflows reliable ([#2213](#2213)) ([`166f27be4b`](166f27b)) - feat(cli): manage repository protection rules ([#2193](#2193)) ([`f94324598d`](f943245)) - feat(cli): add agents archive/unarchive/archived subcommands ([#2173](#2173)) ([`7d7992067b`](7d79920)) - fix(mobile): sanitize Android image uploads ([#2188](#2188)) ([`ee21da90bd`](ee21da9)) - fix(cli): paginate channel directory queries ([#2181](#2181)) ([`03fe19d603`](03fe19d)) - fix(mobile): image upload fails due to unstripped metadata ([#2185](#2185)) ([`37f15b2001`](37f15b2)) - perf(relay): compact Git packs before manifest limits ([#2172](#2172)) ([`80e0ab16b0`](80e0ab1)) - perf(relay): cache Git pack hydration ([#2169](#2169)) ([`a4d82ec722`](a4d82ec)) - fix(relay): bound and observe Git read operations ([#2167](#2167)) ([`5f7c93d9c1`](5f7c93d)) - relay: gate push enqueue on live leases; batch matcher pipeline (T1b/T1a-repair/T2b) ([#2145](#2145)) ([`e43b2d5aac`](e43b2d5)) - relay: add audit logging disable switch ([#2134](#2134)) ([`bf5acabdde`](bf5acab)) - relay: skip TTL deadline bump for known-permanent channels (T1a write-amp) ([#2125](#2125)) ([`2e936d439c`](2e936d4)) - fix(git): carry NIP-OA delegation in auth event ([#2120](#2120)) ([`c12257d57a`](c12257d)) - Route lag-tolerant reads to an optional Postgres read replica ([#2084](#2084)) ([`29c48883d3`](29c4888)) - fix: recover community access visibility ([#2074](#2074)) ([`ca384d082d`](ca384d0)) - feat: proxy feedback-scoped admin attachments ([#2059](#2059)) ([`d7f918e3cb`](d7f918e)) - feat: add read-only deployment moderation dashboard ([#1999](#1999)) ([`68e670e001`](68e670e)) - Bug-bash round 2: table scroll, Goose instructions, workflow mention wake ([#2034](#2034)) ([`64b8fea6dc`](64b8fea)) - Strip media metadata on clients and reject it at the relay ([#2006](#2006)) ([`5cfd69cb0c`](5cfd69c)) - [codex] Hold Git concurrency permits through streaming (BUZZ-SEC-018) ([#1916](#1916)) ([`7baea42abb`](7baea42)) - [codex] Enforce shared relay admission limits (BUZZ-SEC-019) ([#1917](#1917)) ([`73fc0ec6cf`](73fc0ec)) - [codex] Block banned actors from moderation commands (BUZZ-SEC-007) ([#1915](#1915)) ([`caa195ca58`](caa195c)) - [codex] Fix relay WebSocket admission limits ([#1682](#1682)) ([`d3ce971fc7`](d3ce971)) - feat: add invite QR and mobile direct join ([#1957](#1957)) ([`648cbf3610`](648cbf3)) - fix(join-policy): require legal consent on hosted invites ([#1987](#1987)) ([`2e1577f76f`](2e1577f)) - [codex] Prevent actor-tag UI impersonation ([#1931](#1931)) ([`c540ec9678`](c540ec9)) - Scope relay runtime state by community ([#1658](#1658)) ([`d52dedb06f`](d52dedb)) - Apply optional relay join policy across join flows ([#1894](#1894)) ([`6c2d667575`](6c2d667)) - feat(media): require auth for relay media reads ([#1926](#1926)) ([`f308762852`](f308762)) - feat(relay): add community unarchive endpoint ([#1908](#1908)) ([`6b9641db2b`](6b9641d)) - feat(relay): gate Git web GUI separately ([#1901](#1901)) ([`34dc7dec75`](34dc7de)) - mesh: upgrade runtime, enforce membership, add shared compute provider ([#1656](#1656)) ([`54638ff4bb`](54638ff)) - Route Git scratch through configured volume ([#1884](#1884)) ([`2318b3096c`](2318b30)) - feat(relay): gate usage metrics behind stable leader ([#1814](#1814)) ([`59e9821503`](59e9821)) - Relay mesh: cross-pod tunnel + huddle transport (buzz-relay-mesh) ([#1670](#1670)) ([`ccb021d713`](ccb021d)) - feat(push): deliver accepted relay events as wakes ([#1866](#1866)) ([`bffbc5f22c`](bffbc5f)) - fix(db): resolve duplicate migration version ([#1863](#1863)) ([`08ad38a07f`](08ad38a)) - Add private product feedback sidecar ([#1857](#1857)) ([`af190c93e1`](af190c9)) - feat(relay): add durable community archival ([#1834](#1834)) ([`2b15a72675`](2b15a72)) - feat(push): add public APNs gateway ([#1770](#1770)) ([`1c006822e4`](1c00682)) - feat(relay): add atomic community ownership transfer ([#1845](#1845)) ([`52e42ccb9f`](52e42cc)) - Bound NIP-RS retention and search indexing ([#1771](#1771)) ([`1b4703021d`](1b47030)) - Add optional standalone pairing relay to Helm chart ([#1799](#1799)) ([`9b47c8548f`](9b47c85)) - fix(relay): publish membership snapshot on provisioning ([#1761](#1761)) ([`0950d392b7`](0950d39)) - feat(relay): per-community usage metrics ([#1723](#1723)) ([`620822899a`](6208228)) - refactor(desktop): remove vestigial MCP toolsets config ([#1776](#1776)) ([`dfec75b3c0`](dfec75b)) **To release:** merge this PR. The tag and build will happen automatically. Signed-off-by: Will Pfleger <pfleger.will@gmail.com>
## Buzz Relay release v0.2.1 ### Changes since relay-v0.2.0: - fix(sdk): preserve self-mention p tags in message and forum event builders ([block#4975](block#4975)) ([`78c87ae20e`](block@78c87ae)) - feat(desktop): adding rich link previews to messages ([block#3818](block#3818)) ([`1922d49cb2`](block@1922d49)) - feat(relay): accept kind:30179 private managed-agent events at ingest ([block#5133](block#5133)) ([`ad923353a2`](block@ad92335)) - fix(media): require authenticated reads ([block#4610](block#4610)) ([`769ac70b74`](block@769ac70)) - feat(identity): recover desktop identity from a signed-in phone ([block#4845](block#4845)) ([`6eb65919f1`](block@6eb6591)) - ci: prove the relay-driven mesh lifecycle — discover, join, infer, deny — with real nodes ([block#3862](block#3862)) ([`38bf642fcf`](block@38bf642)) - relay: fuzz WebSocket 1012 restart-close timing on graceful drain (BUZZ_DRAIN_JITTER_MS) ([block#4542](block#4542)) ([`e14fff74d0`](block@e14fff7)) - fix(reactions): support max-length custom emoji ([block#3833](block#3833)) ([`2ea9385015`](block@2ea9385)) - fix(channels): restrict private-channel invitations ([block#4612](block#4612)) ([`efe1893dd3`](block@efe1893)) - fix(workflow): bind trigger author to the signed event ([block#4607](block#4607)) ([`885bed35ee`](block@885bed3)) - fix(git): revoke access for banned relay members ([block#4608](block#4608)) ([`997b8caaa4`](block@997b8ca)) - Define private managed agent wire protocol ([block#4593](block#4593)) ([`067c085f37`](block@067c085)) - perf(relay): index channel-id lookups and skip trace-only reads ([block#4647](block#4647)) ([`bc9e6528a7`](block@bc9e652)) - Polish mobile inbox and media flows ([block#4512](block#4512)) ([`feccf4eabc`](block@feccf4e)) - fix(git): allow deleting the default branch ([block#4297](block#4297)) ([`fc598f5f8d`](block@fc598f5)) - feat(projects): add buzz projects CLI commands (NIP-MP kind:30621) ([block#4020](block#4020)) ([`b7bb15122e`](block@b7bb151)) - perf(relay): serve relay-membership checks from the read replica ([block#4124](block#4124)) ([`ac4fa13b8e`](block@ac4fa13)) - fix(relay): allow open relays to set their NIP-11 workspace icon (kind:9033) ([block#3998](block#3998)) ([`5765fc74b7`](block@5765fc7)) - feat(relay): accept kind:30621 multi-repo projects at ingest ([block#3171](block#3171)) ([`cb9701cd30`](block@cb9701c)) - feat(relay): raise hosted community limit to five ([block#3829](block#3829)) ([`10d5a26414`](block@10d5a26)) - fix(relay): align NIP-11 max_limit with REQ ceiling ([block#3635](block#3635)) ([`23f0c26b1c`](block@23f0c26)) - feat(relay): gate kind 30178 team-catalog reads behind the shared tag ([block#3358](block#3358)) ([`114d40d9d3`](block@114d40d)) - fix(db): isolate usage metrics advisory-lock test on scratch DB ([block#3670](block#3670)) ([`dba97eecd9`](block@dba97ee)) - perf(presence): reduce heartbeat frequency ([block#3783](block#3783)) ([`bf139e8d0b`](block@bf139e8)) - feat(mesh): upgrade embedded mesh to v0.74 and harden shared compute (split 1/2 of block#3467) ([block#3741](block#3741)) ([`4933672eb4`](block@4933672)) - feat(replica): portable heartbeat-token fence with snapshot-local reader routing ([block#3268](block#3268)) ([`63496cc1d4`](block@63496cc)) - fix(git): channel binding tooling + author remediation for unbound repos ([block#3626](block#3626)) ([`788b3c002b`](block@788b3c0)) - feat: configure S3 URL addressing style ([block#3400](block#3400)) ([`7012d86d52`](block@7012d86)) - feat(tracing): correlate trace IDs in relay logs ([block#3608](block#3608)) ([`005b5b819a`](block@005b5b8)) - fix(relay): avoid subscription lock inversion ([block#3413](block#3413)) ([`22be8bb351`](block@22be8bb)) - feat(cli): add users set-status command for NIP-38 profile status ([block#3253](block#3253)) ([`60158fce3e`](block@60158fc)) - feat(relay): make Postgres pool size configurable, default 50 ([block#3191](block#3191)) ([`2ce2d71cc3`](block@2ce2d71)) - feat(tracing): add datastore tracing plumbing ([block#2760](block#2760)) ([`e94b9aeda0`](block@e94b9ae)) - feat(invites): add use-limited invite links ([block#3141](block#3141)) ([`d500c2d5cf`](block@d500c2d)) - feat(admin): show reported message content in report detail ([block#3149](block#3149)) ([`f069a85503`](block@f069a85)) - resolve findings ([block#3150](block#3150)) ([`9b0f744804`](block@9b0f744)) - Revert "fix(cli,relay): resolve agents by verified owner" ([block#3168](block#3168)) ([`a041e2d21e`](block@a041e2d)) - fix(cli,relay): resolve agents by verified owner ([block#2615](block#2615)) ([`c3084b36d9`](block@c3084b3)) - fix(security): enforce durable community ban on NIP-43 relay-admin kinds 9030-9033 ([block#3128](block#3128)) ([`e2e0079101`](block@e2e0079)) - fix(security): authorize kind:9000 role changes in both directions ([block#3017](block#3017)) ([`00ecf2cac7`](block@00ecf2c)) - feat(desktop): handle project work from Inbox ([block#3117](block#3117)) ([`c5c4f390b6`](block@c5c4f39)) - feat(relay): make per-owner community limit configurable via BUZZ_MAX_COMMUNITIES_PER_OWNER ([block#2599](block#2599)) ([`2a051a404d`](block@2a051a4)) - feat(relay): add author-only-unless-shared read gate for kind 30175 ([block#2768](block#2768)) ([`ab3af82871`](block@ab3af82)) - fix(core): block IPv6 transition SSRF targets ([block#2801](block#2801)) ([`c26bf5945d`](block@c26bf59)) - fix(workflow): bypass system proxies for webhooks ([block#2800](block#2800)) ([`60a171b19e`](block@60a171b)) - fix(audit): hash created_at at the precision Postgres stores ([block#2638](block#2638)) ([`264a56a226`](block@264a56a)) - feat(desktop): make pull request reviews actionable ([block#2510](block#2510)) ([`9081ab0ec9`](block@9081ab0)) - fix(relay): decompress gzip-encoded git smart-HTTP request bodies ([block#2670](block#2670)) ([`5ca36e7b91`](block@5ca36e7)) - fix(sharing): preserve agent/team snapshot tEXt chunks