964 lines
51 KiB
Text
964 lines
51 KiB
Text
|
use domain/workspace/mod.nu *
|
||
|
|
use platform/user/config.nu [get-active-workspace-details list-workspaces]
|
||
|
|
use tools/nickel/process.nu [ncl-eval, ncl-eval-soft, default-ncl-paths]
|
||
|
|
use domain/workspace/state.nu [state-read state-write state-node-get state-node-set state-node-start state-node-finish state-node-delete state-node-reset log-trim]
|
||
|
|
use platform/clients/orchestrator.nu [orch-url, orch-submit-component, orch-wait-task, orch-health]
|
||
|
|
use orchestration/taskservs/run.nu [run_taskserv]
|
||
|
|
use platform/config/accessor.nu [get-taskservs-path]
|
||
|
|
use domain/utils/templates.nu [on_template_path run_from_template]
|
||
|
|
use primitives/io/service_check.nu [health-gate]
|
||
|
|
use primitives/io/logging.nu [is-debug-enabled is-debug-level]
|
||
|
|
use platform/integrations/context_assembler.nu [build-cabling-credentials-env]
|
||
|
|
|
||
|
|
|
||
|
|
# Cluster-observability slice extracted from components.nu (ADR-026 layer hygiene).
|
||
|
|
use cli/components.nu [cli-err comp-ncl-export comp-ws-path comp-ws-infra comp-server-ssh-info comp-ssh-run comp-state-server comp-resolve-ws-path comp-dag-summary comp-kubectl-ssh comp-kubectl-access]
|
||
|
|
|
||
|
|
# Structured compute capacity for a Hetzner server type code: parsed CPU and RAM
|
||
|
|
# for capacity math (cluster top, the placement simulator) plus a human label.
|
||
|
|
# Unknown codes yield zero capacity and the code itself as label, so callers can
|
||
|
|
# degrade gracefully instead of mis-reporting headroom.
|
||
|
|
export def comp-server-capacity [type_code: string]: nothing -> record {
|
||
|
|
let tbl = {
|
||
|
|
cax11: { cpu: 2, mem_gb: 4, arch: "ARM" }, cax21: { cpu: 4, mem_gb: 8, arch: "ARM" },
|
||
|
|
cax31: { cpu: 8, mem_gb: 16, arch: "ARM" }, cax41: { cpu: 16, mem_gb: 32, arch: "ARM" },
|
||
|
|
cpx11: { cpu: 2, mem_gb: 2, arch: "AMD" }, cpx21: { cpu: 3, mem_gb: 4, arch: "AMD" },
|
||
|
|
cpx31: { cpu: 4, mem_gb: 8, arch: "AMD" }, cpx41: { cpu: 8, mem_gb: 16, arch: "AMD" },
|
||
|
|
cpx51: { cpu: 16, mem_gb: 32, arch: "AMD" },
|
||
|
|
cx22: { cpu: 2, mem_gb: 4, arch: "Intel" }, cx32: { cpu: 4, mem_gb: 8, arch: "Intel" },
|
||
|
|
cx42: { cpu: 8, mem_gb: 16, arch: "Intel" }, cx52: { cpu: 16, mem_gb: 32, arch: "Intel" },
|
||
|
|
}
|
||
|
|
let spec = ($tbl | get -o $type_code)
|
||
|
|
if ($spec == null) {
|
||
|
|
{ type: $type_code, cpu: 0, cpu_millis: 0, mem_gb: 0, mem_bytes: 0, arch: "", label: $type_code }
