Files
aevolution/tools/ops/load-drill.sh
T
dingoandClaude Opus 5.5 46c106409e
Foundation / checks (checks) (push) Successful in 5s
Foundation / checks (client) (push) Successful in 5s
Foundation / checks (ecs) (push) Successful in 4s
Foundation / checks (network) (push) Successful in 4s
Foundation / checks (postgres) (push) Successful in 4s
Foundation / checks (tsan) (push) Successful in 5s
Foundation / checks (deploy) (push) Successful in 4m33s
Foundation / postgres-tools (push) Successful in 2m14s
Foundation / server-image (push) Successful in 5m21s
Foundation / image (push) Successful in 6m18s
feat(admin): session windows in SQL, opened from the admin portal (schema 62)
The dev session window moves from Git to SQL (ADR-0087 amendment): the admin
portal opens, extends (30 min to 4 h from now) and closes it, and the cellular
region re-reads it every 5 s with --session-window-source db. No push, sync or
restart per window; a closed window keeps the region running instead of
crash-looping.

- Schema 62: append-only session_windows (4 h cap, no rewrites),
  load_session_window_v1 for the runtime group, admin_set_session_window_v1
  and admin_session_windows_v1 for a dedicated admin group role, granted and
  audited each migration run when EVO_PG_ADMIN_ROLE is set (aevolution_admin in
  DEV). Schema 61's views are revoked from the runtime group.
- RegionSessions admission switch: while closed every Hello is refused before
  signature work and closing closes open sessions (new sessions were still
  admitted after expiry before).
- Admin API: GET/POST /api/regions/<n>/session-window, window state in the
  regions list; the portal's session-window card with history.
- Git: REGION_SESSION_WINDOW and the pre-push refresh rule are removed; the
  policy check requires the database source and refuses a committed window.

Needs the new image and the aevolution_admin role (with the portal's login in
aevolution-admin-pg) before a sync.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
2026-10-27 15:13:02 -04:00

181 lines
9.7 KiB
Bash

#!/usr/bin/env bash
# M10.9d load drill (ADR-0097) for evolution-dev only: the reference-worker qualification run.
#
# tools/ops/load-drill.sh <tokens-dir> [--context <name>] [--steps N[,N]...] [--hold S]
# [--warmup S] [--grants-per-bot C] [--connect HOST:PORT] [--server-name NAME]
#
# <tokens-dir> holds dev-ca.pem (the current dev session CA) and load-grants.bin, a bundle
# made where the dev issuer key lives:
# evo-session-probe grants --issuer-key issuer.key --world W --frame F --region R \
# --account <HEX32> --accounts <most bots> --count <grants per bot> --lifetime 14400 \
# --out load-grants.bin
# The issuer key never reaches the machine running the drill. Requires kubectl access to
# evolution-dev (patch Deployment/evo-region, read pods and logs), `evo-load` (EVO_LOAD,
# default: on PATH) and python3.
#
# Steps: refuse unless the session window outlasts the drill; enable synthetic inputs on
# evo-region (the drill-time override of ADR-0073, never committed) and wait for the new
# pod to be Ready and for a calm `m10 load-report` (the server prints one every 10 s; the
# governor acts on a full 40-tick window); run evo-load through the public route; take the
# drill pod's whole log with timestamps; then restore REGION_INPUTS to the Git value (also
# on failure) and wait for a Ready pod other than the drill pod, so Argo returns to Synced. The report matches each step's hold
# (evo-load's hold_utc) with the server's reports in it and prints, per step, the client
# verdict and the server's worst tick work, session work, governor state and tick debt.
# Stage and World0 are never touched: the namespace is fixed.
set -euo pipefail
usage() {
echo "usage: $0 <tokens-dir> [--context <name>] [--steps N[,N]...] [--hold S] [--warmup S]" \
"[--grants-per-bot C] [--connect HOST:PORT] [--server-name NAME]" >&2
exit 2
}
[ $# -ge 1 ] || usage
tokens="$1"; shift
k=(kubectl --request-timeout=30s)
steps=20,24,28,32 hold=60 warmup=10 per_bot=4
connect=void.aeternalisgames.com:7400 server_name=void.aeternalisgames.com
while [ $# -gt 0 ]; do
case "$1" in
--context) [ -n "${2:-}" ] || usage; k+=(--context "$2"); shift 2 ;;
--steps) [ -n "${2:-}" ] || usage; steps="$2"; shift 2 ;;
--hold) [ -n "${2:-}" ] || usage; hold="$2"; shift 2 ;;
--warmup) [ -n "${2:-}" ] || usage; warmup="$2"; shift 2 ;;
--grants-per-bot) [ -n "${2:-}" ] || usage; per_bot="$2"; shift 2 ;;
--connect) [ -n "${2:-}" ] || usage; connect="$2"; shift 2 ;;
--server-name) [ -n "${2:-}" ] || usage; server_name="$2"; shift 2 ;;
*) usage ;;
esac
done
ns=evolution-dev # Fixed: qualification runs on the dev reference worker only.
