Two faults, both found by running the reset against the demo rather than by reading it, and both fixed on the box before this commit existed. The reset raced its own alerting container. Writing .env changes alerting's environment, so `docker compose up -d alerting` recreates it -- and the next line posted notification targets to it with no wait. On a busy box that lost: curl returned nothing, json.load threw on an empty string, and set -e killed the script. The damage is in the ordering: `docker compose down -v` runs near the top, so any failure after it leaves the public demo up, empty, and with the simulator still stopped, because the unit is only restarted on the last line. It has been surviving nightly on timing alone. It now polls /healthz for up to 60 seconds and fails loudly before the seed rather than after the wipe. Five dashboard panels were raw ClickHouse SQL, which dashboards refuse: validatePanel rejects query_language "sql" outright, because the time-range picker is injected as leading query terms and a SELECT has nowhere to put them. They were written that way because the pipe syntax has no time bucketing -- no timechart, no bin -- so a genuine time series is not available to a dashboard panel at all. Each is now the breakdown the panel was actually asking for: kubelet events by kind and host, DNS by type and result, queue depth by queue and host, IIS by site and host, HAProxy by backend and balancer. Both mistakes were mine and both were avoidable by reading: the rule about pipe-syntax-only dashboards is stated in terraform/README.md, and I had already read the line that says it. Verified on the demo: 50 hosts, 24 services, 1,922,128 records, all 13 dashboards and all 11 alert rules applied, simulator active. `system` appears on exactly 31 hosts, which is the Linux count -- no Windows host was given a journald stream.
144 lines
6.8 KiB
Bash
Executable File
144 lines
6.8 KiB
Bash
Executable File
#!/usr/bin/env bash
|
|
# Nightly reset for the demo.cairnobs.org stack: wipes every data volume
|
|
# and re-seeds from scratch, so the demo's timestamps stay recent, its
|
|
# incidents stay at the same recent offsets, and storage doesn't grow
|
|
# forever. Mirrors the teardown docs/phase-0-runbook.md and
|
|
# docs/phase-2-runbook.md already document (`docker compose down -v`),
|
|
# plus the seed sequence below.
|
|
#
|
|
# Seeding is data-first, config-second: dashboards and alert rules are
|
|
# applied after the backfill exists, so nothing renders an empty panel or
|
|
# evaluates against an empty table on its first pass.
|
|
#
|
|
# The live half of the demo (/hack/demo-simulator running as
|
|
# cairnobs-demo-simulator.service) is stopped for the duration and
|
|
# started again at the end -- it must not be pushing records into a stack
|
|
# that's being torn down, and it must re-register its agents against the
|
|
# fresh, empty `agents` table afterwards.
|
|
set -euo pipefail
|
|
|
|
DEMO_ROOT=${DEMO_ROOT:-/home/john/cairnobs-demo}
|
|
SEED_DIR="$DEMO_ROOT/hack/demo-seed"
|
|
CERTS="$DEMO_ROOT/hack/dev-certs/out"
|
|
BACKFILL=${BACKFILL:-168h}
|
|
RATE_SCALE=${RATE_SCALE:-0.5}
|
|
SIMULATOR_UNIT=cairnobs-demo-simulator.service
|
|
|
|
cd "$DEMO_ROOT"
|
|
|
|
ADMIN_PASSWORD_FILE="$DEMO_ROOT/.admin-password"
|
|
# Changing DEMO_PASSWORD means rebuilding the web image too: the login
|
|
# page prefills this account's credentials, and they're baked into the
|
|
# bundle at build time from VITE_DEMO_USERNAME/VITE_DEMO_PASSWORD in the
|
|
# demo host's docker-compose.override.yml. Change one without the other
|
|
# and the demo's own login form stops working.
|
|
DEMO_PASSWORD='CairnDemo_2026!'
|
|
# Generated fresh every run, never committed. This account exists only
|
|
# to mint the ALERTING_SERVICE_TOKEN a few lines below, and `docker
|
|
# compose down -v` above has already destroyed the previous one, so the
|
|
# value never needs to outlive a single reset -- there is nothing to
|
|
# remember and therefore no reason to hardcode it. It used to be a
|
|
# literal in this file, which put a working service-account password in
|
|
# the repo: anyone who could read the source could mint an
|
|
# alerting-evaluator token against the live demo at will, and rotating
|
|
# the token achieved nothing while the password that mints it stayed
|
|
# published.
|
|
#
|
|
# Unlike DEMO_PASSWORD above, this one is never shown to a user and is
|
|
# not baked into the web bundle, so randomising it breaks nothing.
|
|
EVALUATOR_PASSWORD="$(openssl rand -base64 24)"
|
|
|
|
echo "=== $(date -u +%FT%TZ) reset starting ==="
|
|
|
|
sudo systemctl stop "$SIMULATOR_UNIT" || true
|
|
|
|
docker compose down -v
|
|
docker compose up -d
|
|
echo "waiting for api to report healthy..."
