refactor(playbooks): host-generic GPU host + GPU LXC playbooks for nh3-ml1
- esh-pve-nvidia-host -> pve-nvidia-host: headers/dkms/build-essential step, nouveau blacklist + guarded unload (refuses if nouveau bound a device) - esh-ml1-lxc -> gpu-lxc: host vars have no defaults (elway aborts on undefined), rootfs storage/startup order parameterized, CT kept out of all-guests vzdump jobs - embed-rerank: Homepage labels take HOST_NAME/HOST_IP, defaults = esh-ml1
This commit is contained in:
@@ -5,7 +5,7 @@ Executes Prime's 2026-09-24 decision (LXC + host driver, not a VFIO VM). Done
|
||||
|
||||
**Host (esh-pve).** NVIDIA **580.178.04** open modules via DKMS from NVIDIA's
|
||||
`-no-compat32.run` (sha256 checked against NVIDIA's published sum), applied
|
||||
live by `playbooks/esh-pve-nvidia-host.yaml`: nouveau was never loaded and the
|
||||
live by `playbooks/pve-nvidia-host.yaml`: nouveau was never loaded and the
|
||||
T400's vfio ids matched nothing, so nothing held the card. Our own
|
||||
`nvidia-persistenced.service` creates `/dev/nvidia{0,ctl,-uvm,-uvm-tools}`
|
||||
**before `pve-guests`** (else `pct start` refuses the `devN:` entries). Retired
|
||||
@@ -15,7 +15,7 @@ rebuilt on purpose (boot path untouched). 580 because the fleet's vLLM
|
||||
not an apt repo: one file on both sides makes the host-module/container-lib
|
||||
version match true by construction; apt would let an upgrade move one side.
|
||||
|
||||
**CT 110 `esh-ml1`** (`playbooks/esh-ml1-lxc.yaml`): unprivileged Debian 12,
|
||||
**CT 110 `esh-ml1`** (`playbooks/gpu-lxc.yaml`): unprivileged Debian 12,
|
||||
nesting+keyctl, 6c/16G/80G `local-lvm`, `10.0.50.80` VLAN 50 (static; UDM pool
|
||||
is `.150–.250`), `startup order=30` after esh-scale/VMs. NVIDIA userspace from
|
||||
the same `.run` with `--no-kernel-modules`; docker-ce; nvidia-container-toolkit
|
||||
|
||||
@@ -167,7 +167,7 @@ Pre-flight done before shutdown:
|
||||
8. **pfi-gx10:** if Prime ran the AC-pull test on this visit, verify it came back
|
||||
by itself.
|
||||
9. **Next work:** the GPU's purpose is TBD with Prime; likely the second embed/rerank
|
||||
backend. Reuse `playbooks/esh-pve-nvidia-host.yaml` / `esh-ml1-lxc.yaml` (make
|
||||
backend. Reuse `playbooks/pve-nvidia-host.yaml` / `gpu-lxc.yaml` (make
|
||||
them host-generic). nh3-pve runs kernel 6.8.12-11 with **no matching
|
||||
proxmox-headers installed**, on PVE 8.4.1 (Debian 12 template only). If the LM
|
||||
port was cabled, MEBx provisioning can be done through the NanoKVM.
|
||||
|
||||
@@ -1,15 +1,25 @@
|
||||
# esh-ml1 — the GPU LXC on esh-pve that serves embedding + reranking on the
|
||||
# RTX 2000E Ada (16 GB). Runs against the HYPERVISOR and drives the container
|
||||
# through `pct`, so it works before the container has any users or keys.
|
||||
# A GPU LXC on a Proxmox host whose NVIDIA driver lives on the HYPERVISOR.
|
||||
# Host-generic since 2026-09-25 (was esh-ml1-lxc). Runs against the HYPERVISOR
|
||||
# and drives the container through `pct`, so it works before the container has
|
||||
# any users or keys.
|
||||
#
|
||||
# Prerequisite: playbooks/esh-pve-nvidia-host.yaml (host kernel module, device
|
||||
# nodes, persistence unit ordered before pve-guests).
|
||||
# CT host IP role
|
||||
# 110 esh-ml1 esh-pve 10.0.50.80 v50 fleet embed/rerank (TEI) + reward seat
|
||||
# 109 nh3-ml1 nh3-pve 10.100.50.80 v50 embed/rerank (TEI), second RTX 2000
|
||||
#
|
||||
# Prerequisite: playbooks/pve-nvidia-host.yaml on the same host (kernel module,
|
||||
# device nodes, persistence unit ordered before pve-guests).
|
||||
#
|
||||
# HOST-SPECIFIC VARS HAVE NO DEFAULTS, on purpose: elway refuses to run with an
|
||||
# undefined {{ var }}, so forgetting one aborts before anything is touched
|
||||
# instead of silently building esh-ml1's twin. Pass them all with --var (the
|
||||
# exact invocations are at the bottom of this header and in each server README).
|
||||
#
|
||||
# WHAT IT BUILDS
|
||||
# - CT {{ ctid }} `esh-ml1`, unprivileged, Debian 12 (the template the host's
|
||||
# own bookworm matches), nesting+keyctl for Docker, static 10.0.50.80 on
|
||||
# VLAN 50 (outside the UDM's .150–.250 DHCP pool), onboot AFTER esh-scale
|
||||
# and the two VMs so a GPU fault can never delay DNS or the mesh route.
|
||||
# - CT {{ ctid }}, unprivileged, Debian 12 (PVE < 8.4.20 rejects the Debian 13
|
||||
# template, and both hosts are bookworm), nesting+keyctl for Docker, static
|
||||
# IP outside the site's DHCP pool, onboot AFTER the site's core guests
|
||||
# (startup order 30) so a GPU fault can never delay DNS or the mesh route.
|
||||
# - /dev/nvidia0, nvidiactl, nvidia-uvm, nvidia-uvm-tools bound in via
|
||||
# `devN:` (PVE handles the cgroup rules and node creation).
|
||||
# - NVIDIA userspace from THE SAME .run the host installed, with
|
||||
@@ -22,25 +32,34 @@
|
||||
# LXC cannot write the device cgroup; the LXC's own devN rules already
|
||||
# confine the GPU).
|
||||
#
|
||||
# NOT BACKED UP, on purpose: esh-pve's vzdump job lists vmids explicitly and
|
||||
# this CT is not in it. Everything here is rebuilt by this playbook plus
|
||||
# stacks/embed-rerank; the models re-download from Hugging Face.
|
||||
# NOT BACKED UP, on purpose: everything here is rebuilt by this playbook plus
|
||||
# the stack, and the models re-download from Hugging Face. esh-pve's vzdump job
|
||||
# lists vmids explicitly, so it never included 110. nh3-pve's job is `all 1`,
|
||||
# so the playbook adds the CT to that job's `exclude` list.
|
||||
#
|
||||
# Run: scripts/elway root@esh-pve --playbook playbooks/esh-ml1-lxc.yaml
|
||||
# Run (esh-ml1):
|
||||
# scripts/elway root@esh-pve --playbook playbooks/gpu-lxc.yaml \
|
||||
# --var ctid=110 --var hostname=esh-ml1 --var ip_cidr=10.0.50.80/24 \
|
||||
# --var ip_addr=10.0.50.80 --var gateway=10.0.50.1 --var vlan=50 \
|
||||
# --var rootfs_storage=local-lvm \
|
||||
# --var "role=the fleet embedding + reranking service (TEI)"
|
||||
# Run (nh3-ml1):
|
||||
# scripts/elway root@nh3-pve.nh3.internal --playbook playbooks/gpu-lxc.yaml \
|
||||
# --var ctid=109 --var hostname=nh3-ml1 --var ip_cidr=10.100.50.80/24 \
|
||||
# --var ip_addr=10.100.50.80 --var gateway=10.100.50.1 --var vlan=50 \
|
||||
# --var rootfs_storage=local-zfs \
|
||||
# --var "role=second embed/rerank backend (TEI)"
|
||||
|
||||
vars:
