feat(#3255): hives stop owning the knowledge webhook, and clean up their own

A webhook has exactly one target URL, so every hive registering one
against the shared internal/knowledge repository was last-writer-wins
rather than idempotent: all but the most recent silently stopped
receiving deliveries. The swarm controller holds the single registration
and now addresses an event to each hive over the queue instead.

This is a migration, not a deletion. Not registering any more fixes
nothing on a hive that has already run — the hook it created persists on
the forge, so the contention would survive on exactly the deployments
that have it while fresh installs looked fixed. The hive that created a
hook removes it.

It removes only its OWN, matched on the full URL rather than the
/webhook/knowledge suffix. A hook with that suffix and a different base
belongs to another hive, possibly one not yet upgraded, and deleting it
would break that hive's knowledge sync until it caught up. Reaping a
neighbour's registration is the behaviour being removed here; doing it
while fixing it would only invert the direction.

The predecessor did reap by suffix, to clear loopback hooks left by an
older single-hive layout. That was safe when a hive was alone on its
forge and is not safe now. The hive-side registrars also acted as reapers
of hooks under their own path, which is why the swarm hook lives under
/webhook/forge/; removing this registrar removes that reaper too.
Intended, and stated because no reviewer would infer it from the diff.

The receive endpoint goes with it. A live HMAC-verified
/webhook/knowledge that nothing can legitimately reach would tell the
next reader that this is how a hive learns about knowledge changes.

Docs move in the same commit: docs/swarm/README.md said two hooks exist
per swarm-wide repo and neither should be deleted, which is now true for
agent-configs and wrong for internal/knowledge — a half-correct
description being worse than an uncorrected one.
This commit is contained in:
atlas 2026-08-19 19:25:30 +02:00 committed by mara
commit d2a550e685
6 changed files with 111 additions and 218 deletions

View file

@ -38,18 +38,23 @@ create a new document.
hive-c0re maintains the local clone at
`/var/lib/hyperhive/knowledge` via two paths:
1. **Forgejo push webhook**`ensure_webhook` registers a push
hook on `internal/knowledge` at startup pointing at
`https://<hive_domain>/webhook/knowledge` (routed through the
gateway, avoiding the Forgejo SSRF guard that blocks loopback
delivery). On any push to main (including merge commits) hive-c0re
runs `git pull` so agents see the new content on their next turn.
The endpoint is protected by an auto-generated HMAC secret that
hive-c0re verifies on every delivery.
1. **Swarm event** — the swarm controller holds the single push hook on
`internal/knowledge` (see `docs/swarm/README.md` § Swarm-wide forge
webhooks). On any push to main, including merge commits, it sends an
event to every hive over the swarm queue and each hive runs `git
pull`, so agents see the new content on their next turn.
On a swarm, a **second** hook on the same repo points at the
swarm controller — see `docs/swarm/README.md` § Swarm-wide forge
webhooks. Both are expected; neither should be deleted.
A hive that is offline when the event is sent does not get it on
reconnect — the periodic pull below is what closes that gap. So one
hive briefly showing older `/knowledge` content than another is
expected, and resolves by itself within the fallback interval.
**Do not add a per-hive hook.** A webhook has exactly one target
URL, so a second registration against the same repo does not add a
recipient — it takes delivery away from whoever registered first.
Earlier versions had each hive register its own; hive-c0re now
removes its own leftover at startup, so no operator step is needed
to migrate.
2. **Periodic pull** — a background task in `hive-c0re::main`
pulls on a fixed cadence as a fallback (webhook missed, c0re