|
|
until [ "$(docker inspect -f '{{.State.Health.Status}}' cairnobs-api 2>/dev/null)" = "healthy" ]; do sleep 2; done
|
|
|
|
# -seed-admin is idempotent and prints the password once -- capture it
|
|
# fresh each run rather than reusing a stale one from a prior reset.
|
|
ADMIN_PASSWORD=$(docker compose run --rm api -seed-admin 2>&1 | grep '^ password:' | awk '{print $2}')
|
|
echo "$ADMIN_PASSWORD" > "$ADMIN_PASSWORD_FILE"
|
|
chmod 600 "$ADMIN_PASSWORD_FILE"
|
|
|
|
export CAIRNOBSCTL_API_URL=http://localhost:8080
|
|
export CAIRNOBSCTL_ALERTING_API_URL=http://localhost:8081
|
|
ADMIN_TOKEN=$(echo "$ADMIN_PASSWORD" | ./bin/cairnobsctl users login admin)
|
|
export CAIRNOBSCTL_TOKEN="$ADMIN_TOKEN"
|
|
|
|
# The password reaches the CLI on stdin only -- `--password <value>` was
|
|
# removed deliberately (see cli/cmd/cairnobsctl/cmd_users.go).
|
|
echo "$DEMO_PASSWORD" | ./bin/cairnobsctl users create demo --role viewer >/dev/null
|
|
echo "$EVALUATOR_PASSWORD" | ./bin/cairnobsctl users create alerting-evaluator --role viewer >/dev/null
|
|
|
|
SVCTOKEN=$(echo "$EVALUATOR_PASSWORD" | ./bin/cairnobsctl users login alerting-evaluator)
|
|
printf 'COMPOSE_PROFILES=single-tenant\nALERTING_SERVICE_TOKEN=%s\n' "$SVCTOKEN" > .env && chmod 600 .env
|
|
docker compose up -d alerting
|
|
|
|
# Wait for it to answer before posting to it. Writing the .env above
|
|
# changes alerting's environment, so `up -d` *recreates* the container
|
|
# rather than leaving it running -- and the next line used to curl it
|
|
# immediately. On a busy box that lost the race: curl returned nothing,
|
|
# `json.load` got an empty string, and the reset died after `down -v` had
|
|
# already wiped everything, leaving the public demo up, empty, and with
|
|
# the simulator still stopped because the unit only restarts at the very
|
|
# end. Cheap to wait; expensive not to.
|
|
echo "waiting for alerting to answer..."
|
|
for i in $(seq 1 60); do
|
|
if curl -sf -o /dev/null http://localhost:8081/healthz; then break; fi
|
|
if [ "$i" -eq 60 ]; then
|
|
echo "!! alerting did not become ready within 60s -- stopping before the seed" >&2
|
|
exit 1
|
|
fi
|
|
sleep 1
|
|
done
|
|
|
|
# Three notification targets so the Alerts page shows rules routed to
|
|
# different destinations, the way a real deployment splits ops/security/
|
|
# platform. The URLs are deliberately inert placeholders on a domain
|
|
# reserved for documentation -- nothing is actually notified.
|
|
create_target() {
|
|
curl -s -X POST http://localhost:8081/targets \
|
|
-H "Authorization: Bearer $ADMIN_TOKEN" -H 'Content-Type: application/json' \
|
|
-d "{\"name\":\"$1\",\"kind\":\"webhook\",\"webhook_url\":\"$2\"}" \
|
|
| python3 -c 'import json,sys; print(json.load(sys.stdin)["id"])'
|
|
}
|
|
TARGET_OPS=$(create_target 'Ops on-call (placeholder)' 'https://example.com/webhooks/ops-oncall')
|
|
TARGET_SECURITY=$(create_target 'Security team (placeholder)' 'https://example.com/webhooks/security')
|
|
TARGET_PLATFORM=$(create_target 'Platform team (placeholder)' 'https://example.com/webhooks/platform')
|
|
|
|
echo "backfilling $BACKFILL of synthetic history..."
|
|
./bin/demo-simulator \
|
|
-addr 127.0.0.1:4317 -ca "$CERTS/ca.pem" -cert "$CERTS/client.pem" -key "$CERTS/client-key.pem" \
|
|
-backfill "$BACKFILL" -rate-scale "$RATE_SCALE" -live=false
|
|
|
|
# A handful of Windows-shaped records from the dedicated fixture as well:
|
|
# it's the tool the Windows ingest path is actually verified with, so
|
|
# keeping its output present means the demo and that check agree.
|
|
docker run --rm --network host -v "$DEMO_ROOT":/src -w /src/hack/windows-fixture -e GOCACHE=/tmp/gocache golang:1.25-bookworm \
|
|
go run . --addr 127.0.0.1:4317 --ca /src/hack/dev-certs/out/ca.pem --cert /src/hack/dev-certs/out/client.pem --key /src/hack/dev-certs/out/client-key.pem --count 5
|
|
|
|
echo "applying dashboards..."
|
|
for f in "$SEED_DIR"/dashboards/*.json; do
|
|
./bin/cairnobsctl dashboards apply "$f" >/dev/null
|
|
echo " $(basename "$f")"
|
|
done
|
|
|
|
echo "applying alert rules..."
|
|
TMP=$(mktemp -d)
|
|
trap 'rm -rf "$TMP"' EXIT
|
|
for f in "$SEED_DIR"/alerts/*.json.template; do
|
|
out="$TMP/$(basename "${f%.template}")"
|
|
sed -e "s/__TARGET_OPS__/$TARGET_OPS/" \
|
|
-e "s/__TARGET_SECURITY__/$TARGET_SECURITY/" \
|
|
-e "s/__TARGET_PLATFORM__/$TARGET_PLATFORM/" "$f" > "$out"
|
|
./bin/cairnobsctl alerts apply "$out" >/dev/null
|
|
echo " $(basename "$out")"
|
|
done
|
|
|
|
sudo systemctl start "$SIMULATOR_UNIT"
|
|
|
|
echo "=== $(date -u +%FT%TZ) reset complete ==="
|