|
||
|
|
} else {
|
||
|
|
{
|
||
|
|
type: $type_code,
|
||
|
|
cpu: $spec.cpu,
|
||
|
|
cpu_millis: ($spec.cpu * 1000),
|
||
|
|
mem_gb: $spec.mem_gb,
|
||
|
|
mem_bytes: ($spec.mem_gb * 1024 * 1024 * 1024),
|
||
|
|
arch: $spec.arch,
|
||
|
|
label: $"($spec.cpu)CPU ($spec.mem_gb)GB ($spec.arch)",
|
||
|
|
}
|
||
|
|
}
|
||
|
|
}
|
||
|
|
|
||
|
|
# Translate Hetzner server type codes to a human-readable CPU/RAM/arch label.
|
||
|
|
def hetzner-type-label [type_code: string]: nothing -> string {
|
||
|
|
(comp-server-capacity $type_code).label
|
||
|
|
}
|
||
|
|
|
||
|
|
def status-dot [status: string]: nothing -> string {
|
||
|
|
match $status {
|
||
|
|
"running" => "●"
|
||
|
|
"off" => "○"
|
||
|
|
"starting" => "◐"
|
||
|
|
_ => "○"
|
||
|
|
}
|
||
|
|
}
|
||
|
|
|
||
|
|
# Load server runtime + declared metadata for all servers in a workspace infra.
|
||
|
|
# Cache mode: .servers-state.json (status/public_ip/location) + settings.ncl (type/private_ip).
|
||
|
|
# Live mode: hcloud API directly.
|
||
|
|
# Returns record keyed by hostname.
|
||
|
|
export def comp-load-server-info [
|
||
|
|
ws_root: string
|
||
|
|
infra: string
|
||
|
|
--live
|
||
|
|
]: nothing -> record {
|
||
|
|
let state_path = ($ws_root | path join "infra" $infra ".servers-state.json")
|
||
|
|
let cache = if ($state_path | path exists) { open $state_path } else { {} }
|
||
|
|
|
||
|
|
let declared = do -i {
|
||
|
|
let raw = (comp-ncl-export $ws_root $"infra/($infra)/settings.ncl")
|
||
|
|
$raw | get -o servers | default []
|
||
|
|
| reduce --fold {} {|srv, acc|
|
||
|
|
$acc | insert $srv.hostname {
|
||
|
|
type: ($srv.server_type? | default ""),
|
||
|
|
private_ip: ($srv | get -o networking.private_ip | default ""),
|
||
|
|
provider: ($srv.provider? | default "hetzner"),
|
||
|
|
}
|
||
|
|
}
|
||
|
|
} | default {}
|
||
|
|
|
||
|
|
if $live {
|
||
|
|
let res = (do { ^hcloud server list -o json } | complete)
|
||
|
|
if $res.exit_code != 0 { return $declared }
|
||
|
|
($res.stdout | from json)
|
||
|
|
| reduce --fold {} {|h, acc|
|
||
|
|
let d = ($declared | get -o $h.name | default {})
|
||
|
|
$acc | insert $h.name {
|
||
|
|
status: $h.status,
|
||
|
|
type: ($h | get -o server_type.name | default ($d.type? | default "")),
|
||
|
|
location: ($h | get -o datacenter.location.name | default ""),
|
||
|
|
public_ip: ($h | get -o public_net.ipv4.ip | default ""),
|
||
|
|
private_ip: ($d.private_ip? | default ""),
|
||
|
|
provider: ($d.provider? | default "hetzner"),
|
||
|
|
}
|
||
|
|
}
|
||
|
|
} else {
|
||
|
|
$cache | items {|hostname entry|
|
||
|
|
let d = ($declared | get -o $hostname | default {})
