feat(P09): collector + aggregator (C-11/C-12/C-14) + drift aggregation (REQ-107)

scripts/orca-aggregate.sh: 10s aggregator, cluster.json merge + drift-events rsync + remediation trigger (P10b stub). scripts/orca- watchdog.sh: C-11 starvation detection. internal/cli/collector.go: orca collector start/stop/status. Tests: CLI + bats. ---ci--- project: orca phase: 09 milestone: v0.11 status: execute ---/ci---
2026-08-07 06:14:16 +00:00
parent 5f92196625
commit 5cbe3020d3
6 changed files with 828 additions and 0 deletions
@@ -0,0 +1,126 @@
+#!/usr/bin/env bats
+# Tests for scripts/orca-aggregate.sh + orca-watchdog.sh (P09, C-11,
+# REQ-107). Hermetic: exercises the aggregator against a fake peers.env
+# + state dir in a temp dir; the SSH path is exercised only when a
+# fake ssh shim is on PATH. The watchdog is exercised with a stale
+# cluster.json to confirm starvation detection.
+
+load test_helper
+
+@test "orca-aggregate.sh exists and is executable" {
+	[ -f "$SCRIPTS_DIR/orca-aggregate.sh" ]
+	[ -x "$SCRIPTS_DIR/orca-aggregate.sh" ]
+}
+
+@test "orca-watchdog.sh exists and is executable" {
+	[ -f "$SCRIPTS_DIR/orca-watchdog.sh" ]
+	[ -x "$SCRIPTS_DIR/orca-watchdog.sh" ]
+}
+
+@test "orca-aggregate.sh handles missing peers.env gracefully" {
+	tmp="$(mktemp -d)"
+	run env \
+		ORCA_PEERS_ENV="$tmp/nonexistent.env" \
+		ORCA_STATE_DIR="$tmp/state" \
+		ORCA_REMOTE_STATE_DIR="$tmp/remote-state" \
+		ORCA_REMOTE_DRIFT_DIR="$tmp/remote-drift" \
+		ORCA_APPLIED_DIR="$tmp/applied" \
+		bash "$SCRIPTS_DIR/orca-aggregate.sh"
+	assert_status 0 "$status"
+	# cluster.json written atomically with an empty peers array.
+	[ -f "$tmp/state/cluster.json" ]
+	assert_contains "$(cat "$tmp/state/cluster.json")" '"peers":[]'
+}
+
+@test "orca-aggregate.sh skips drift section when drift-events absent" {
+	tmp="$(mktemp -d)"
+	# No applied dir and no remote drift dir on lead → drift section
+	# must skip (no drift-events-aggregated.json written).
+	run env \
+		ORCA_PEERS_ENV="$tmp/nonexistent.env" \
+		ORCA_STATE_DIR="$tmp/state" \
+		ORCA_REMOTE_STATE_DIR="$tmp/remote-state" \
+		ORCA_REMOTE_DRIFT_DIR="$tmp/absent-drift" \
+		ORCA_APPLIED_DIR="$tmp/absent-applied" \
+		bash "$SCRIPTS_DIR/orca-aggregate.sh"
+	assert_status 0 "$status"
+	[ -f "$tmp/state/cluster.json" ]
+	# Drift aggregation skipped → aggregated file must NOT exist.
+	[ ! -f "$tmp/state/drift-events-aggregated.json" ]
+}
+
+@test "orca-watchdog.sh detects starvation with old cluster.json" {
+	tmp="$(mktemp -d)"
+	mkdir -p "$tmp/state"
+	cluster="$tmp/state/cluster.json"
+	echo '{}' >"$cluster"
+	# Set mtime to 120s ago so age (120) >= threshold (default 30).
+	old_ts="$(date -d '120 seconds ago' +%Y%m%d%H%M.%S)"
+	touch -t "$old_ts" "$cluster"
+	# logger is available in CI; capture the alert by overriding PATH
+	# with a fake logger that records to a file.
+	fake_bin="$tmp/bin"
+	mkdir -p "$fake_bin"
+	cat >"$fake_bin/logger" <<'LOGGER_EOF'
+#!/usr/bin/env bash
+echo "logger: $*" >>"$LOGGER_OUT"
+LOGGER_EOF
+	chmod +x "$fake_bin/logger"
+	env \
+		ORCA_STATE_DIR="$tmp/state" \
+		ORCA_WATCHDOG_THRESHOLD=30 \
+		LOGGER_OUT="$tmp/logger.out" \
+		PATH="$fake_bin:$PATH" \
+		bash "$SCRIPTS_DIR/orca-watchdog.sh"
+	[ -f "$tmp/logger.out" ]
+	out="$(cat "$tmp/logger.out")"
+	assert_contains "$out" "aggregator starvation detected"
+}
+
+@test "orca-watchdog.sh passes when cluster.json is fresh" {
+	tmp="$(mktemp -d)"
+	mkdir -p "$tmp/state"
+	echo '{}' >"$tmp/state/cluster.json"
+	# Fresh mtime (now); age 0 < threshold.
+	fake_bin="$tmp/bin"
+	mkdir -p "$fake_bin"
+	cat >"$fake_bin/logger" <<'LOGGER_EOF'
+#!/usr/bin/env bash
+echo "logger: $*" >>"$LOGGER_OUT"
+LOGGER_EOF
+	chmod +x "$fake_bin/logger"
+	env \
+		ORCA_STATE_DIR="$tmp/state" \
+		ORCA_WATCHDOG_THRESHOLD=30 \
+		LOGGER_OUT="$tmp/logger.out" \
+		PATH="$fake_bin:$PATH" \
+		bash "$SCRIPTS_DIR/orca-watchdog.sh"
+	if [ -f "$tmp/logger.out" ]; then
+		out="$(cat "$tmp/logger.out")"
+		! assert_contains "$out" "aggregator starvation detected" || {
+			echo "unexpected starvation alert for fresh cluster.json" >&2
+			return 1
+		}
+	fi
+}
+
+@test "orca-watchdog.sh alerts when cluster.json absent" {
+	tmp="$(mktemp -d)"
+	fake_bin="$tmp/bin"
+	mkdir -p "$fake_bin"
+	cat >"$fake_bin/logger" <<'LOGGER_EOF'
+#!/usr/bin/env bash
+echo "logger: $*" >>"$LOGGER_OUT"
+LOGGER_EOF
+	chmod +x "$fake_bin/logger"
+	env \
+		ORCA_STATE_DIR="$tmp/state" \
+		ORCA_WATCHDOG_THRESHOLD=30 \
+		LOGGER_OUT="$tmp/logger.out" \
+		PATH="$fake_bin:$PATH" \
+		bash "$SCRIPTS_DIR/orca-watchdog.sh"
+	[ -f "$tmp/logger.out" ]
+	out="$(cat "$tmp/logger.out")"
+	assert_contains "$out" "aggregator starvation detected"
+	assert_contains "$out" "cluster.json absent"
+}