through media sanitization ([block#2438](block#2438)) ([`b096b0a15a`](block@b096b0a)) - fix(relay): send 1012 restart close to all clients on graceful drain ([block#2575](block#2575)) ([`1911c69aa2`](block@1911c69)) - fix(media): sanitize animated image uploads ([block#2524](block#2524)) ([`8f8f5fa5a4`](block@8f8f5fa)) - fix(channels): strip leading hash prefixes from names ([block#2250](block#2250)) ([`d0ab3fdb05`](block@d0ab3fd)) - feat(relay): make Redis pool size configurable, default 16 ([block#2521](block#2521)) ([`bcc3e13069`](block@bcc3e13)) - feat(desktop+acp): spawn a harness per (agent, community) pair at GUI startup — warm sockets, lazy LLM pool ([block#2122](block#2122)) ([`61cc738ee8`](block@61cc738)) - feat(media): add S3-truth per-community storage sweep ([block#2044](block#2044)) ([`bd37a4d584`](block@bd37a4d)) - feat(relay): log NIP-98 pubkey attribution on HTTP bridge requests ([block#2206](block#2206)) ([`7e34bee62c`](block@7e34bee)) - Revert "feat(relay): inventory unreachable Git objects" ([block#2275](block#2275)) ([`0fb820f9bf`](block@0fb820f)) - feat(relay): inventory unreachable Git objects ([block#2264](block#2264)) ([`3afc9dae15`](block@3afc9da)) - relay: add author_type label to buzz_events_stored_total ([block#2243](block#2243)) ([`b9f54c43fe`](block@b9f54c4)) - fix(git): make project branch workflows reliable ([block#2213](block#2213)) ([`166f27be4b`](block@166f27b)) - feat(cli): manage repository protection rules ([block#2193](block#2193)) ([`f94324598d`](block@f943245)) - feat(cli): add agents archive/unarchive/archived subcommands ([block#2173](block#2173)) ([`7d7992067b`](block@7d79920)) - fix(mobile): sanitize Android image uploads ([block#2188](block#2188)) ([`ee21da90bd`](block@ee21da9)) - fix(cli): paginate channel directory queries ([block#2181](block#2181)) ([`03fe19d603`](block@03fe19d)) - fix(mobile): image upload fails due to unstripped metadata ([block#2185](block#2185)) ([`37f15b2001`](block@37f15b2)) - perf(relay): compact Git packs before manifest limits ([block#2172](block#2172)) ([`80e0ab16b0`](block@80e0ab1)) - perf(relay): cache Git pack hydration ([block#2169](block#2169)) ([`a4d82ec722`](block@a4d82ec)) - fix(relay): bound and observe Git read operations ([block#2167](block#2167)) ([`5f7c93d9c1`](block@5f7c93d)) - relay: gate push enqueue on live leases; batch matcher pipeline (T1b/T1a-repair/T2b) ([block#2145](block#2145)) ([`e43b2d5aac`](block@e43b2d5)) - relay: add audit logging disable switch ([block#2134](block#2134)) ([`bf5acabdde`](block@bf5acab)) - relay: skip TTL deadline bump for known-permanent channels (T1a write-amp) ([block#2125](block#2125)) ([`2e936d439c`](block@2e936d4)) - fix(git): carry NIP-OA delegation in auth event ([block#2120](block#2120)) ([`c12257d57a`](block@c12257d)) - Route lag-tolerant reads to an optional Postgres read replica ([block#2084](block#2084)) ([`29c48883d3`](block@29c4888)) - fix: recover community access visibility ([block#2074](block#2074)) ([`ca384d082d`](block@ca384d0)) - feat: proxy feedback-scoped admin attachments ([block#2059](block#2059)) ([`d7f918e3cb`](block@d7f918e)) - feat: add read-only deployment moderation dashboard ([block#1999](block#1999)) ([`68e670e001`](block@68e670e)) - Bug-bash round 2: table scroll, Goose instructions, workflow mention wake ([block#2034](block#2034)) ([`64b8fea6dc`](block@64b8fea)) - Strip media metadata on clients and reject it at the relay ([block#2006](block#2006)) ([`5cfd69cb0c`](block@5cfd69c)) - [codex] Hold