jcoffey-dev
bcb9a01cd6
The demo had 75k generic records across eight host-0N/service pairs, one dashboard, one alert rule, and -- because nothing ever called AgentControl.CheckIn -- a completely empty Agents page. /hack/demo-simulator replaces the generic data with a fictional but coherent fleet: 14 hosts running nginx, an API tier, workers, Postgres, Redis, mail, Linux journals and Windows event logs, whose messages and attributes look like what those services actually write. It backfills a week (~370k records, ~20s) and then keeps running. Running continuously is the point, not an implementation detail. Three things the demo has to show are only true if data keeps arriving: the Agents page marks a host stale once check-ins stop, alert rules evaluate over trailing windows and would freeze in one state against a static dataset, and any "last 15 minutes" view is empty on data that stopped growing overnight. It also emits metrics/heartbeats and answers CheckIn faithfully enough that the remote-config editor's pending -> applied transition works end to end. Seeded incidents give the data something to find: an api-02 outage with matching slow queries on db-01, 5xx at the edge and cascading job failures; an SSH probe burst; a spam wave; a disk filling up; and one decommissioned host left deliberately stale. /hack/demo-seed holds the rest of the deployment -- the nightly reset, eight dashboards (64 panels, every viz type but line), eleven alert rules across three notification targets, and the systemd unit. Rule thresholds are calibrated against what the simulator actually produces: the first pass had four rules whose thresholds the traffic could never reach and one that fired during normal operation. No line charts: dashboard panels reject the raw-SQL escape hatch, and the pipe language has no time-bucketing, so a real time axis isn't expressible today. Noted in demo-seed/README.md rather than papered over.
demo-seed
Everything the public demo deployment (demo.cairnobs.org) is built out
of, kept in the repo rather than only on the box so a demo can be rebuilt
from scratch and so its dashboards and rules are reviewable like any
other code.
reset-demo.sh— the nightly reset. Wipes every volume, brings the stack back up, re-seeds users, notification targets, a week of synthetic history, dashboards, and alert rules. Run from cron at 04:00 on the demo host.dashboards/*.json— one file per dashboard, in the shapePOST /dashboards/importconsumes (identical to whatGET /dashboards/{id}/exportand the web UI's Export JSON button produce, so a dashboard edited in the UI can be exported straight back into this directory).alerts/*.json.template— one file per alert rule, in the shapePOST /rulesconsumes.__TARGET_OPS__/__TARGET_SECURITY__/__TARGET_PLATFORM__are substituted at apply time with the IDs of the three notification targetsreset-demo.shcreates: every reset starts from an empty database, so the IDs can't be baked in.cairnobs-demo-simulator.service— systemd unit for the live half of the demo,/hack/demo-simulator. Installed at/etc/systemd/system/on the demo box.
Prefilled login
The demo's login page comes up with the read-only demo account already
in both fields, so a visitor doesn't need credentials handed to them.
That's a build-time opt-in, off everywhere else: the web image is built
with VITE_DEMO_USERNAME/VITE_DEMO_PASSWORD (set in the demo host's
docker-compose.override.yml), and the login page prefills only when it
has both. Any deployment that doesn't set them gets the ordinary empty
form -- see web/src/lib/api.ts's demoUsername.
The password is baked into the static bundle, which is fine for exactly
this case and nothing else: a Viewer-role account on a deployment whose
database is wiped and reseeded nightly. It has to match DEMO_PASSWORD
in reset-demo.sh, and changing that means rebuilding the web image.
Why the demo needs a long-running process
Three things the demo has to show are only true if data keeps arriving, and no amount of one-shot seeding fixes any of them:
- Agents. The Agents page is populated by the
AgentControl.CheckInRPC, and marks a host stale once it stops calling in. A fleet seeded once at 04:00 is entirely stale by 04:10. - Alerts. Rules evaluate over trailing windows (
earliest=-5m). Against a frozen dataset every rule settles into a permanent state within minutes and the Alerts page never moves again. - Recent views. A "last 15 minutes" dashboard, or a query for what just happened, is empty on a dataset that stopped growing overnight.
So the demo runs demo-simulator continuously, and reset-demo.sh only
handles the parts that genuinely are one-shot: the history behind the
present, and the dashboards and rules themselves.
Editing a dashboard
Change it in the web UI, hit Export JSON, and drop the file in
dashboards/ — the export shape and the import shape are the same one.
The next reset picks it up. (Note that panel IDs and the dashboard ID are
not part of that shape: every reset creates them fresh.)
What the queries can't do yet
Every panel here uses table, bar, top_n, single_stat, or
heatmap. None uses line, because a line chart needs a time axis and
the query language has no time-bucketing function — stats count by <field> groups by literal column values, so there's no equivalent of
Splunk's bin/timechart to group by hour or minute. Raw SQL could
express it (toStartOfHour(timestamp)), but dashboard panels reject the
SQL escape hatch by design, since the time-range picker works by
prepending earliest=/latest= terms to a pipe-syntax query
(see api/dashboards/types.go's validatePanel).
That gap is the one real thing standing between these dashboards and a conventional observability overview screen.