|
||||
ctid: 110
|
||||
hostname: esh-ml1
|
||||
ip_cidr: 10.0.50.80/24
|
||||
ip_addr: 10.0.50.80
|
||||
gateway: 10.0.50.1
|
||||
vlan: 50
|
||||
# Host-specific, NO defaults (see header): ctid, hostname, ip_cidr, ip_addr,
|
||||
# gateway, vlan, rootfs_storage, role.
|
||||
cores: 6
|
||||
memory_mb: 16384
|
||||
swap_mb: 2048
|
||||
rootfs_gb: 80
|
||||
startup_order: 30
|
||||
template: debian-12-standard_12.12-1_amd64.tar.zst
|
||||
# Must equal playbooks/esh-pve-nvidia-host.yaml's driver_version.
|
||||
# Must equal playbooks/pve-nvidia-host.yaml's driver_version.
|
||||
driver_version: 580.178.04
|
||||
stage_dir: /root/nvidia
|
||||
infra_ops_pubkey: "ssh-ed25519 AAAAC3NzaC1lZDI1NTE5AAAAIN+1HBwfXrkfTYWdcnWCjLJ6VLAGC87gxH5h5vKaaA3c infra-ops@pfi-fleet"
|
||||
@@ -55,14 +74,33 @@ steps:
|
||||
pct create {{ ctid }} local:vztmpl/{{ template }} \
|
||||
--hostname {{ hostname }} --unprivileged 1 --features nesting=1,keyctl=1 \
|
||||
--cores {{ cores }} --memory {{ memory_mb }} --swap {{ swap_mb }} \
|
||||
--rootfs local-lvm:{{ rootfs_gb }} \
|
||||
--rootfs {{ rootfs_storage }}:{{ rootfs_gb }} \
|
||||
--net0 name=eth0,bridge=vmbr0,firewall=1,gw={{ gateway }},ip={{ ip_cidr }},tag={{ vlan }},type=veth \
|
||||
--onboot 1 --startup order=30,up=10 \
|
||||
--onboot 1 --startup order={{ startup_order }},up=10 \
|
||||
--dev0 /dev/nvidia0 --dev1 /dev/nvidiactl \
|
||||
--dev2 /dev/nvidia-uvm --dev3 /dev/nvidia-uvm-tools \
|
||||
--description "esh-ml1 — RTX 2000E Ada: the fleet embedding + reranking service (TEI). Built by eshpfi playbooks/esh-ml1-lxc.yaml; stack stacks/embed-rerank. NOT in the vzdump job on purpose (rebuildable)."
|
||||
--description "{{ hostname }} — RTX 2000E Ada: {{ role }}. Built by eshpfi playbooks/gpu-lxc.yaml; stack stacks/embed-rerank. NOT backed up on purpose (rebuildable)."
|
||||
when: "! pct status {{ ctid }} >/dev/null 2>&1"
|
||||
|
||||
# A job with `all 1` would pick the new CT up tonight. Add it to that job's
|
||||
# exclude list; jobs that name their vmids (esh-pve's) are left alone.
|
||||
- name: Keep CT {{ ctid }} out of every all-guests vzdump job
|
||||
shell: |
|
||||
set -e
|
||||
CT={{ ctid }} JOBS="$(pvesh get /cluster/backup --output-format json)" python3 - <<'PY'
|
||||
import json, os, subprocess
|
||||
ct = os.environ["CT"]
|
||||
for j in json.loads(os.environ["JOBS"]):
|
||||
if str(j.get("all", 0)) != "1":
|
||||
continue
|
||||
ex = [x for x in str(j.get("exclude", "")).split(",") if x]
|
||||
if ct in ex:
|
||||
continue
|
||||
subprocess.run(["pvesh", "set", "/cluster/backup/" + j["id"], "--exclude", ",".join(ex + [ct])], check=True)
|
||||
print("excluded", ct, "from", j["id"])
|
||||
PY
|
||||
when: "awk -v ct={{ ctid }} 'function fl(){ if(isv&&all&&!hit) need=1 } /^[^ \\t]/{fl(); isv=($1==\"vzdump:\"); all=0; hit=0; next} isv&&$1==\"all\"&&$2==\"1\"{all=1} isv&&$1==\"exclude\"{n=split($2,a,\",\"); for(i=1;i<=n;i++) if(a[i]==ct) hit=1} END{fl(); exit !need}' /etc/pve/jobs.cfg"
|
||||
|
||||
- name: Start the container
|
||||
shell: pct start {{ ctid }} && sleep 5
|
||||
when: "! pct status {{ ctid }} | grep -q running"
|
||||
@@ -168,7 +206,7 @@ steps:
|
||||
set -euo pipefail
|
||||
install -d /etc/systemd/system/docker.service.d
|
||||
cat > /etc/systemd/system/docker.service.d/override.conf <<'EOC'
|
||||
# Homepage discovery — see eshpfi playbooks/esh-ml1-lxc.yaml
|
||||
# Homepage discovery — see eshpfi playbooks/gpu-lxc.yaml
|
||||
[Service]
|
||||
ExecStart=
|
||||
ExecStart=/usr/bin/dockerd -H fd:// -H tcp://{{ ip_addr }}:2375 --containerd=/run/containerd/containerd.sock
|
||||
@@ -180,7 +218,7 @@ steps:
|
||||
|
||||
verify:
|
||||
- name: Container is running with onboot set, started after the core guests
|
||||
shell: "pct status {{ ctid }} | grep -q running && pct config {{ ctid }} | grep -q '^onboot: 1' && pct config {{ ctid }} | grep -q '^startup: order=30'"
|
||||
shell: "pct status {{ ctid }} | grep -q running && pct config {{ ctid }} | grep -q '^onboot: 1' && pct config {{ ctid }} | grep -q '^startup: order={{ startup_order }}'"
|
||||
changed_when: "false"
|
||||
|
||||
- name: nvidia-smi inside the CT reports the host's driver version
|
||||
@@ -195,3 +233,7 @@ verify:
|
||||
shell: |
|
||||
pct exec {{ ctid }} -- sh -c 'test "$(id -u infra-ops)" = 850 && test "$(id -u vh)" = 1000 && test "$(getent group docker | cut -d: -f3)" = 851 && id -nG infra-ops | grep -qw docker'
|
||||
changed_when: "false"
|
||||
|
||||
- name: CT {{ ctid }} is in no all-guests vzdump job
|
||||
shell: "! awk -v ct={{ ctid }} 'function fl(){ if(isv&&all&&!hit) need=1 } /^[^ \\t]/{fl(); isv=($1==\"vzdump:\"); all=0; hit=0; next} isv&&$1==\"all\"&&$2==\"1\"{all=1} isv&&$1==\"exclude\"{n=split($2,a,\",\"); for(i=1;i<=n;i++) if(a[i]==ct) hit=1} END{fl(); exit !need}' /etc/pve/jobs.cfg"
|
||||
changed_when: "false"
