#!/usr/bin/env bash
# T51: baseEjectionTime backoff behavior (outlierDetection consecutive5xxErrors)
#
# NOTE on manifest.yaml: the DestinationRule host is set to
# mixed-echo.istio-vt-t51.svc.cluster.local, NOT ...svc.homelab.local.
# This cluster's kubeadm clusterDomain is homelab.local, but Istio's internal
# Service registry / route-domain generation independently uses cluster.local
# regardless of the kubelet's actual --cluster-domain. Using svc.homelab.local
# in the DestinationRule host (or as request Host header) leaves
# outlier_detection completely unattached to the outbound cluster (verified
# via istio-proxy /config_dump) and routes traffic through the ALLOW_ANY
# wildcard vhost to PassthroughCluster instead. This script targets the
# Service ClusterIP directly so the Host header matches Istio's route-config
# domains (which include the raw ClusterIP) and the correctly-named
# "outbound|80||mixed-echo...svc.cluster.local" cluster is exercised.
set -euo pipefail

NS=istio-vt-t51
SCRIPT_DIR="$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)"

cleanup() {
  echo "[cleanup] deleting namespace ${NS}"
  kubectl delete namespace "${NS}" --wait=false --ignore-not-found
}
trap cleanup EXIT

echo "[setup] creating namespace ${NS}"
kubectl create namespace "${NS}"
kubectl label namespace "${NS}" istio-injection=enabled

echo "[setup] applying manifest.yaml"
kubectl apply -f "${SCRIPT_DIR}/manifest.yaml"

echo "[wait] client pod Ready"
kubectl -n "${NS}" wait --for=condition=Ready pod/client --timeout=90s

echo "[wait] good-mixed / bad-mixed deployments available"
kubectl -n "${NS}" wait --for=condition=available deploy/good-mixed deploy/bad-mixed --timeout=60s

BAD_IP=$(kubectl -n "${NS}" get pod -l health=bad -o jsonpath='{.items[0].status.podIP}')
CLUSTER_IP=$(kubectl -n "${NS}" get svc mixed-echo -o jsonpath='{.spec.clusterIP}')
echo "[info] BAD_IP=${BAD_IP} CLUSTER_IP=${CLUSTER_IP}"

echo "[info] sanity: confirm outlier_detection is attached to the named cluster"
kubectl -n "${NS}" exec client -c istio-proxy -- curl -s localhost:15000/config_dump 2>/dev/null | python3 -c "
import json,sys
d=json.load(sys.stdin)
for c in d['configs']:
    if c.get('@type','').endswith('ClustersConfigDump'):
        for dc in c.get('dynamic_active_clusters',[]):
            cluster = dc['cluster']
            if 'mixed-echo' in cluster.get('name',''):
                print('CLUSTER:', cluster['name'])
                print('outlier_detection:', json.dumps(cluster.get('outlier_detection','NONE')))
"

OUT="/tmp/ejection_timeline_$$.txt"
echo "[observe] 75 x 2s polling loop -> ${OUT}"
( for i in $(seq 1 75); do
    TS=$(date +%s)
    FLAGS=$(kubectl -n "${NS}" exec client -c istio-proxy -- curl -s localhost:15000/clusters 2>/dev/null | grep "${BAD_IP}" | grep -o 'health_flags::[a-zA-Z_,/]*')
    echo "${TS} ${FLAGS}"
    kubectl -n "${NS}" exec client -c curl -- curl -s -o /dev/null --max-time 2 "http://${CLUSTER_IP}/" >/dev/null 2>&1
    sleep 2
  done ) > "${OUT}"

echo "[result] timeline:"
cat "${OUT}"

echo "[analysis] contiguous EJECTED/HEALTHY segments (based on health_flags containing 'failed_outlier_check' vs 'healthy'; note: actual Envoy 1.30 flag string is lowercase '/failed_outlier_check', not the spec's assumed 'FAILED_OUTLIER_CHECK')"
python3 -c "
lines = open('${OUT}').read().strip().split(chr(10))
rows = []
for l in lines:
    parts = l.split(' ', 1)
    ts = int(parts[0])
    flag = parts[1].strip() if len(parts) > 1 else ''
    state = 'EJECTED' if 'failed_outlier_check' in flag else ('HEALTHY' if 'healthy' in flag else 'UNKNOWN')
    rows.append((ts, state, flag))
segments = []
cur_state = rows[0][1]
seg_start = rows[0][0]
for i in range(1, len(rows)):
    ts, state, flag = rows[i]
    if state != cur_state:
        segments.append((cur_state, seg_start, rows[i-1][0]))
        cur_state = state
        seg_start = ts
segments.append((cur_state, seg_start, rows[-1][0]))
t0 = rows[0][0]
for s in segments:
    state, start, end = s
    print(f'{state:8s} rel_start={start-t0:4d}s rel_end={end-t0:4d}s last-seen-span={end-start:4d}s')
"

echo "[done] see analysis above: baseEjectionTime=10s; first ejection segment should be ~10s, subsequent repeat ejections should be markedly longer (backoff)."