View file

@ -356,18 +356,23 @@ itself — a `push` hook on `internal/knowledge` and a `pull_request` hook
on the `agent-configs` org, both under
`https://<swarm.ui.domain>/webhook/forge/`.
**Two hooks exist per swarm-wide repo: each hive's own, plus the
controller's.** Both are expected — **do not delete either.** Removing a
hive's hook stops that hive acting on knowledge pushes and config PRs;
removing the controller's just gets recreated on its next start.
The controller **interprets** a delivery and sends hives a specific
message — *the knowledge repo changed*, *deploy agent `foo` at rev
`abc123`* — rather than forwarding forge payloads for each hive to
re-derive. Approval happens once, at the swarm level: a hive receives a
decision, not an event to adjudicate. Today the controller logs each
verified delivery and sends nothing, because the swarm→hive channel does
not exist yet; the hive-side hooks are what act in the meantime.
decision, not an event to adjudicate.
**`internal/knowledge` is on that path.** The controller's is the only
hook on it: hives no longer register their own, and each removes its
leftover at startup. A webhook has exactly one target URL, so per-hive
registration never added a recipient — it took delivery away from
whichever hive registered before it.
**The `agent-configs` org is not yet.** Each hive still registers its own
`pull_request` hook there, so that repo has two — the hive's and the
controller's — and **both are expected; do not delete either.** Removing
a hive's stops it acting on config PRs; removing the controller's just
gets recreated on its next start.
Nothing to configure. The hooks are registered only when this host also
serves the swarm UI vhost — that is what publishes the endpoint, and a

View file

@ -184,7 +184,6 @@ pub async fn serve(
.routes(routes!(schedules::post_schedule_resume))
.routes(routes!(schedules::post_schedule_fire_now))
.routes(routes!(schedules::post_rebuild_queue_cancel))
.routes(routes!(webhook::post_webhook_knowledge))
.routes(routes!(webhook::post_webhook_config_pr))
.routes(routes!(approvals::post_approve))
.routes(routes!(approvals::post_deny))
@ -430,7 +429,6 @@ mod router_build_probe {
.routes(routes!(schedules::post_schedule_resume))
.routes(routes!(schedules::post_schedule_fire_now))
.routes(routes!(schedules::post_rebuild_queue_cancel))
.routes(routes!(webhook::post_webhook_knowledge))
.routes(routes!(webhook::post_webhook_config_pr))
.routes(routes!(approvals::post_approve))
.routes(routes!(approvals::post_deny))

View file

@ -1,12 +1,18 @@
//! Forgejo webhook endpoints.
//!
//! - **`/webhook/knowledge`** — push events on `internal/knowledge` trigger a
//! `git pull` on the local clone so agents see up-to-date docs.
//! - **`/webhook/config-pr`** — `pull_request` events on any `agent-configs/*`
//! repo queue a [`hive_sh4re::approvals::ApprovalKind::MergeConfigPr`] approval row
//! so the operator can review + approve the merge from the dashboard.
//!
//! Both endpoints are reached via the gateway (HTTPS, public domain URL) so
//! There was a second endpoint here, `/webhook/knowledge`, which pulled the
//! local `internal/knowledge` clone on push. It is gone along with the
//! per-hive registration that fed it: a webhook has exactly one target URL,
//! so every hive registering one against the shared repository was
//! last-writer-wins. The swarm controller now holds the single registration
//! and addresses an event to each hive over the queue, which
//! [`crate::workers::knowledge`] documents.
//!
//! The endpoint is reached via the gateway (HTTPS, public domain URL) so
//! Forgejo's SSRF guard does not block delivery. Each delivery is verified
//! against the `X-Hub-Signature-256` HMAC header Forgejo attaches; the
//! shared secret is auto-generated at startup and persisted to
@ -42,104 +48,6 @@ fn verify_hmac(state: &AppState, headers: &HeaderMap, body: &Bytes) -> Result<()
crate::webhook_secret::verify_signature(secret, body, sig).map_err(|e| e.to_string())
}
// ── knowledge webhook ──────────────────────────────────────────────────────────
/// Minimal Forgejo push-webhook payload — only the fields we care about.
#[derive(Deserialize)]
pub(super) struct PushWebhookPayload {
#[serde(rename = "ref")]
git_ref: Option<String>,
repository: Option<PushWebhookRepo>,
}
#[derive(Deserialize)]
pub(super) struct PushWebhookRepo {
full_name: Option<String>,
}
/// POST `/webhook/knowledge` — Forgejo push webhook for
/// `internal/knowledge`.
///
/// Runs `git pull` on the local clone so agents see up-to-date documents
/// on their next turn.
///
/// Expected Forgejo webhook configuration:
/// - URL: `https://<HYPERHIVE_HIVE_DOMAIN>/webhook/knowledge`
/// - Content type: `application/json`
/// - Event: "Push" (fires on merge commits to main as well)
/// - Secret: auto-generated HMAC key (see [`crate::webhook_secret`])
///
/// The gateway routes `/webhook/` → hive-c0re; the HMAC secret protects
/// the endpoint from unauthenticated callers.
#[utoipa::path(
post,
path = "/webhook/knowledge",
request_body(
content = String,
content_type = "application/json",
description = "Forgejo push-webhook payload, taken as raw bytes \
(not a typed extractor) so HMAC verification runs \
over the exact wire bytes before any JSON parsing"
),
responses(
(status = 200, description = "processed (pull triggered or ignored)", body = String),
(status = 400, description = "invalid JSON payload"),
(status = 401, description = "bad or missing HMAC signature"),
(status = 503, description = "HMAC secret unavailable at startup"),
),
tag = "webhook"
)]
pub(super) async fn post_webhook_knowledge(
State(state): State<AppState>,
headers: HeaderMap,
body: Bytes,
) -> Response {
if let Err(e) = verify_hmac(&state, &headers, &body) {
tracing::warn!("webhook/knowledge: HMAC verification failed: {e}");
let status = if e.contains("unavailable") {
StatusCode::SERVICE_UNAVAILABLE
} else {
StatusCode::UNAUTHORIZED
};
return (status, e).into_response();
}
let payload = match serde_json::from_slice::<PushWebhookPayload>(&body) {
Ok(p) => p,
Err(e) => {
tracing::warn!("webhook/knowledge: JSON parse error: {e}");
return (StatusCode::BAD_REQUEST, "invalid JSON").into_response();
}
};
let expected_repo = format!("{}/{}", crate::knowledge::ORG, crate::knowledge::REPO);
let full_name = payload
.repository
.as_ref()
.and_then(|r| r.full_name.as_deref())
.unwrap_or("");
if full_name != expected_repo {
tracing::debug!(
full_name,
"webhook/knowledge: ignoring push from unexpected repo"
);
return (StatusCode::OK, "ignored").into_response();
}
let git_ref = payload.git_ref.as_deref().unwrap_or("");
if git_ref != "refs/heads/main" {
tracing::debug!(git_ref, "webhook/knowledge: ignoring non-main push");
return (StatusCode::OK, "ignored").into_response();
}
tracing::info!("webhook/knowledge: pull triggered by push to {expected_repo}");
let coord = state.coord.clone();
tokio::spawn(async move {
if let Err(e) = crate::knowledge::pull(&coord).await {
tracing::warn!(error = ?e, "webhook/knowledge: pull failed");
}
});
(StatusCode::OK, "ok").into_response()
}
// ── config-PR webhook ──────────────────────────────────────────────────────────
/// Minimal Forgejo `pull_request`-webhook payload.

View file

@ -179,10 +179,18 @@ async fn run_matrix_sweep() -> Result<()> {
}
}
/// Boot-time Forgejo webhook registration as a DAG node — see
/// Boot-time Forgejo webhook management as a DAG node — see
/// [`NodeKind::WebhookRegister`]. Mirrors the guard chain the
/// `tokio::spawn` block it replaced used: no-op (not an error) when the
/// HMAC secret, core token, or hive domain aren't available yet.
///
/// The node now does one of each: it still registers the config-PR hook,
/// and it *removes* the knowledge one. A knowledge push is delivered to
/// the swarm controller, which addresses an event to each hive over the
/// queue — so a hive holding its own registration is holding a shared
/// resource only one party can own. The removal runs every boot rather
/// than behind a marker because it is already idempotent: it is a no-op
/// the moment the hook is gone.
async fn run_webhook_register() -> Result<()> {
let Ok(webhook_secret) = crate::webhook_secret::load_or_generate() else {
tracing::debug!("webhook secret unavailable; skipping hook registration");
@ -198,10 +206,8 @@ async fn run_webhook_register() -> Result<()> {
tracing::debug!("HYPERHIVE_HIVE_DOMAIN unset; skipping webhook registration");
return Ok(());
};
if let Err(e) =
crate::workers::knowledge::ensure_webhook(&token, &domain, &webhook_secret).await
{
tracing::warn!(error = ?e, "knowledge: ensure_webhook failed");
if let Err(e) = crate::workers::knowledge::remove_webhook(&token, &domain).await {
tracing::warn!(error = ?e, "knowledge: remove_webhook failed");
}
if let Err(e) = crate::forge::ensure_config_pr_webhook(&token, &domain, &webhook_secret).await {
tracing::warn!(error = ?e, "forge: ensure_config_pr_webhook failed");