|
||||
@@ -1,14 +1,18 @@
|
||||
# esh-pve — NVIDIA driver on the HYPERVISOR, for the RTX 2000 Ada that an LXC
|
||||
# (esh-ml1, embedding + reranking) uses.
|
||||
# NVIDIA driver on a Proxmox HYPERVISOR, for a GPU that an LXC uses through the
|
||||
# host's kernel module. Host-generic since 2026-09-25 (was esh-pve-nvidia-host).
|
||||
#
|
||||
# host GPU LXC run
|
||||
# esh-pve RTX 2000E Ada (01:00.0) esh-ml1 (CT 110) 2026-09-24
|
||||
# nh3-pve RTX 2000E Ada (01:00.0) nh3-ml1 (CT 109) 2026-09-25
|
||||
#
|
||||
# WHY A HOST DRIVER: Prime's decision 2026-09-24 — LXC + host driver, NOT a
|
||||
# VFIO VM. Both esh-pve hangs this year (the 2026-08-19 hard freeze and the
|
||||
# 2026-09-10 vzdump that left `kvm -id 102` in D state for 14 days) came from
|
||||
# VFIO passthrough on this box. An LXC shares the host kernel's driver, so no
|
||||
# VFIO passthrough on that box. An LXC shares the host kernel's driver, so no
|
||||
# passthrough, no pinned VM RAM. Accepted cost: a driver on the hypervisor,
|
||||
# rebuilt by DKMS on every PVE kernel update, and the LXC's userspace libraries
|
||||
# must be the SAME version as this kernel module (see
|
||||
# playbooks/esh-ml1-lxc.yaml, which installs from the same .run file).
|
||||
# must be the SAME version as this kernel module (see playbooks/gpu-lxc.yaml,
|
||||
# which installs from the same .run file).
|
||||
#
|
||||
# WHY THE .run FILE, NOT AN APT REPO: the LXC must match the host module
|
||||
# exactly. One file, installed with the kernel module here and with
|
||||
@@ -19,26 +23,45 @@
|
||||
# WHY 580: the fleet's vLLM image (vllm/vllm-openai:v0.24.0) is a CUDA 13.0
|
||||
# build and needs a >= 580 driver. fv-ml1 runs 580.65.06; this is the newest
|
||||
# point release on the same branch. `-no-compat32` because a headless compute
|
||||
# box has no use for 32-bit libraries.
|
||||
# box has no use for 32-bit libraries. ONE version fleet-wide for the RTX 2000
|
||||
# hosts, so a container image that works on one works on the other.
|
||||
#
|
||||
# NO REBOOT NEEDED: nouveau is blacklisted and was never loaded, and the T400's
|
||||
# vfio-pci ids no longer match any device, so nothing holds the card. The
|
||||
# module loads live. (esh-pve is ESH's only DNS and mesh route; a reboot takes
|
||||
# the whole site dark from outside — avoid one.)
|
||||
# NO REBOOT NEEDED, on either host, and each PVE host is its site's DNS/mesh
|
||||
# path, so avoid one. esh-pve: nouveau was blacklisted and never loaded, and the
|
||||
# T400's vfio-pci ids matched nothing. nh3-pve: nouveau auto-loaded at boot but
|
||||
# FAILED to bind the Ada card (`gsp ctor failed: -2`, no GSP firmware), so the
|
||||
# module holds nothing and unloads cleanly. The unload step refuses if nouveau
|
||||
# has actually bound a device; then the blacklist plus a reboot is the only way.
|
||||
#
|
||||
# Upgrading the driver: bump driver_version + driver_sha256 here AND in
|
||||
# playbooks/esh-ml1-lxc.yaml, run this, then that. Until both have run, CUDA
|
||||
# in the container fails with "driver/library version mismatch".
|
||||
# Upgrading the driver: bump driver_version + driver_sha256 here AND
|
||||
# driver_version in playbooks/gpu-lxc.yaml, run this on every GPU host, then
|
||||
# gpu-lxc.yaml for every GPU LXC. Until both have run, CUDA in the container
|
||||
# fails with "driver/library version mismatch".
|
||||
#
|
||||
# Run: scripts/elway root@esh-pve --playbook playbooks/esh-pve-nvidia-host.yaml
|
||||
# Run: scripts/elway root@esh-pve --playbook playbooks/pve-nvidia-host.yaml
|
||||
# scripts/elway root@nh3-pve.nh3.internal --playbook playbooks/pve-nvidia-host.yaml
|
||||
|
||||
vars:
|
||||