|
||
|
|
{
|
||
|
|
key: $hostname,
|
||
|
|
value: {
|
||
|
|
status: ($entry.status? | default ""),
|
||
|
|
type: ($d.type? | default ""),
|
||
|
|
location: ($entry.location? | default ""),
|
||
|
|
public_ip: ($entry.public_ip? | default ""),
|
||
|
|
private_ip: ($d.private_ip? | default ""),
|
||
|
|
provider: ($d.provider? | default "hetzner"),
|
||
|
|
},
|
||
|
|
}
|
||
|
|
} | reduce --fold {} {|it, acc| $acc | insert $it.key $it.value}
|
||
|
|
}
|
||
|
|
}
|
||
|
|
|
||
|
|
# One-line server info badge for the TASKSERV section header.
|
||
|
|
export def comp-server-badge [info: record]: nothing -> string {
|
||
|
|
if ($info | is-empty) { return "" }
|
||
|
|
let dot = (status-dot ($info.status? | default ""))
|
||
|
|
let status = ($info.status? | default "")
|
||
|
|
let type_l = if ($info.type? | default "" | is-not-empty) {
|
||
|
|
hetzner-type-label $info.type
|
||
|
|
} else { "" }
|
||
|
|
let loc = ($info.location? | default "")
|
||
|
|
let pub = ($info.public_ip? | default "")
|
||
|
|
let priv = ($info.private_ip? | default "")
|
||
|
|
let ip_str = if ($pub | is-not-empty) and ($priv | is-not-empty) {
|
||
|
|
$"($pub) / ($priv)"
|
||
|
|
} else if ($pub | is-not-empty) { $pub } else { $priv }
|
||
|
|
let parts = [$"($dot) ($status)" $type_l $loc $ip_str] | where { $in | is-not-empty }
|
||
|
|
if ($parts | is-empty) { "" } else { $parts | str join " " }
|
||
|
|
}
|
||
|
|
|
||
|
|
def cluster-cache-path [ws_root: string, infra: string]: nothing -> string {
|
||
|
|
$ws_root | path join "infra" $infra ".cluster-state.json"
|
||
|
|
}
|
||
|
|
|
||
|
|
# Read the declared k0s version from settings.ncl (settings.k0s.version).
|
||
|
|
export def comp-settings-k0s-version [ws_root: string, infra: string]: nothing -> string {
|
||
|
|
let raw = (comp-ncl-export $ws_root $"infra/($infra)/settings.ncl")
|
||
|
|
$raw | get -o settings | default {} | get -o k0s | default {} | get -o version | default ""
|
||
|
|
}
|
||
|
|
|
||
|
|
# Fetch cluster summary counts via the declared remote kubectl (single SSH call).
|
||
|
|
# Returns the same shape as comp-cluster-summary's result record. node/pod totals
|
||
|
|
# come from the caller's already-fetched k8s_data; this adds ns/dply/pvc counts.
|
||
|
|
def comp-kubectl-summary-counts [ws_root: string, infra: string, version: string, k8s_data: record]: nothing -> record {
|
||
|
|
let nodes = ($k8s_data.nodes? | default [])
|
||
|
|
let pods = ($k8s_data.pods? | default [])
|
||
|
|
let nodes_total = ($nodes | length)
|
||
|
|
let nodes_ready = ($nodes | where {|n|
|
||
|
|
($n | get -o status | default {} | get -o conditions | default []) | any { ($in.type? == "Ready") and ($in.status? == "True") }
|
||
|
|
} | length)
|
||
|
|
let pods_total = ($pods | length)
|
||
|
|
let pods_running = ($pods | where { ($in | get -o status | default {} | get -o phase | default "") == "Running" } | length)
|
||
|
|
let cp_ip = ($nodes | get 0? | default {} | get -o status | default {} | get -o addresses | default [] | where { $in.type? == "InternalIP" } | get 0? | get -o address | default "")
|
||
|
|
let r = (comp-kubectl-ssh $ws_root $infra "get namespaces,deployments,pvc -A --no-headers -o json")
|
||
|
|
let counts = if $r.exit_code == 0 {
|
||
|
|
let items = (try { $r.stdout | from json | get -o items | default [] } catch { [] })
|
||
|
|
{
|
||
|
|
ns: ($items | where { ($in.kind? | default "") == "Namespace" } | length),
|
||
|
|
dply: ($items | where { ($in.kind? | default "") == "Deployment" } | length),
|
||
|
|
pvc: ($items | where { (($in.kind? | default "") == "PersistentVolumeClaim") and (($in.status?.phase? | default "") == "Bound") } | length),
