#!/bin/sh
# Oikos deploy script — triggered by Gitea webhook on push to dtoro/oikos.
# Runs on mac-mini as non-root user via launchd unit oikos-deploy-webhook.service.
# Phase 6: CI-gated, version-tagged images, rolling restart, pre-deploy pg_dump.
#
# Plans implemented here:
#   D1 — CI gate: blocks deploy unless Gitea reports a green run for the SHA.
#   D2 — versioned images: tags every built image v$VERSION (from VERSION file),
#        keeps the last 3 tags per service for rollback.

# Notify on deploy failure via Matrix. Uses Oikos API to raise an event
# so the scheduler picks it up and alerts via the notifier.
notify_deploy_failure() {
    local reason="$1"
    local sha="${SHA:-unknown}"
    echo "NOTIFY: deploy failed — $reason"
    # Try to raise an event through the Oikos API (best-effort, silent failure)
    if [ -n "${OIKOS_API_TOKEN:-}" ]; then
        curl -sf -X POST "http://localhost:8090/api/v1/events" \
            -H "Authorization: Bearer $OIKOS_API_TOKEN" \
            -H "Content-Type: application/json" \
            -d "{\"type\":\"deploy.failed\",\"severity\":\"critical\",\"source\":\"webhook\",\"data\":{\"sha\":\"$sha\",\"reason\":\"$reason\"}}" \
            >/dev/null 2>&1 || true
    fi
    # Also try Matrix directly via the notifier's webhook endpoint if configured
    if [ -n "${MATRIX_WEBHOOK_URL:-}" ]; then
        curl -sf -X POST "$MATRIX_WEBHOOK_URL" \
            -H "Content-Type: application/json" \
            -d "{\"msgtype\":\"m.text\",\"body\":\"🚨 Deploy failed: $reason (sha: $sha)\"}" \
            >/dev/null 2>&1 || true
    fi
}

set -e

REPO_DIR="${REPO_DIR:-$PWD}"
COMPOSE_FILE="${COMPOSE_FILE:-docker-compose.yml}"
PROFILE="${PROFILE:-full}"
HEALTH_URL="${HEALTH_URL:-http://localhost:8090/healthz}"
DUMP_DIR="${DUMP_DIR:-/opt/oikos/backups}"
RETRIES=${RETRIES:-30}
SLEEP=${SLEEP:-2}

# CI gate (D1). Set GITEA_URL + GITEA_TOKEN to enable; without them the gate
# is skipped with a warning (dev/local deploys). Owner/repo default to the
# canonical homelab repo but can be overridden or derived from the git remote.
GITEA_URL="${GITEA_URL:-}"
GITEA_TOKEN="${GITEA_TOKEN:-}"
GITEA_OWNER="${GITEA_OWNER:-dtoro}"
GITEA_REPO="${GITEA_REPO:-oikos}"
CI_POLL_INTERVAL="${CI_POLL_INTERVAL:-15}"
CI_TIMEOUT="${CI_TIMEOUT:-1200}"

# Serialize deploys: the webhook runs this script in a background goroutine and
# the CI gate can hold a deploy open for many minutes, so a second push during
# that window would otherwise race on git/pg_dump/compose. mkdir is atomic on
# POSIX (no flock dependency — macOS lacks it). The stale-pid check recovers
# if a previous deploy was SIGKILLed.
LOCKDIR="${LOCKDIR:-/tmp/oikos-deploy.lock}"
if ! mkdir "$LOCKDIR" 2>/dev/null; then
    oldpid=$(cat "$LOCKDIR/pid" 2>/dev/null || echo "")
    if [ -n "$oldpid" ] && kill -0 "$oldpid" 2>/dev/null; then
        echo "deploy already in progress (pid $oldpid) — exiting"
        exit 0
    fi
    echo "removing stale deploy lock (pid ${oldpid:-?} not running)"
    rm -rf "$LOCKDIR"
    mkdir "$LOCKDIR"
fi
echo $$ > "$LOCKDIR/pid"
trap 'rc=$?; rm -rf "$LOCKDIR" 2>/dev/null || true; if [ "$_ok" != "1" ]; then notify_deploy_failure "deploy aborted (exit $rc)"; fi' EXIT
_ok=0

cd "$REPO_DIR"

echo "=== oikos deploy: $(date) ==="