driver_version: 580.178.04
|
||||
# From download.nvidia.com/XFree86/Linux-x86_64/580.178.04/*.run.sha256sum
|
||||
driver_sha256: 1833c8c5c53481dfab46df442c3de26ec5d4c35084e03d94ad93746d29899750
|
||||
stage_dir: /root/nvidia
|
||||
# What `nvidia-smi --query-gpu=name` must contain on this host.
|
||||
gpu_match: RTX 2000
|
||||
|
||||
steps:
|
||||
# DKMS needs the headers for the RUNNING kernel (built now) and the series
|
||||
# meta-package (so each future kernel's headers arrive with it and DKMS
|
||||
# rebuilds on upgrade). nh3-pve had neither, nor dkms or a compiler.
|
||||
- name: Kernel headers (running kernel + series meta), dkms, build tools
|
||||
shell: |
|
||||
set -e
|
||||
export DEBIAN_FRONTEND=noninteractive
|
||||
k=$(uname -r); series=$(echo "$k" | cut -d. -f1,2)
|
||||
apt-get update -qq
|
||||
apt-get install -y -qq "proxmox-headers-$k" "proxmox-headers-$series" dkms build-essential
|
||||
when: "! dpkg -s proxmox-headers-$(uname -r) proxmox-headers-$(uname -r | cut -d. -f1,2) dkms build-essential >/dev/null 2>&1"
|
||||
|
||||
- name: Stage the driver .run and check it against NVIDIA's published sha256
|
||||
shell: |
|
||||
set -e
|
||||
@@ -53,27 +76,45 @@ steps:
|
||||
fi
|
||||
when: "! echo '{{ driver_sha256 }} {{ stage_dir }}/NVIDIA-Linux-x86_64-{{ driver_version }}-no-compat32.run' | sha256sum -c --status 2>/dev/null"
|
||||
|
||||
# The T400 was passed to VM 102 through vfio-pci by id. The card and the VM
|
||||
# are both gone; the ids match nothing. Kept (not deleted) so the retirement
|
||||
# is readable later.
|
||||
# esh-pve only: the T400 was passed to VM 102 through vfio-pci by id. The card
|
||||
# and the VM are both gone; the ids match nothing. Kept (not deleted) so the
|
||||
# retirement is readable later.
|
||||
- name: Retire the T400's vfio-pci id binding
|
||||
shell: mv /etc/modprobe.d/vfio.conf {{ stage_dir }}/vfio.conf.retired-2026-09-24
|
||||
when: "grep -q '10de:1ff2' /etc/modprobe.d/vfio.conf 2>/dev/null"
|
||||
|
||||
# blacklist.conf blacklisted `nvidia` alongside nouveau (VM-passthrough era,
|
||||
# when the host must never touch the card). nouveau stays blacklisted.
|
||||
# esh-pve only: blacklist.conf blacklisted `nvidia` alongside nouveau
|
||||
# (VM-passthrough era, when the host must never touch the card). nouveau
|
||||
# stays blacklisted.
|
||||
- name: Stop blacklisting the nvidia module (nouveau stays blacklisted)
|
||||
shell: |
|
||||
cp -n /etc/modprobe.d/blacklist.conf {{ stage_dir }}/blacklist.conf.pre-nvidia
|
||||
sed -i '/^blacklist nvidia$/d' /etc/modprobe.d/blacklist.conf
|
||||
when: "grep -qx 'blacklist nvidia' /etc/modprobe.d/blacklist.conf"
|
||||
when: "grep -qx 'blacklist nvidia' /etc/modprobe.d/blacklist.conf 2>/dev/null"
|
||||
|
||||
# nouveau is not in PVE's initramfs (checked on nh3-pve: 0 files), so the
|
||||
# rootfs modprobe.d is where udev's autoload reads the blacklist. No
|
||||
# initramfs rebuild or ESP sync needed.
|
||||
- name: Blacklist nouveau
|
||||
shell: |
|
||||
printf '# nvidia owns the GPU (the LXC uses it) — see playbooks/pve-nvidia-host.yaml\nblacklist nouveau\noptions nouveau modeset=0\n' \
|
||||
> /etc/modprobe.d/blacklist-nouveau.conf
|
||||
when: "! grep -rqx 'blacklist nouveau' /etc/modprobe.d/"
|
||||
|
||||
- name: Unload nouveau (only when it holds no device)
|
||||
shell: |
|
||||
set -e
|
||||
if ls /sys/bus/pci/drivers/nouveau/ 2>/dev/null | grep -q '^0000:'; then
|
||||
echo "nouveau is BOUND to a device: it is blacklisted now, but only a reboot releases it" >&2
|
||||
exit 1
|
||||
fi
|
||||
modprobe -r nouveau
|
||||
when: "grep -q '^nouveau ' /proc/modules"
|
||||
|
||||
- name: Install the driver with DKMS (open kernel modules)
|
||||
# --no-opengl-files: headless compute. --kernel-module-type=open: Ada
|
||||
# (Turing+) is supported by the open modules and fv-ml1 runs them too.
|
||||
# DKMS rebuilds the module whenever a new proxmox-headers package lands;
|
||||
# `proxmox-headers-6.8` (the meta-package) is installed so new headers
|
||||
# arrive with each new kernel.
|
||||
# DKMS rebuilds the module whenever a new proxmox-headers package lands.
|
||||
shell: |
|
||||
sh {{ stage_dir }}/NVIDIA-Linux-x86_64-{{ driver_version }}-no-compat32.run \
|
||||
--dkms --silent --no-x-check --no-opengl-files --kernel-module-type=open \
|
||||
@@ -82,7 +123,7 @@ steps:
|
||||
|
||||
- name: Load nvidia + nvidia_uvm at every boot
|
||||
shell: |
|
||||
printf '# RTX 2000 Ada for the esh-ml1 LXC — see playbooks/esh-pve-nvidia-host.yaml\nnvidia\nnvidia_uvm\n' \
|
||||
printf '# GPU for the host GPU LXC — see playbooks/pve-nvidia-host.yaml\nnvidia\nnvidia_uvm\n' \
|
||||
> /etc/modules-load.d/nvidia.conf
|
||||
when: "! grep -qx 'nvidia_uvm' /etc/modules-load.d/nvidia.conf 2>/dev/null"
|
||||
|
||||
@@ -95,7 +136,7 @@ steps:
|
||||
- name: Install the GPU-init + persistence unit, ordered before pve-guests
|
||||
shell: |
|
||||
cat > /etc/systemd/system/nvidia-persistenced.service <<'EOF'
|
||||
# See playbooks/esh-pve-nvidia-host.yaml
|
||||
# See playbooks/pve-nvidia-host.yaml
|
||||
[Unit]
|
||||
Description=NVIDIA persistence daemon + device nodes for the GPU LXC
|
||||
Before=pve-guests.service
|
||||
@@ -116,8 +157,8 @@ steps:
|
||||
when: "! systemctl is-active --quiet nvidia-persistenced.service || ! grep -q 'Before=pve-guests.service' /etc/systemd/system/nvidia-persistenced.service"
|
||||
|
||||
verify:
|
||||
- name: nvidia-smi sees the RTX 2000 Ada at the pinned driver version
|
||||
shell: nvidia-smi --query-gpu=name,driver_version --format=csv,noheader | grep -q 'RTX 2000.*, {{ driver_version }}$'
|
||||
- name: nvidia-smi sees the {{ gpu_match }} at the pinned driver version
|
||||
shell: nvidia-smi --query-gpu=name,driver_version --format=csv,noheader | grep -q '{{ gpu_match }}.*, {{ driver_version }}$'
|
||||
changed_when: "false"
|
||||
|
||||
- name: Kernel module in use is the DKMS-built {{ driver_version }}
|
||||
@@ -132,6 +173,10 @@ verify:
|
||||
shell: nvidia-smi --query-gpu=persistence_mode --format=csv,noheader | grep -qx Enabled
|
||||
changed_when: "false"
|
||||
|
||||
- name: Survives a reboot (modules-load + unit enabled before pve-guests)
|
||||
shell: grep -qx nvidia_uvm /etc/modules-load.d/nvidia.conf && systemctl is-enabled --quiet nvidia-persistenced.service && ! grep -qx 'blacklist nvidia' /etc/modprobe.d/blacklist.conf
|
||||
- name: Survives a reboot (modules-load + unit enabled + nouveau blacklisted)
|
||||
shell: grep -qx nvidia_uvm /etc/modules-load.d/nvidia.conf && systemctl is-enabled --quiet nvidia-persistenced.service && ! grep -qx 'blacklist nvidia' /etc/modprobe.d/blacklist.conf 2>/dev/null && grep -rqx 'blacklist nouveau' /etc/modprobe.d/
|
||||
changed_when: "false"
|
||||
|
||||
- name: DKMS will rebuild for the next kernel (series headers meta installed)
|
||||
shell: dpkg -s proxmox-headers-$(uname -r | cut -d. -f1,2) >/dev/null 2>&1
|
||||
changed_when: "false"
|
||||
@@ -111,14 +111,17 @@ free.