tool="${EVO_LOAD:-evo-load}"
py="$(command -v python3 || command -v python || true)"
for f in dev-ca.pem load-grants.bin; do
[ -f "$tokens/$f" ] || { echo "missing $tokens/$f" >&2; exit 2; }
done
command -v "$tool" > /dev/null || [ -x "$tool" ] || { echo "evo-load not found (set EVO_LOAD)" >&2; exit 2; }
[ -n "$py" ] || { echo "python3 is required" >&2; exit 2; }
selector=app.kubernetes.io/name=evo-region
work="$(mktemp -d)"
chmod 700 "$work"
drill_pod=""
say() { echo "$(date -u +%H:%M:%S) $*"; }
fail() { say "FAIL $*"; exit 1; }
env_of() {
"${k[@]}" get deployment -n "$ns" evo-region \
-o jsonpath="{.spec.template.spec.containers[0].env[?(@.name==\"$1\")].value}"
}
ready_pod() { # Waits up to $2 seconds for a Ready pod other than $1.
local old="$1" limit="${2:-240}"
for _ in $(seq 1 $((limit * 5))); do
for p in $("${k[@]}" get pods -n "$ns" -l "$selector" \
-o go-template='{{range .items}}{{if not .metadata.deletionTimestamp}}{{.metadata.name}}{{"\n"}}{{end}}{{end}}'); do
[ "$p" = "$old" ] && continue
[ "$("${k[@]}" get pod -n "$ns" "$p" -o jsonpath='{.status.conditions[?(@.type=="Ready")].status}' 2> /dev/null)" = True ] &&
{ echo "$p"; return 0; }
done
sleep 0.2
done
return 1
}
restore() {
set +e
say "restoring the Git configuration of evo-region (REGION_INPUTS empty)"
"${k[@]}" set env -n "$ns" deployment/evo-region REGION_INPUTS= > /dev/null
# Ready means a pod other than the drill's, i.e. the restored configuration.
ready_pod "${drill_pod:-none}" 240 > /dev/null && say "restored: evo-region Ready" ||
say "WARN evo-region not Ready after restore"
[ "$("${k[@]}" get deployment -n "$ns" evo-region \
-o jsonpath='{.spec.template.spec.containers[0].env[?(@.name=="REGION_INPUTS")].value}')" = "" ] ||
say "WARN REGION_INPUTS is not empty after restore"
say "drill files kept in $work"
}
world="$(env_of REGION_WORLD)" region="$(env_of REGION_ID)" frame="$(env_of REGION_FRAME)"
[ -n "$world" ] && [ -n "$region" ] && [ -n "$frame" ] || fail "cannot read the region identity"
[ -z "$(env_of REGION_INPUTS)" ] || fail "REGION_INPUTS is already set: another drill may be running"
# The drill must end before the session window closes: the window is in SQL (ADR-0087
# amendment), so open it in the admin portal for at least the drill's duration first.
count=$(echo "$steps" | tr ',' '\n' | wc -l)
most=$(echo "$steps" | tr ',' '\n' | sort -n | tail -1)
needed=$(( count * (hold + warmup + most / 10 + 5) + 600 ))
say "the drill needs about ${needed} s of open session window (admin portal)"
size=$(wc -c < "$tokens/load-grants.bin")
[ $(( size / 160 )) -ge $(( most * per_bot )) ] ||
fail "load-grants.bin holds $(( size / 160 )) grants; $most bots x $per_bot need $(( most * per_bot ))"
image="$("${k[@]}" get deployment -n "$ns" evo-region -o jsonpath='{.spec.template.spec.containers[0].image}')"
say "drill: world $world region $region image $image steps $steps hold ${hold}s"
trap restore EXIT
before="$("${k[@]}" get pods -n "$ns" -l "$selector" -o jsonpath='{.items[0].metadata.name}')"
say "enabling synthetic inputs in $ns (drill-time override, restored on exit)"
"${k[@]}" set env -n "$ns" deployment/evo-region REGION_INPUTS=development-v1 > /dev/null
pod="$(ready_pod "$before" 240)" || fail "no Ready evo-region pod with synthetic inputs"
drill_pod="$pod"