# Resolve the SHA we are ABOUT to deploy from the remote (read-only: no working
# tree mutation yet) so the CI gate can run before anything is touched. Keep the
# full SHA (REMOTE_FULL) for the post-pull equality check; the 12-char form is
# only for display and the Gitea status API (which accepts any unique prefix).
REMOTE_FULL=$(git ls-remote origin refs/heads/main 2>/dev/null | awk '{print $1}')
if [ -z "$REMOTE_FULL" ]; then
    echo "ERROR: could not resolve origin/main (offline?) — aborting before any change"
    exit 1
fi
REMOTE_SHA=$(printf '%s' "$REMOTE_FULL" | cut -c1-12)
echo "remote SHA: $REMOTE_SHA"

# ── 1. CI gate (plan D1) ──────────────────────────────────────────────────
# Runs BEFORE pg_dump/pull/build: a red or hung pipeline must not leave the
# tree half-deployed. On failure|error it refuses; on success it proceeds; on
# "no CI signal at all" (Actions unconfigured / token rejected) it warns and
# proceeds rather than bricking every deploy.
echo "[1/8] verify CI status for $REMOTE_SHA"
verify_ci() {
    sha=$1
    if [ -z "$GITEA_URL" ] || [ -z "$GITEA_TOKEN" ]; then
        echo "SKIP: GITEA_URL/GITEA_TOKEN not set — CI gate disabled. Set both to enforce."
        return 0
    fi
    # Derive owner/repo from the origin remote when the defaults don't apply.
    origin=$(git remote get-url origin 2>/dev/null || echo "")
    seg=
    case "$origin" in
        *@*:*)  # SSH: git@host:owner/repo.git
            seg=${origin##*:}; seg=${seg%.git}
            ;;
        http://*|https://*)  # HTTPS: scheme://host/owner/repo.git
            seg=${origin#*://}; seg=${seg#*/}; seg=${seg%.git}
            ;;
    esac
    case "$seg" in
        */*)
            GITEA_OWNER=${seg%%/*}
            GITEA_REPO=${seg#*/}
            ;;
    esac

    api="$GITEA_URL/api/v1/repos/$GITEA_OWNER/$GITEA_REPO/commits/$sha/status"
    body=$(mktemp)
    elapsed=0
    saw_ci=0          # became 1 once we observed a real status (pending/success/...)
    no_signal=0       # consecutive responses with no usable status
    while [ "$elapsed" -lt "$CI_TIMEOUT" ]; do
        # Pass the token via curl --config stdin so it never appears in argv
        # (visible via ps). Don't use -f: we want the HTTP code on 4xx.
        code=$(printf 'header = "Authorization: token %s"\n' "$GITEA_TOKEN" | \
            curl -sS -o "$body" -w '%{http_code}' --config - "$api" 2>/dev/null) || code="000"
        state=$(sed -n 's/.*"state"[[:space:]]*:[[:space:]]*"\([^"]*\)".*/\1/p' "$body" | head -n1)

        case "$code" in
            200)
                case "$state" in
                    success)
                        rm -f "$body"
                        echo "CI: green for $sha after ${elapsed}s"
                        return 0
                        ;;
                    failure|error)
                        rm -f "$body"
                        echo "ERROR: CI $state for $sha — refusing to deploy."
                        echo "       See $GITEA_URL/$GITEA_OWNER/$GITEA_REPO/actions"
                        return 1
                        ;;
                    pending|"")
                        saw_ci=1
                        no_signal=0
                        ;;
                esac
                ;;
            404)
                # No status checks exist for this commit (Actions not configured
                # / no runner has reported). Can't gate — warn + proceed.
                rm -f "$body"
                echo "WARN: Gitea has no CI status for $sha (404)."
                echo "      Is Gitea Actions configured with a runner for $GITEA_OWNER/$GITEA_REPO?"
                echo "      Proceeding without a gate."
                return 0
                ;;
            401|403)
                rm -f "$body"
                echo "WARN: GITEA_TOKEN rejected by Gitea ($code) — cannot verify CI."
                echo "      Fix the token to enforce the gate; proceeding without one."
                return 0
                ;;
            *)
                # Network blip / 5xx / 000: retry, but count as no-signal.
                no_signal=$((no_signal + 1))
                ;;
        esac

        # If we never get a usable signal after ~1 min, assume CI is unreachable
        # rather than burn the full timeout and brick deploys.
        if [ "$saw_ci" -eq 0 ] && [ "$no_signal" -ge 4 ]; then
            rm -f "$body"
            echo "WARN: no CI signal from Gitea after ${elapsed}s (last code=$code)."
            echo "      CI may be down or misconfigured; proceeding without a gate."
            return 0
        fi

        sleep "$CI_POLL_INTERVAL"
        elapsed=$((elapsed + CI_POLL_INTERVAL))
        printf '\rCI: waiting (%ss, code=%s state=%s)...' "$elapsed" "$code" "${state:-none}"
    done
    rm -f "$body"
    echo ""
    echo "ERROR: CI did not reach a terminal state within ${CI_TIMEOUT}s for $sha — refusing to deploy."
    return 1
}
verify_ci "$REMOTE_SHA" || exit 1