|
||||
|
||||
## How it is built
|
||||
|
||||
1. [`playbooks/esh-pve-nvidia-host.yaml`](../../playbooks/esh-pve-nvidia-host.yaml)
|
||||
1. [`playbooks/pve-nvidia-host.yaml`](../../playbooks/pve-nvidia-host.yaml)
|
||||
— driver **580.178.04** (open modules, DKMS) on the **hypervisor**, the
|
||||
`nvidia-persistenced` unit that creates the device nodes before
|
||||
`pve-guests`, and removal of the old VFIO/blacklist config.
|
||||
2. [`playbooks/esh-ml1-lxc.yaml`](../../playbooks/esh-ml1-lxc.yaml) — the CT,
|
||||
2. [`playbooks/gpu-lxc.yaml`](../../playbooks/gpu-lxc.yaml) — the CT,
|
||||
`dev0–3` GPU nodes, the NVIDIA userspace from the **same `.run`** with
|
||||
`--no-kernel-modules`, fleet ids (infra-ops 850, docker 851, vh 1000),
|
||||
docker-ce and nvidia-container-toolkit with `no-cgroups = true`.
|
||||
Host-generic since 2026-09-25 (it also builds nh3-ml1). Its host vars have
|
||||
**no defaults**, so pass esh-ml1's set: the exact `--var` line is in the
|
||||
playbook header under "Run (esh-ml1)".
|
||||
3. `scripts/deploy-stack.sh esh-ml1 embed-rerank`, then `docker compose up -d`.
|
||||
|
||||
Both playbooks are idempotent; re-run them to repair.
|
||||
@@ -142,7 +145,7 @@ build and this CT will fail to start at the next boot.
|
||||
|---|---|---|
|
||||
| **Beszel** (host + GPU telemetry) | NVIDIA agent `henrygd/beszel-agent-nvidia:0.18.7`, `stacks/beszel` + `hosts/esh-ml1.yaml`. Hub system `ridkfdpwfq3f730`. GPU util, VRAM, power and temperature are sampled. | alerts → infra-ops: Status down 2 m, Disk >85% 5 m, CPU >95% 15 m, Memory >90% 10 m, Temperature >85 °C 5 m (GPU included) |
|
||||
| **Uptime Kuma** (service) | `Embed — Qwen3 0.6B (TEI, esh-ml1)` → `:8001/health` (#27); `Rerank — bge-v2-m3 (TEI, esh-ml1)` → `:8013/health` (#28). TEI's health runs the backend. | alerts → infra-ops via althing-alert-bridge; `stacks/uptimekuma/monitors.yaml` |
|
||||
| **Homepage** | three cards under *AI - Eval & Retrieval*. dockerd exposes tcp/2375 on `10.0.50.80` only (fleet norm, `playbooks/esh-ml1-lxc.yaml`). | `stacks/homepage/conf/docker.yaml` → `esh-ml1-docker` |
|
||||
| **Homepage** | three cards under *AI - Eval & Retrieval*. dockerd exposes tcp/2375 on `10.0.50.80` only (fleet norm, `playbooks/gpu-lxc.yaml`). | `stacks/homepage/conf/docker.yaml` → `esh-ml1-docker` |
|
||||
| **Dozzle** (logs) | agent `v10.4.1` on `10.0.50.80:7007`, compose dir `dozzle-agent` | hub on ana-docker :8088 |
|
||||
|
||||
The reward seat has **no Kuma check on purpose**: seats are outside Kuma's lane,
|
||||
|
||||
@@ -59,7 +59,7 @@ and reranking ([`servers/esh-ml1/README.md`](../esh-ml1/README.md)).
|
||||
|
||||
- **Driver 580.178.04, open kernel modules, DKMS**, from NVIDIA's `.run`
|
||||
(`/root/nvidia/`). Applied by
|
||||
[`playbooks/esh-pve-nvidia-host.yaml`](../../playbooks/esh-pve-nvidia-host.yaml)
|
||||
[`playbooks/pve-nvidia-host.yaml`](../../playbooks/pve-nvidia-host.yaml)
