Skip to content

Build the lab Windows domain on Saruman (ADR-0029) #414

Closed

Description

@Gerrrt

ADR-0029 sized the lab's Windows domain and gave it a namespace and a clock;
build-the-lab-domain.md is the
procedure. #265 closed on the decision and the runbook, and the build lost its
tracker — the roadmap still says "decided, not built", and two issues wait
behind it. This is the tracker.

What is decided

Six guests on Saruman, the .50 decade of ImaginationLAN: bahamut and
leviathan (domain controllers), titan and ramuh (members), carbuncle
and siren (endpoints). Servers continuous, endpoints per session. Namespace
ad.matrix.elysium / NetBIOS AD, clock from the gateway, DNS inward only,
domain scraped by the lab's Prometheus and never published. Hicks reaches it
by the rule ADR-0031 keeps open, and build-the-lab-domain.md §10 records
what a workstation then reaches on VLAN 30 — the list #228's narrowing waits
for.

What waits on it

  • Run Wazuh on Saruman — the heaviest component in ADR-0007 #266 Wazuh and Run Velociraptor on Saruman #267 Velociraptor: ADR-0030 puts both in stacks/soc/ on a
    second guest, odin at 10.0.30.60, and both are agentless without this
    domain — "an agentless Wazuh still has nothing to say".
  • ADR-0014's range: an attack VM pointed at an estate with no domain teaches
    nothing (build-the-playground.md §0).
  • The three domain rules already live in stacks/lab/prometheus/rules/lab.rules.yaml
    with promtool tests, and the commented-out windows scrape job in the lab's
    prometheus.yaml waits to be uncommented once six machines answer — the
    file says why it landed commented.

Order of work

The runbook's, §1 through §11, at Saruman from a Hicks workstation: create
the six VMs, install and address, promote bahamut with the clock set before
anything joins, the second DC, join the members and build the tiers, the
authentication generator, windows_exporter and the licence clock, turn on
the scrape, verify, record what Hicks now reaches, write it down. No purchase;
evenings.

stacks/soc/ can be authored and CI-validated before this is built — the
stacks/lab precedent — and then waits only on this.


Corrected 2026-09-19

A gate this issue did not name. ADR-0029's sizing — six guests, servers continuous, endpoints per session — rests on ~90 derived write IOPS that nobody has measured. #418 closed with the SSDs fitted and unassigned: no logical drive, no thin pool, alexander still on the HDD mirror, no fio. roadmap.md: "an unassigned drive changes no duty cycle either — the sentence above holds until the runbook's step 8 measures both arrays." That work is #527, and it precedes this.

Also worth settling before starting, because #448 raises it: whether this domain is built by hand from the runbook's eleven sections and later rebuilt from the pipeline, or built from the pipeline the first time. This issue stays the tracker either way; the order of work above assumes the first.


Corrected 2026-09-20

The gate named yesterday has cleared: #527 closed on 2026-09-20 — logical drive 2, both arrays measured at 4 KiB QD1 (the 7.2K mirror 741 IOPS, the SSD mirror 7,952), alexander moved to large_data and rebooted from flash. ADR-0029's sizing now rests on a measured number, and it is two orders of magnitude above the ~90 it assumed for the spindles. Nothing precedes this any more. The order of work above stands; the build-by-hand-then-rebuild question with #448 is still open and still this issue's to settle before starting.

Activity

  1. added 2 commits that reference this issue on Sep 17, 2026
  2. added
    labstacks/lab on alexander (VLAN 30)
    on Sep 19, 2026
  3. Gerrrt commented on Sep 24, 2026

    @Gerrrt
    OwnerAuthor

    Decided 2026-09-23: by hand first, then rebuilt from the pipeline

    This settles the question left open on 2026-09-19 and 2026-09-20. The domain is built by hand from build-the-lab-domain.md §1–§11, and later rebuilt from #448's pipeline (with #440's templates and #445's provisioning).