View file

@ -6,15 +6,18 @@
//! contribute by forking the repo and opening PRs — they never write
//! to the bind-mounted path inside the container.
//!
//! hive-c0re maintains the local clone. A Forgejo webhook notifies it
//! on push to main so agents always see an up-to-date snapshot. The
//! webhook is auto-created by [`ensure_webhook`] at startup. A
//! hive-c0re maintains the local clone. It learns that the repository
//! moved from the **swarm controller**, which owns the one Forgejo
//! webhook and addresses an event to each hive over the queue; a
//! periodic pull in `main.rs` provides a fallback cadence.
use std::collections::BTreeMap;
//!
//! A hive used to register that webhook itself, pointing at its own
//! `/webhook/knowledge`. A webhook has exactly one target URL, so with
//! more than one hive that was last-writer-wins rather than idempotent —
//! every hive but the most recent silently stopped receiving deliveries.
//! [`remove_webhook`] is the migration off it.
use anyhow::{Context, Result};
use forgejo_api::structs::{CreateHookOption, CreateHookOptionConfig, CreateHookOptionType};
use crate::coordinator::Coordinator;
use crate::forge::{core_auth_header, forge_git_url};
@ -142,103 +145,71 @@ async fn seed_readme(core_token: &str) -> Result<()> {
}
}
/// Ensure a Forgejo push webhook for `internal/knowledge` exists and
/// points at hive-c0re's `/webhook/knowledge` endpoint. Idempotent —
/// lists existing hooks first and skips creation when one is already
/// targeting the correct URL.
/// Delete this hive's own `internal/knowledge` push webhook if it is
/// still registered, so the swarm controller is the only party holding
/// one.
///
/// `hive_domain` is the public domain name of the hive; the webhook URL is
/// `https://<hive_domain>/webhook/knowledge` (routed through the gateway,
/// avoiding the Forgejo SSRF guard that blocks loopback delivery).
/// # Why this is a migration and not just a deletion
///
/// `webhook_secret` is the HMAC secret Forgejo will attach as
/// `X-Hub-Signature-256` on each delivery; hive-c0re verifies this header
/// in the dashboard webhook handler (`post_webhook_knowledge`).
/// Not registering any more fixes nothing on a hive that has already
/// run: the hook it created persists on the forge, so the contention
/// this removes would survive on exactly the deployments that have it
/// while fresh installs looked fixed. The hive that created a hook is
/// the one that removes it.
///
/// Called at startup alongside [`ensure_local_clone`]. No-op when the
/// core token is absent (forge not yet provisioned).
pub async fn ensure_webhook(
core_token: &str,
hive_domain: &str,
webhook_secret: &str,
) -> Result<()> {
/// # It removes only its OWN hook, never a neighbour's
///
/// The match is the full URL, not the `/webhook/knowledge` suffix. A
/// hook with that suffix and a different base belongs to *another hive* —
/// one that may not have been upgraded yet — and deleting it would break
/// its knowledge sync until it was. Reaping a neighbour's registration is
/// the very behaviour this issue is about; doing it in the name of fixing
/// it would just invert the direction.
///
/// (The predecessor did reap by suffix, to clear loopback hooks left by
/// an older single-hive layout. That was safe when a hive was alone on
/// its forge and is not safe now.)
///
/// A listing failure is an error rather than a silent skip: there is no
/// create attempt left to fall through to, so swallowing it would leave