|
||||
**live, with no reboot**: nouveau was never loaded and nothing held the card.
|
||||
- **`nvidia-persistenced.service`** (ours, in `/etc/systemd/system`) runs
|
||||
`nvidia-modprobe -c0 -u` and the persistence daemon **before
|
||||
|
||||
@@ -1,6 +1,6 @@
|
||||
# esh-ml1 (CT 110 on esh-pve) — Beszel agent with NVIDIA GPU telemetry for the
|
||||
# RTX 2000E Ada (utilization, VRAM, temperature, power). Docker-in-LXC: the
|
||||
# NVIDIA container toolkit runs with no-cgroups=true (playbooks/esh-ml1-lxc.yaml).
|
||||
# NVIDIA container toolkit runs with no-cgroups=true (playbooks/gpu-lxc.yaml).
|
||||
# No extra filesystems: the root filesystem holds everything, models included.
|
||||
services:
|
||||
beszel-agent:
|
||||
|
||||
@@ -1,4 +1,10 @@
|
||||
# embed-rerank tunables (esh-ml1, TEI). Copy to `.env` on the server.
|
||||
# embed-rerank tunables (TEI). Copy to `.env` on the server.
|
||||
|
||||
# Per-host Homepage labels. Unset = esh-ml1's values. nh3-ml1 sets:
|
||||
# HOST_NAME=nh3-ml1
|
||||
# HOST_IP=10.100.50.80
|
||||
#HOST_NAME=esh-ml1
|
||||
#HOST_IP=10.0.50.80
|
||||
|
||||
# TEI image tag. `89-` = the Ada Lovelace (sm_89) build; a different GPU
|
||||
# generation needs a different prefix (see the TEI README's image table).
|
||||
|
||||
@@ -1,6 +1,8 @@
|
||||
# embed-rerank — THE fleet's embedding + reranking service, on esh-ml1 (CT 110 on
|
||||
# esh-pve, RTX 2000E Ada, 16 GB). Served by Hugging Face Text Embeddings
|
||||
# Inference (TEI).
|
||||
# Inference (TEI). A second instance runs on nh3-ml1 (CT 109 on nh3-pve, the same
|
||||
# card) since 2026-09-25; the per-host bits are HOST_NAME / HOST_IP in .env, and
|
||||
# their defaults are esh-ml1's, so esh-ml1's live .env needs no change.
|
||||
#
|
||||
# Prime, 2026-09-25: "TEI is embed/reranker server for esh-ml1 and the FLEET in
|
||||
# general, in future." It replaced vLLM here the same day, after a side-by-side
|
||||
@@ -30,7 +32,7 @@
|
||||
# ports, and in practice only the LiteLLM gateway does (verified from the seats'
|
||||
# logs 2026-09-25: every request matched a gateway spend-log row).
|
||||
#
|
||||
# Host setup: playbooks/esh-pve-nvidia-host.yaml, then playbooks/esh-ml1-lxc.yaml.
|
||||
# Host setup: playbooks/pve-nvidia-host.yaml, then playbooks/gpu-lxc.yaml.
|
||||
# Tunables live in .env.
|
||||
|
||||
name: embed-rerank
|
||||
@@ -76,10 +78,10 @@ services:
|
||||
start_period: 60s
|
||||
labels:
|
||||
- homepage.group=AI - Eval & Retrieval
|
||||
- homepage.name=Embed — Qwen3 0.6B (TEI, esh-ml1)
|
||||
- homepage.name=Embed — Qwen3 0.6B (TEI, ${HOST_NAME:-esh-ml1})
|
||||
- homepage.icon=mdi-vector-arrange-below
|
||||
- homepage.description=Fleet embeddings (qwen3-embedding) via TEI on esh-ml1
|
||||
- homepage.href=http://10.0.50.80:${EMBED_PORT}/docs
|
||||
- homepage.description=Fleet embeddings (qwen3-embedding) via TEI on ${HOST_NAME:-esh-ml1}
|
||||
- homepage.href=http://${HOST_IP:-10.0.50.80}:${EMBED_PORT}/docs
|
||||
|
||||
tei-rerank:
|
||||
image: ghcr.io/huggingface/text-embeddings-inference:${TEI_TAG}
|
||||
@@ -117,7 +119,7 @@ services:
|
||||
start_period: 60s
|
||||
labels:
|
||||
- homepage.group=AI - Eval & Retrieval
|
||||
- homepage.name=Rerank — bge-v2-m3 (TEI, esh-ml1)
|
||||
- homepage.name=Rerank — bge-v2-m3 (TEI, ${HOST_NAME:-esh-ml1})
|
||||
- homepage.icon=mdi-sort-variant
|
||||
- homepage.description=Fleet reranker (reranker) via TEI on esh-ml1
|
||||
- homepage.href=http://10.0.50.80:${RERANK_PORT}/docs
|
||||
- homepage.description=Fleet reranker (reranker) via TEI on ${HOST_NAME:-esh-ml1}
|
||||
- homepage.href=http://${HOST_IP:-10.0.50.80}:${RERANK_PORT}/docs
|
||||
|
||||
@@ -32,7 +32,7 @@ irv-ml1-docker:
|
||||
port: 2375
|
||||
|
||||
# esh-ml1 — CT 110 on esh-pve: the fleet embed/rerank (TEI) + reward seats.
|
||||
# dockerd listens only on its own address (playbooks/esh-ml1-lxc.yaml), 2026-09-25.
|
||||
# dockerd listens only on its own address (playbooks/gpu-lxc.yaml), 2026-09-25.
|
||||
esh-ml1-docker:
|
||||
host: 10.0.50.80
|
||||
port: 2375
|
||||
|
||||
Reference in New Issue
Block a user