Git concurrency permits through streaming (BUZZ-SEC-018) ([block#1916](block#1916)) ([`7baea42abb`](block@7baea42)) - [codex] Enforce shared relay admission limits (BUZZ-SEC-019) ([block#1917](block#1917)) ([`73fc0ec6cf`](block@73fc0ec)) - [codex] Block banned actors from moderation commands (BUZZ-SEC-007) ([block#1915](block#1915)) ([`caa195ca58`](block@caa195c)) - [codex] Fix relay WebSocket admission limits ([block#1682](block#1682)) ([`d3ce971fc7`](block@d3ce971)) - feat: add invite QR and mobile direct join ([block#1957](block#1957)) ([`648cbf3610`](block@648cbf3)) - fix(join-policy): require legal consent on hosted invites ([block#1987](block#1987)) ([`2e1577f76f`](block@2e1577f)) - [codex] Prevent actor-tag UI impersonation ([block#1931](block#1931)) ([`c540ec9678`](block@c540ec9)) - Scope relay runtime state by community ([block#1658](block#1658)) ([`d52dedb06f`](block@d52dedb)) - Apply optional relay join policy across join flows ([block#1894](block#1894)) ([`6c2d667575`](block@6c2d667)) - feat(media): require auth for relay media reads ([block#1926](block#1926)) ([`f308762852`](block@f308762)) - feat(relay): add community unarchive endpoint ([block#1908](block#1908)) ([`6b9641db2b`](block@6b9641d)) - feat(relay): gate Git web GUI separately ([block#1901](block#1901)) ([`34dc7dec75`](block@34dc7de)) - mesh: upgrade runtime, enforce membership, add shared compute provider ([block#1656](block#1656)) ([`54638ff4bb`](block@54638ff)) - Route Git scratch through configured volume ([block#1884](block#1884)) ([`2318b3096c`](block@2318b30)) - feat(relay): gate usage metrics behind stable leader ([block#1814](block#1814)) ([`59e9821503`](block@59e9821)) - Relay mesh: cross-pod tunnel + huddle transport (buzz-relay-mesh) ([block#1670](block#1670)) ([`ccb021d713`](block@ccb021d)) - feat(push): deliver accepted relay events as wakes ([block#1866](block#1866)) ([`bffbc5f22c`](block@bffbc5f)) - fix(db): resolve duplicate migration version ([block#1863](block#1863)) ([`08ad38a07f`](block@08ad38a)) - Add private product feedback sidecar ([block#1857](block#1857)) ([`af190c93e1`](block@af190c9)) - feat(relay): add durable community archival ([block#1834](block#1834)) ([`2b15a72675`](block@2b15a72)) - feat(push): add public APNs gateway ([block#1770](block#1770)) ([`1c006822e4`](block@1c00682)) - feat(relay): add atomic community ownership transfer ([block#1845](block#1845)) ([`52e42ccb9f`](block@52e42cc)) - Bound NIP-RS retention and search indexing ([block#1771](block#1771)) ([`1b4703021d`](block@1b47030)) - Add optional standalone pairing relay to Helm chart ([block#1799](block#1799)) ([`9b47c8548f`](block@9b47c85)) - fix(relay): publish membership snapshot on provisioning ([block#1761](block#1761)) ([`0950d392b7`](block@0950d39)) - feat(relay): per-community usage metrics ([block#1723](block#1723)) ([`620822899a`](block@6208228)) - refactor(desktop): remove vestigial MCP toolsets config ([block#1776](block#1776)) ([`dfec75b3c0`](block@dfec75b)) **To release:** merge this PR. The tag and build will happen automatically. Signed-off-by: Will Pfleger <pfleger.will@gmail.com>
Summary
CI now proves the full Buzz shared-compute join story end to end: a member can discover another member's served model through the Buzz relay alone and run inference over the mesh, while a non-member gets nothing — the relay rejects its auth, and the mesh refuses to route for it even holding a leaked endpoint address.