|
||
|
|
}
|
||
|
|
} else { { ns: 0, dply: 0, pvc: 0 } }
|
||
|
|
{
|
||
|
|
version: $version,
|
||
|
|
namespaces: $counts.ns,
|
||
|
|
deployments: $counts.dply,
|
||
|
|
pvcs_bound: $counts.pvc,
|
||
|
|
nodes_ready: $nodes_ready,
|
||
|
|
nodes_total: $nodes_total,
|
||
|
|
pods_running: $pods_running,
|
||
|
|
pods_total: $pods_total,
|
||
|
|
cp_ip: $cp_ip,
|
||
|
|
last_sync: (date now | format date "%Y-%m-%dT%H:%M:%SZ"),
|
||
|
|
}
|
||
|
|
}
|
||
|
|
|
||
|
|
# Fetch (--live) or read cached cluster summary: version, ns/dply/pvc counts, node+pod status.
|
||
|
|
# Live: second lightweight SSH (counts only — separate from comp-ssh-k8s-data heavy JSON call).
|
||
|
|
# Cache: writes .cluster-state.json on live fetch; reads it on non-live call.
|
||
|
|
export def comp-cluster-summary [
|
||
|
|
ws_root: string
|
||
|
|
infra: string
|
||
|
|
k8s_data: record # node/pod data already fetched by comp-ssh-k8s-data
|
||
|
|
version: string # declared k0s/k8s version from settings — caller resolves via comp-settings-k0s-version
|
||
|
|
--live
|
||
|
|
]: nothing -> record {
|
||
|
|
let empty = {
|
||
|
|
version: "", namespaces: 0, deployments: 0, pvcs_bound: 0, last_sync: "",
|
||
|
|
nodes_ready: 0, nodes_total: 0, pods_running: 0, pods_total: 0, cp_ip: "",
|
||
|
|
}
|
||
|
|
if not $live {
|
||
|
|
let cache = (cluster-cache-path $ws_root $infra)
|
||
|
|
let base = if ($cache | path exists) { open $cache } else { $empty }
|
||
|
|
return ($base | merge { version: (if ($version | is-not-empty) { $version } else { $base.version? | default "" }) })
|
||
|
|
}
|
||
|
|
# Live: counts come from the declared remote kubectl (SSH to control-plane).
|
||
|
|
# provisioning never runs kubectl locally — see comp-kubectl-access.
|
||
|
|
let result = (comp-kubectl-summary-counts $ws_root $infra $version $k8s_data)
|
||
|
|
$result | to json | save -f (cluster-cache-path $ws_root $infra)
|
||
|
|
$result
|
||
|
|
}
|
||
|
|
|
||
|
|
# One-line cluster summary badge for the CLUSTER COMPONENTS header.
|
||
|
|
export def comp-cluster-badge [summary: record]: nothing -> string {
|
||
|
|
if ($summary | is-empty) { return "" }
|
||
|
|
let ver = ($summary.version? | default "")
|
||
|
|
let ns = ($summary.namespaces? | default 0)
|
||
|
|
let pods = ($summary.pods_total? | default 0)
|
||
|
|
if ($ver | is-empty) and $ns == 0 and $pods == 0 { return "" }
|
||
|
|
let cp_str = if ($summary.cp_ip? | default "" | is-not-empty) { $"CP: ($summary.cp_ip)" } else { "" }
|
||
|
|
let ver_str = if ($ver | is-not-empty) { $ver } else { "" }
|
||
|
|
let nodes_str = if ($summary.nodes_total? | default 0) > 0 { $"($summary.nodes_ready)/($summary.nodes_total) nodes" } else { "" }
|
||
|
|
let pods_str = if ($summary.pods_total? | default 0) > 0 { $"($summary.pods_running)/($summary.pods_total) pods" } else { "" }
|
||
|
|
let ns_str = if $ns > 0 { $"($ns) ns" } else { "" }
|
||
|
|
let dply_str = if ($summary.deployments? | default 0) > 0 { $"($summary.deployments) dply" } else { "" }
|
||
|
|
let pvcs_str = if ($summary.pvcs_bound? | default 0) > 0 { $"($summary.pvcs_bound) pvcs" } else { "" }
|
||
|
|
[$cp_str $ver_str $nodes_str $pods_str $ns_str $dply_str $pvcs_str] | where { $in | is-not-empty } | str join " "
|
||
|
|
}
|
||
|
|
|
||
|
|
# Parse a Kubernetes resource quantity (e.g. "529Mi", "1Gi", "1024Ki", "2G") to
|
||
|
|
# bytes for numeric sorting and capacity math. Unparseable input yields 0.
|
||
|
|
def k8s-quantity-bytes [q: string]: nothing -> int {
|
||
|
|
let s = ($q | str trim)
|
||
|
|
if ($s | is-empty) { return 0 }
|
||
|
|
let m = ($s | parse --regex '(?<num>[0-9.]+)(?<unit>[A-Za-z]*)' | get 0?)