/// the hook in place with nothing said. The caller logs and continues —
/// boot does not depend on this.
pub async fn remove_webhook(core_token: &str, hive_domain: &str) -> Result<()> {
// The typed client carries no per-request timeout, so each call is
// wrapped in one: this runs as a detached startup task, and a forge
// that accepts connections but never answers would otherwise hang
// it forever (and the hourly pull fallback masks the missing hook).
// that accepts connections but never answers would otherwise hang it
// forever.
const HTTP_TIMEOUT: std::time::Duration = std::time::Duration::from_secs(10);
let target_url = format!("https://{hive_domain}/webhook/knowledge");
let own_url = format!("https://{hive_domain}/webhook/knowledge");
let client = crate::forge::api(core_token)?;
// List existing hooks — skip creation if ours is already there.
// Best-effort like the raw-HTTP predecessor: a listing failure
// falls through to the create attempt.
let listed = tokio::time::timeout(HTTP_TIMEOUT, client.repo_list_hooks(ORG, REPO).all())
.await
.map_err(anyhow::Error::from)
.and_then(|r| r.map_err(anyhow::Error::from));
match listed {
Ok(hooks) => {
let already_exists = hooks.iter().any(|h| {
h.config
.as_ref()
.and_then(|c| c.get("url"))
.map(String::as_str)
== Some(target_url.as_str())
});
if already_exists {
tracing::debug!(%target_url, "knowledge: push webhook already configured");
return Ok(());
}
// Delete stale hooks that point at our path but a different base
// (e.g. old loopback hooks from before the SSRF-bypass migration).
for h in &hooks {
let hook_url = h
.config
.as_ref()
.and_then(|c| c.get("url"))
.map_or("", String::as_str);
if hook_url.ends_with("/webhook/knowledge")
&& hook_url != target_url
&& let Some(id) = h.id
{
tracing::info!(hook_url, "knowledge: deleting stale webhook (wrong base)");
let _ = tokio::time::timeout(
HTTP_TIMEOUT,
client.repo_delete_hook(ORG, REPO, id).send(),
)
.await;
}
}
}
Err(e) => {
tracing::debug!(error = %e, "knowledge: listing hooks failed; attempting create");
}
}
// Create the webhook.
let mut additional = BTreeMap::new();
additional.insert("secret".to_owned(), webhook_secret.to_owned());
let hook = CreateHookOption {
active: Some(true),
authorization_header: None,
branch_filter: None,
config: CreateHookOptionConfig {
content_type: "json".to_owned(),
url: url::Url::parse(&target_url).context("parse webhook target url")?,
additional,
},
events: Some(vec!["push".to_owned()]),
r#type: CreateHookOptionType::Forgejo,
};
tokio::time::timeout(HTTP_TIMEOUT, client.repo_create_hook(ORG, REPO, hook))
let hooks = tokio::time::timeout(HTTP_TIMEOUT, client.repo_list_hooks(ORG, REPO).all())
.await
.map_err(anyhow::Error::from)
.and_then(|r| r.map_err(anyhow::Error::from))
.with_context(|| format!("create webhook for {ORG}/{REPO}"))?;
tracing::info!(%target_url, "knowledge: push webhook created");
.with_context(|| format!("list webhooks for {ORG}/{REPO}"))?;
for h in &hooks {
let hook_url = h
.config
.as_ref()
.and_then(|c| c.get("url"))
.map_or("", String::as_str);
if hook_url == own_url
&& let Some(id) = h.id
{
tokio::time::timeout(HTTP_TIMEOUT, client.repo_delete_hook(ORG, REPO, id).send())
.await
.map_err(anyhow::Error::from)
.and_then(|r| r.map_err(anyhow::Error::from))
.with_context(|| format!("delete webhook {id} for {ORG}/{REPO}"))?;
tracing::info!(
%own_url,
"knowledge: removed this hive's push webhook — the swarm controller owns it now"
);
}
}
Ok(())
}