This is deliberately different from mesh-llm's own CI smokes (which bootstrap two nodes with a hand-carried invite token / mdns): here the relay is the control plane, exactly like the desktop app:
buzz-admin(kind:13534 NIP-43 roster); C is not.serveTargets[].endpointAddr, covered by an endpoint-binding signature — the exact payload shape the desktop coordinator publishes.TrustPolicy::Allowlist.dial_endpoint_addrstep). No out-of-band token.restricted: not a relay member— successful auth or any unrelated connect error fails the run), and dialing the leaked endpoint must not produce a routed inference — while the trusted client re-proves inference immediately afterwards, so a dead serve node can't masquerade as an admission denial.What's in the PR
crates/buzz-relay/examples/mesh_relay_lifecycle_smoke.rs— the harness. One process per node (mesh-llm keeps process-global state under~/.mesh-llm), orchestrator + serve/client/stranger roles, byte-identical binding payloads todesktop/src-tauri/src/mesh_llm/identity.rs(called out with keep-in-sync comments). Child stdout is pumped through a reader thread so every wait has a hard deadline; timed-out children are killed; exit statuses are checked.scripts/ci-mesh-lifecycle-smoke.sh— provisions a membership-gated relay (throwaway owner + signing identities viabuzz-admin generate-key), runs the harness, cleans up. Fails fast if :3000 is already occupied (a stale open relay would mask gating).scripts/start-relay-for-tests.sh— gains opt-in NIP-43 membership env passthrough (BUZZ_REQUIRE_RELAY_MEMBERSHIP+RELAY_OWNER_PUBKEY+BUZZ_RELAY_PRIVATE_KEY). Default behavior unchanged..github/workflows/mesh-lifecycle.yml— separate, path-filtered, non-required workflow (mesh paths, the harness's dependency crates,Cargo.lock, dispatch), pinned toubuntu-24.04. Caches the mesh native runtime + HF model keyed on the lockfile hash, so a mesh pin bump rolls the runtime cache. Uploads relay + harness logs on failure.Scope
This is an independent protocol harness: it speaks the same wire protocol and payload shapes as the desktop but re-implements the binding/verification logic (the desktop crate is outside the workspace). Regressions inside the desktop's own discovery filtering are the desktop unit tests' job; what this smoke proves is that the relay + mesh-llm SDK + admission stack support the lifecycle end to end.
Relationship to mesh-llm's CI
Follows the shape mesh-llm's own CI proved stable (tiny CPU model, one runner, multiple real mesh-llm processes over real QUIC — cf. their
ci-two-node-client-serving-smoke.sh), but swaps the token bootstrap for the relay-driven lifecycle, which is the part only Buzz can test.Validation
Green on GitHub Actions (ubuntu-24.04) across three runs, including after rebases onto the mesh v0.74 upgrade (#3467) and latest main:
Also validated locally on macOS.
cargo fmt --all --checkandcargo clippy -p buzz-relay --all-targets -- -D warningspass.Notes
[dev-dependencies]pin automatically, so it doubles as a canary for future mesh upgrades (it already caught the v0.73.1 → v0.74.0 bump during development).MESH_CLIENT_WINDOW_SECS/MESH_STRANGER_WINDOW_SECSif shared runners prove slow — pin a longer window in the workflow env rather than re-running the job.