|
||
|
|
if ($m == null) { return 0 }
|
||
|
|
let num = ($m.num | into float)
|
||
|
|
let mult = (match $m.unit {
|
||
|
|
"Ki" => 1024.0, "Mi" => 1048576.0, "Gi" => 1073741824.0, "Ti" => 1099511627776.0,
|
||
|
|
"k" | "K" => 1000.0, "M" => 1000000.0, "G" => 1000000000.0, "T" => 1000000000000.0,
|
||
|
|
_ => 1.0,
|
||
|
|
})
|
||
|
|
(($num * $mult) | math round | into int)
|
||
|
|
}
|
||
|
|
|
||
|
|
# Operator-relevant node labels: drop the well-known system/cloud domains so only
|
||
|
|
# placement-meaningful labels (node_class, can_use_fip.*, custom) remain. Returns
|
||
|
|
# a "k=v k=v" string (valueless labels shown as bare keys).
|
||
|
|
def node-operator-labels [labels: record]: nothing -> string {
|
||
|
|
let system = ["kubernetes.io", "k8s.io", "hetzner.cloud", "topology.", "beta.", "csi."]
|
||
|
|
$labels | transpose key value
|
||
|
|
| where {|r| not ($system | any {|p| $r.key | str contains $p }) }
|
||
|
|
| each {|r|
|
||
|
|
# Boolean labels (value "true") and valueless labels show as the bare key —
|
||
|
|
# if it's present it's set. Non-boolean labels keep key=value.
|
||
|
|
if (($r.value | default "" | is-empty) or ($r.value == "true")) { $r.key } else { $"($r.key)=($r.value)" }
|
||
|
|
}
|
||
|
|
| str join " "
|
||
|
|
}
|
||
|
|
|
||
|
|
# Live per-node resource usage via `kubectl top`, enriched with the declared
|
||
|
|
# Hetzner server type and its nominal capacity. provisioning never runs kubectl
|
||
|
|
# locally — metrics arrive over the bounded remote-kubectl SSH path
|
||
|
|
# (comp-kubectl-ssh). kubectl's own percentages are relative to node *allocatable*
|
||
|
|
# (physical RAM minus kernel/kube reservations), which is why the nominal column
|
||
|
|
# differs from what the % implies. --pods appends the heaviest pods by memory.
|
||
|
|
export def "main cluster top" [
|
||
|
|
mode?: string = "" # "pods" appends the heaviest pods by memory
|
||
|
|
--workspace (-w): string = ""
|
||
|
|
--infra (-i): string = ""
|
||
|
|
--limit (-l): int = 15
|
||
|
|
--out (-o): string = "" # json | yaml | table (default terminal view)
|
||
|
|
]: nothing -> any {
|
||
|
|
let pods = ($mode in ["pods", "p"])
|
||
|
|
let ws_root = (comp-ws-path $workspace)
|
||
|
|
if ($ws_root | is-empty) { cli-err "no workspace resolved" }
|
||
|
|
let infra_name = if ($infra | is-not-empty) { $infra } else { comp-ws-infra $ws_root }
|
||
|
|
if ($infra_name | is-empty) { cli-err $"no infra found under ($ws_root)/infra" }
|
||
|
|
|
||
|
|
let nodes_res = (comp-kubectl-ssh $ws_root $infra_name "top nodes --no-headers")
|
||
|
|
if $nodes_res.exit_code != 0 {
|
||
|
|
cli-err $"kubectl top nodes failed \(metrics-server reachable?\): ($nodes_res.stderr | str trim)"
|
||
|
|
}
|
||
|
|
let servers = (comp-load-server-info $ws_root $infra_name)
