=== CMD: kubectl apply -f manifest.yaml ===
pod/client created
deployment.apps/paralleltiming-echo created
service/paralleltiming-echo created
=== CMD: kubectl -n istio-vt-t79 wait --for=condition=available deploy/paralleltiming-echo --timeout=60s ===
deployment.apps/paralleltiming-echo condition met
=== CMD: kubectl -n istio-vt-t79 wait --for=condition=Ready pod/client --timeout=60s ===
pod/client condition met
=== CMD 1: kubectl apply -f manifest.yaml (already applied above; re-noting) ===
(applied earlier - see top of this file)

=== CMD 2: POD=$(kubectl -n istio-vt-t79 get pod -l app=paralleltiming-echo -o jsonpath='{.items[0].metadata.name}') ===
POD=paralleltiming-echo-55f7966f47-5srfr

=== CMD 3: T0=$(date +%s); kubectl -n istio-vt-t79 delete pod $POD --wait=true; T1=$(date +%s); echo total_termination_seconds=$((T1-T0)) ===
pod "paralleltiming-echo-55f7966f47-5srfr" deleted from istio-vt-t79 namespace
total_termination_seconds=61

=== DIAGNOSTIC: unexpected result investigation ===
Primary spec run (commands 1-3) gave total_termination_seconds=61, which is neither
close to the serial-sum hypothesis (35s) nor the parallel/max hypothesis (20-30s)
per pass_criteria. Investigated root cause with extra (non-spec) diagnostics below.

--- Event check on first terminated pod ---
$ kubectl -n istio-vt-t79 get events (filtered for paralleltiming, Killing/FailedPreStopHook)
paralleltiming-echo-55f7966f47-5srfr Killing "Stopping container app"
paralleltiming-echo-55f7966f47-5srfr Killing "Stopping container istio-proxy"
paralleltiming-echo-55f7966f47-5srfr FailedPreStopHook "PreStopHook failed"

--- Root cause check: does the app image have a shell? ---
$ kubectl -n istio-vt-t79 exec <app-pod> -c app -- sh -c "echo hi"
error: ... OCI runtime exec failed: exec failed: unable to start container process:
exec: "sh": executable file not found in $PATH: unknown

=> CONFIRMED: hashicorp/http-echo:0.2.3 is a minimal/scratch-style image with NO shell.
   The manifest's preStop hook (["sh","-c","sleep 15"]) FAILS INSTANTLY (exec error),
   it never actually sleeps 15s. This is a test-fixture defect, not an Istio behavior.

--- Re-run #2 (fresh pod) with per-container state polling every ~1s during termination ---
POD=paralleltiming-echo-55f7966f47-9tzmd
elapsed(kubectl delete --wait)=61.1s
Delete triggered at approx 2026-07-05T00:23:25.49Z (poll loop timestamp, next tick after 1s sleep)
Poll of containerStatuses[*].state every ~1s shows BOTH containers "running" continuously
until timestamp 1783211066.24 (poll tick), at which point:
  app:         terminated {exitCode:137, reason:"Error",     finishedAt:"2026-07-05T00:24:25Z"}
  istio-proxy: terminated {exitCode:0,   reason:"Completed",  finishedAt:"2026-07-05T00:23:45Z"}

Delta analysis (relative to delete-trigger ~00:23:25.49Z):
  istio-proxy finishedAt 00:23:45Z  -> ~19.5s elapsed  ~= terminationDrainDuration (20s), clean exit (code 0)
  app         finishedAt 00:24:25Z  -> ~59.5s elapsed  ~= terminationGracePeriodSeconds (60s), SIGKILL (code 137)

=> MECHANISM CONFIRMED:
   1. istio-proxy (Envoy/pilot-agent) received SIGTERM at pod-delete time (no preStop
      hook of its own) and completed its internal graceful drain in ~20s, matching
      proxy.istio.io/config terminationDrainDuration=20s, exiting cleanly (code 0)
      INDEPENDENTLY of the app container's state.
   2. app container's preStop hook failed instantly (missing shell in image) so it
      never actually slept 15s; more importantly the http-echo binary does not trap/
      handle SIGTERM (classic PID-1-without-init-handler issue), so it kept running
      until kubelet's terminationGracePeriodSeconds (60s) expired and force-killed it
      (SIGKILL, exit 137).
   3. Because istio-proxy fully exited at ~20s WHILE the app container was still very
      much alive (running) until ~60s, this is direct, unambiguous proof that Envoy's
      drain and the app container's shutdown/termination proceed IN PARALLEL/independently
      -- if they were serial (e.g. Envoy waiting on app container exit first), Envoy could
      not have exited at t=20s while app was still running.
   4. However, the specific NUMERIC pass_criteria in the spec (total pod-delete wall time
      in ~20-30s window) is NOT met -- observed total (~61s) is dominated by an unrelated
      confound (broken preStop fixture + app not handling SIGTERM => full TGPS + SIGKILL),
      not by the intended preStop(15s)-vs-drain(20s) race. The chosen test fixture
      (hashicorp/http-echo:0.2.3, TGPS=60s) does not correctly exercise the intended
      comparison for the *aggregate* total_termination_seconds metric, even though the
      underlying architecture (parallel container shutdown) IS independently confirmed
      by the per-container timestamp evidence above.