# Warm up: two consecutive calm reports (refusing=0 shed=0, about 10 s apart), so the
# governor's window no longer holds the restart's first ticks when the bots arrive (the
# second reference run met one calm report, then a refusal at the first login).
calm=""
for _ in $(seq 1 60); do
sleep 3
last2="$("${k[@]}" logs -n "$ns" --since=25s "$pod" 2> /dev/null | grep "m10 load-report" | tail -2)"
last="$(echo "$last2" | tail -1)"
[ "$(echo "$last2" | grep -c "refusing=0 shed=0")" = 2 ] && { calm=1; break; }
done
[ -n "$calm" ] || fail "the region did not report calm twice within 3 minutes of starting: $last"
say "region calm: $last"
say "running evo-load against $connect"
set +e
"$tool" --connect "$connect" --server-name "$server_name" --ca "$tokens/dev-ca.pem" \
--grants "$tokens/load-grants.bin" --grants-per-bot "$per_bot" --world "$world" \
--frame "$frame" --region "$region" --steps "$steps" --hold "$hold" --warmup "$warmup" \
--threads 8 --ramp-ms 500 --per-client > "$work/load.log" 2>&1
load_exit=$?
set -e
sleep 12 # One more server report after the last hold.
"${k[@]}" logs -n "$ns" --timestamps "$pod" > "$work/server.log" 2> /dev/null ||
say "WARN could not read the drill pod's log"
grep -E "^load (step|response_ms|outcome_ms|snapshots|starved|verdict|result)" "$work/load.log" || true
[ "$load_exit" = 0 ] || say "WARN evo-load exited $load_exit (see $work/load.log)"
"$py" - "$work/load.log" "$work/server.log" <<'PY'
import re, sys
from datetime import datetime
load, server = sys.argv[1], sys.argv[2]
reports = []
for line in open(server, encoding='utf-8', errors='replace'):
m = re.match(r'(\S+) m10 load-report (.*)', line.strip())
if not m:
continue
stamp = m.group(1)[:19]
t = datetime.strptime(stamp, '%Y-%m-%dT%H:%M:%S').replace(tzinfo=None)
epoch = int((t - datetime(1970, 1, 1)).total_seconds())
fields = dict(kv.split('=') for kv in m.group(2).split())
reports.append((epoch, {k: int(v) for k, v in fields.items()}))
print(f"drill server reports={len(reports)}")
for line in open(load, encoding='utf-8', errors='replace'):
m = re.match(r'load step N=(\d+) .*hold_utc=(\d+)\.\.(\d+)', line)
if not m:
continue
n, t0, t1 = int(m.group(1)), int(m.group(2)), int(m.group(3))
# A report covers the 2 s before it: only reports wholly inside the hold count.
inside = [r for t, r in reports if t0 + 2 <= t <= t1]
if not inside:
print(f"drill step N={n} server_reports=0")
continue
worst = lambda key: max(r.get(key, 0) for r in inside)
mean = lambda key: sum(r.get(key, 0) for r in inside) / len(inside)
sessions = max(1, worst('sessions'))
# Budgets v2 (schemas/network/m10-budgets-v2.md): tick work p95 <= 25 ms over all ticks,
# ticks that commit a batch <= 50 ms; session layer p99 <= 0.5 ms per session; no debt.
within = (worst('p95_us') <= 25000 and worst('commit_max_us') <= 50000 and
worst('session_p99_us') <= 500 * sessions and worst('tick_debt') == 0)
print(f"drill step N={n} server_reports={len(inside)} sessions_max={worst('sessions')} "
f"work_p95_max_ms={worst('p95_us') / 1000:.1f} work_p99_max_ms={worst('p99_us') / 1000:.1f} "
f"commit_max_ms={worst('commit_max_us') / 1000:.1f} other_max_ms={worst('other_max_us') / 1000:.1f} "
f"session_p95_max_ms={worst('session_p95_us') / 1000:.1f} "
f"session_p99_max_ms={worst('session_p99_us') / 1000:.1f} "
f"per_session_p99_ms={worst('session_p99_us') / 1000 / sessions:.3f} "
f"poll_avg_ms={mean('poll_avg_us') / 1000:.2f} replicate_avg_ms={mean('replicate_avg_us') / 1000:.2f} "
f"refusing_max={worst('refusing')} shed_max={worst('shed')} "
f"tick_debt_max={worst('tick_debt')} server_within_budget={int(within)}")
PY
say "drill complete (evo-load exit $load_exit); restoring"