    Why this order:

    Nothing precedes the build. #527 cleared the IOPS gate, and Saruman read today with about 109 GB of memory available, 48 threads, and large_data at 8 % of 876 GiB. The two endpoints wait on the buy table's Windows 11 Pro keys; the four servers do not.

    🤖 Generated with Claude Code

  4. Gerrrt commented on Sep 26, 2026

    @Gerrrt
    OwnerAuthor

    2026-09-26: both Windows 11 Pro keys were bought on 2026-09-25 and are activated on carbuncle and siren. Their DHCP reservations (10.0.30.54/.55) are on morpheus. The endpoints aren't joined yet, because the DCs aren't built. The roadmap buy table is now empty (#665).

  5. Gerrrt commented on Sep 26, 2026

    @Gerrrt
    OwnerAuthor

    Built by hand 2026-09-24 to 2026-09-25 — steps 1–8 done, verified on the running domain

    Built by hand from build-the-lab-domain.md, the order decided on 2026-09-23 (by hand first, then #448 rebuilds). All six guests exist on Saruman, joined to ad.matrix.elysium / NetBIOS AD.

    Guest VMID Addr OU Role Windows Eval / licence
    bahamut 150 10.0.30.50 (static) Domain Controllers DC, PDC Server 2025 Std (Desktop) 178 d (~2027-03-22)
    leviathan 151 10.0.30.51 (static) Domain Controllers DC Server 2025 Std (Desktop) 178 d
    titan 152 10.0.30.52 (DHCP resv) Servers member Server 2025 Std (Desktop) 179 d
    ramuh 153 10.0.30.53 (DHCP resv) Servers member Server 2025 Std (Desktop) 179 d
    carbuncle 154 10.0.30.54 (DHCP resv) Workstations endpoint Windows 11 Pro activated
    siren 155 10.0.30.55 (DHCP resv) Workstations endpoint Windows 11 Pro activated

    The two endpoints used the two Windows 11 Pro keys (retail), which leave the buy table.

    What was done (runbook §1–§8)

    • §1–2: six VMs on large_data (ssd=1), firewall=0 per NIC, joined; DCs static, the other four on morpheus DHCP reservations; DNS at the two DCs; IPv6/LLMNR/NetBIOS/WPAD left at defaults (§0's "do not harden").
    • §3–4: bahamut promoted with the clock set from 10.0.30.1 before anything joined (it reads 5.5 ms off morpheus); leviathan the second DC; replication clean (repadmin /replsummary 0 fails); root hints removed on both; each DC's own DNS set partner-first-then-loopback.
    • §5: OUs (Tier0/1/2, Servers, Workstations); one admin per tier; Tier 0 Admins in Domain Admins; the Tier 0 deny-logon GPO linked to Servers + Workstations (the control a violation becomes an event against); titan shares with a decoy. The exploitable contents (users, roastable/delegation weaknesses) are left to Give the lab domain users and deliberate weaknesses, one tag at a time #449 by design — nothing observes them yet.
    • §7: windows_exporter v0.31.8 on all six, firewalled to alexander (10.0.30.40); licence-clock task on the four servers.
    • §8: the windows scrape turned on; all six up=1 on the lab Prometheus. Clock check healthy: bahamut=NTP, leviathan=NT5DS, neither on Local CMOS.

    Repo caught up with reality

    What remains to close this

    Flagged for a decision

    🤖 Generated with Claude Code

  6. Gerrrt commented on Sep 26, 2026

    @Gerrrt
    OwnerAuthor

    2026-09-26: a dependency check. §6 needs one ordinary domain user for a 15-minute scheduled task (runbook §6), but it's recorded here as waiting on #449's population. #449 waits on this issue and on #448, which is the end of the automation chain (#440 → #445 → #448). As written, this issue can't close until milestone 5 is nearly done, and #449 waits on it: a cycle.

    Two ways out. Make a single ordinary user by hand for §6's task now and leave #449 the population and the weakness tags. Or record §6 as deferred to #448's rebuild and close this on §10 and §11.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

No one assigned

    Labels

    featureA new capability or a changelabstacks/lab on alexander (VLAN 30)

    Projects

    No projects

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions