hyperhive/hive-c0re/src/forge/config_pr_poll.rs

//! Polling fallback for the config-PR webhook.
//!
//! The webhook (`/webhook/config-pr`) is the primary path for detecting open
//! PRs on `agent-configs/*` repos and queuing `MergeConfigPr` approvals.
//! But webhooks can be missed — hive-c0re might be down when a PR is opened,
//! or Forgejo might fail a delivery.
//!
//! This module provides [`poll_open_config_prs`], called periodically from
//! `main.rs`, which scans all `agent-configs/*` repos for open PRs that have
//! no pending `MergeConfigPr` approval yet, and queues one. Idempotent: PRs
//! that already have a pending approval are skipped.

use std::sync::Arc;

use anyhow::Result;
use forgejo_api::structs::{RepoListPullRequestsQuery, RepoListPullRequestsQueryState};

use crate::coordinator::Coordinator;
use crate::forge::CONFIG_ORG;

const HTTP_TIMEOUT: std::time::Duration = std::time::Duration::from_secs(15);

/// Scan every repo in `agent-configs` for open PRs that have no pending
/// `MergeConfigPr` approval yet, and queue one for each gap found.
///
/// Designed to be called on a periodic timer (e.g. every 5 minutes) as a
/// fault-tolerance backstop for the Forgejo webhook. The webhook fires
/// immediately; this catches anything the webhook missed.
pub async fn poll_open_config_prs(core_token: &str, coord: &Arc<Coordinator>) -> Result<()> {
    let client = crate::forge::api(core_token)?;

    // List all repos in agent-configs org.
    let repos = tokio::time::timeout(HTTP_TIMEOUT, client.org_list_repos(CONFIG_ORG).all())
        .await
        .map_err(anyhow::Error::from)
        .and_then(|r| r.map_err(anyhow::Error::from))?;

    // (agent, pr_number) of every OPEN config PR seen this sweep, plus the set
    // of agents whose PR list was fetched successfully. The reconcile pass
    // below uses these to cancel pending approvals whose PR is no longer open,
    // without wrongly cancelling one when a repo's list failed transiently.
    let mut open_prs: std::collections::HashSet<(String, u64)> = std::collections::HashSet::new();
    let mut scanned_agents: std::collections::HashSet<String> = std::collections::HashSet::new();

    for repo in repos {
        let Some(repo_name) = repo.name.as_deref() else {
            continue;
        };
        // The repo name is the agent name (agent-configs/<agent>).
        let agent = repo_name;

        let query = RepoListPullRequestsQuery {
            state: Some(RepoListPullRequestsQueryState::Open),
            sort: None,
            milestone: None,
            labels: None,
            poster: None,
            base: None,
            head: None,
        };

        let prs = match tokio::time::timeout(
            HTTP_TIMEOUT,
            client
                .repo_list_pull_requests(CONFIG_ORG, repo_name, query)
                .all(),
        )
        .await
        {
            Ok(Ok(prs)) => {
                scanned_agents.insert(agent.to_owned());
                prs
            }
            Ok(Err(e)) => {
                tracing::debug!(
                    %agent, error = %e,
                    "config-pr poll: listing PRs failed, skipping repo"
                );
                continue;
            }
            Err(_) => {
                tracing::debug!(
                    %agent,
                    "config-pr poll: timeout listing PRs, skipping repo"
                );
                continue;
            }
        };

        for pr in prs {
            let Some(pr_number) = pr.number.and_then(|n| u64::try_from(n).ok()) else {
                continue;
            };
            open_prs.insert((agent.to_owned(), pr_number));

            // `submit_merge_config_pr` is idempotent + PR-drift aware: it
            // no-ops when an approval pinned to this PR's *current* head is
            // already pending, and cancels+re-queues when the head has drifted.
            // So the poll can call it unconditionally — it backstops both a
            // missed `opened` webhook (no approval yet) and a missed
            // `synchronize` (stale approval whose head moved).
            let description = format!("PR #{pr_number} on {CONFIG_ORG}/{agent} (poll fallback)");
            if let Err(e) = crate::socket_server::submit_merge_config_pr(
                coord,
                agent,
                pr_number,
                Some(&description),
                "poll", // submitter — identifies the polling path in the audit trail
            )
            .await
            {
                tracing::warn!(
                    %agent, %pr_number, error = ?e,
                    "config-pr poll: failed to reconcile MergeConfigPr approval"
                );
            }
        }
    }

    // Reconcile: cancel pending approvals whose PR is no longer open.
    reconcile_stale_config_pr_approvals(coord, &open_prs, &scanned_agents);

    Ok(())
}

/// Cancel pending `MergeConfigPr` approvals whose PR is no longer open — merged
/// (incl. outside the approval flow) or closed. Without this the card lingers on
/// the dashboard forever, since the webhook only signals opened PRs and the
/// add-loop only ever queues.
///
/// `open_prs` is the set of `(agent, pr_number)` seen open this sweep and
/// `scanned_agents` the agents whose PR list was fetched successfully — the
/// reconcile is guarded to those so a transient list failure can't cancel a
/// still-valid approval.
fn reconcile_stale_config_pr_approvals(
    coord: &Arc<Coordinator>,
    open_prs: &std::collections::HashSet<(String, u64)>,
    scanned_agents: &std::collections::HashSet<String>,
) {
    let pending = match coord.approvals.pending() {
        Ok(p) => p,
        Err(e) => {
            tracing::warn!(error = ?e, "config-pr poll: pending() failed, skipping reconcile");
            return;
        }
    };
    for a in pending {
        if a.kind != hive_sh4re::approvals::ApprovalKind::MergeConfigPr
            || !scanned_agents.contains(a.agent.as_str())
        {
            continue;
        }
        let Ok(pr_number) = a.commit_ref.parse::<u64>() else {
            continue;
        };
        if open_prs.contains(&(a.agent.to_string(), pr_number)) {
            continue;
        }
        match coord
            .approvals
            .mark_cancelled(a.id, "config-pr poll (PR no longer open)")
        {
            Ok(_) => {
                tracing::info!(
                    agent = %a.agent, pr_number, id = a.id,
                    "config-pr poll: cancelled stale MergeConfigPr approval (PR merged/closed)"
                );
                coord.emit_approval_resolved(crate::coordinator::ApprovalResolved {
                    id: a.id,
                    agent: a.agent.as_str(),
                    approval_kind: "merge_config_pr",
                    sha_short: a
                        .fetched_sha
                        .as_deref()
                        .map(|s| s[..s.len().min(12)].to_owned()),
                    status: "cancelled",
                    note: Some("PR merged/closed outside the approval".to_owned()),
                    description: a.description.clone(),
                });
            }
            Err(e) => {
                tracing::warn!(
                    agent = %a.agent, id = a.id, error = ?e,
                    "config-pr poll: failed to cancel stale MergeConfigPr approval"
                );
            }
        }
    }
}