# ── 2. Pre-deploy pg_dump for rollback safety (plan O1) ───────────────────
echo "[2/8] pre-deploy pg_dump"
DUMP_FILE="$DUMP_DIR/pre-deploy-$REMOTE_SHA.sql"
mkdir -p "$DUMP_DIR"
docker compose exec -T postgres pg_dump -U oikos oikos > "$DUMP_FILE" 2>/dev/null || \
    echo "WARNING: pg_dump failed — rollback will not have a recovery point"

# ── 3. Update working tree to the verified commit ─────────────────────────
echo "[3/8] git pull (ff-only)"
git pull --ff-only origin main
# TOCTOU guard: if origin/main advanced during the CI wait + pg_dump, the pull
# fast-forwards PAST the SHA we verified without re-checking its CI. Refuse
# rather than ship an unverified commit — a retry will verify the new tip.
PULLED_FULL=$(git rev-parse HEAD)
if [ "$PULLED_FULL" != "$REMOTE_FULL" ]; then
    echo "ERROR: origin/main advanced during deploy (verified $REMOTE_SHA, now at $(git rev-parse --short HEAD)) — aborting; retry verifies the new tip"
    exit 1
fi
SHA=$(git rev-parse --short HEAD)
echo "SHA (deployed): $SHA"

# Resolve the deploy version AFTER pull (D2) so the tag matches the code being
# built. Compose interpolates $OIKOS_VERSION into each service's image: tag.
VERSION_FILE="$REPO_DIR/VERSION"
if [ -f "$VERSION_FILE" ]; then
    OIKOS_VERSION="v$(head -n1 "$VERSION_FILE" | tr -d '[:space:]')"
    export OIKOS_VERSION
    echo "VERSION: $OIKOS_VERSION"
else
    echo "WARNING: VERSION file missing — images will use :latest (rollback unavailable)"
fi

# ── 4. Build version-tagged images (plan D2) ──────────────────────────────
echo "[4/8] docker compose build"
DOCKER_BUILDKIT=1 docker compose --profile "$PROFILE" build \
    --build-arg BUILDKIT_INLINE_CACHE=1

# ── 5. Rolling restart ────────────────────────────────────────────────────
echo "[5/8] docker compose up -d"
docker compose --profile "$PROFILE" up -d --remove-orphans

# ── 6. Prune old image tags — keep the 3 newest per service so rollback ────
#    (OIKOS_VERSION=v0.x.y docker compose up) stays available. The repo list
#    is derived from compose so it can't drift from the image: names.
echo "[6/8] prune old image tags (keep 3)"
if [ -n "$OIKOS_VERSION" ]; then
    images=$(docker compose --profile "$PROFILE" config --images 2>/dev/null || true)
    if [ -z "$images" ]; then
        images="oikos-api oikos-scheduler oikos-notifier oikos-migrate oikos-seed oikos-nomos oikos-web"
    fi
    printf '%s\n' $images | sed 's/:.*//' | grep '^oikos-' | sort -u | while read -r repo; do
        docker image ls "$repo" --format '{{.Tag}}' 2>/dev/null | grep '^v' | sort -rV | tail -n +4 | while read -r tag; do
            docker rmi "$repo:$tag" >/dev/null 2>&1 || true
        done
    done
fi

# ── 7. Health check wait ──────────────────────────────────────────────────
echo "[7/8] health check"
healthy=0
for i in $(seq 1 $RETRIES); do
    if curl -sf "$HEALTH_URL" > /dev/null 2>&1; then
        echo "healthy after ${i}s"
        healthy=1
        break
    fi
    sleep "$SLEEP"
done
if [ "$healthy" -ne 1 ]; then
    echo "ERROR: health check failed after $((RETRIES * SLEEP))s"
    exit 1
fi

# ── 8. Seed secrets into Infisical (idempotent) ───────────────────────────
echo "[8/8] seed secrets"
if [ -f "$REPO_DIR/scripts/seed-secrets.sh" ]; then
    REPO_DIR="$REPO_DIR" sh "$REPO_DIR/scripts/seed-secrets.sh" || \
        echo "WARNING: secret seeding failed"
else
    echo "SKIP: seed-secrets.sh not found"
fi

# All steps completed successfully — clear failure trap
_ok=1
exit 0