|
||
|
|
|
||
|
|
# Per-node requested/limited memory vs allocatable — the bin-packing signal
|
||
|
|
# kubectl top omits (top reports live usage, not scheduler reservations).
|
||
|
|
# Sourced from pod specs via one SSH JSON fetch; pods in a terminal phase are
|
||
|
|
# excluded to match kubectl's "Allocated resources". no-req counts containers
|
||
|
|
# with no memory request (invisible weight that defeats the scheduler).
|
||
|
|
let access = (comp-kubectl-access $ws_root $infra_name)
|
||
|
|
let k8s = (comp-ssh-k8s-data $ws_root $infra_name $access.host)
|
||
|
|
let alloc = ($k8s.nodes | reduce --fold {} {|n, acc|
|
||
|
|
$acc | insert ($n.metadata.name) (k8s-quantity-bytes ($n.status?.allocatable?.memory? | default "0"))
|
||
|
|
})
|
||
|
|
let labelmap = ($k8s.nodes | reduce --fold {} {|n, acc|
|
||
|
|
$acc | insert ($n.metadata.name) (node-operator-labels ($n.metadata?.labels? | default {}))
|
||
|
|
})
|
||
|
|
let agg = ($k8s.pods
|
||
|
|
| where { (($in.status?.phase? | default "") not-in ["Succeeded", "Failed"]) and (($in.spec?.nodeName? | default "") | is-not-empty) }
|
||
|
|
| reduce --fold {} {|p, acc|
|
||
|
|
let node = ($p.spec.nodeName)
|
||
|
|
let cts = ($p.spec?.containers? | default [])
|
||
|
|
let req = ($cts | reduce --fold 0 {|c, s| $s + (k8s-quantity-bytes ($c.resources?.requests?.memory? | default "0")) })
|
||
|
|
let lim = ($cts | reduce --fold 0 {|c, s| $s + (k8s-quantity-bytes ($c.resources?.limits?.memory? | default "0")) })
|
||
|
|
let nor = ($cts | where { ($in.resources?.requests?.memory? | default "" | is-empty) } | length)
|
||
|
|
let cur = ($acc | get -o $node | default { req: 0, lim: 0, noreq: 0 })
|
||
|
|
$acc | upsert $node { req: ($cur.req + $req), lim: ($cur.lim + $lim), noreq: ($cur.noreq + $nor) }
|
||
|
|
})
|
||
|
|
|
||
|
|
let node_rows = ($nodes_res.stdout | str replace --all "\r" "" | lines
|
||
|
|
| where { ($in | str trim | is-not-empty) }
|
||
|
|
| each {|ln|
|
||
|
|
let c = ($ln | str trim | split row --regex '\s+')
|
||
|
|
let name = ($c | get 0? | default "")
|
||
|
|
let type = ($servers | get -o $name | default {} | get -o type | default "")
|
||
|
|
let cap = (comp-server-capacity $type)
|
||
|
|
let a = ($alloc | get -o $name | default 0)
|
||
|
|
let g = ($agg | get -o $name | default { req: 0, lim: 0, noreq: 0 })
|
||
|
|
{
|
||
|
|
node: $name,
|
||
|
|
type: (if ($type | is-not-empty) { $type } else { "—" }),
|
||
|
|
nominal: (if $cap.mem_gb > 0 { $"($cap.cpu)c/($cap.mem_gb)Gi" } else { "—" }),
|
||
|
|
cpu: ($c | get 1? | default ""),
|
||
|
|
"cpu%": ($c | get 2? | default ""),
|
||
|
|
mem: ($c | get 3? | default ""),
|
||
|
|
"mem%": ($c | get 4? | default ""),
|
||
|
|
"req%": (if $a > 0 { $"(($g.req * 100 / $a) | math round)%" } else { "—" }),
|
||
|
|
"lim%": (if $a > 0 { $"(($g.lim * 100 / $a) | math round)%" } else { "—" }),
|
||
|
|
"no-req": $g.noreq,
|
||
|
|
labels: ($labelmap | get -o $name | default ""),
|
||
|
|
}
|
||
|
|
})
|
||
|
|
|
||
|
|
let pod_rows = if $pods {
|
||
|
|
let pods_res = (comp-kubectl-ssh $ws_root $infra_name "top pods --all-namespaces --no-headers")
|
||
|
|
if $pods_res.exit_code != 0 {
|
||
|
|
print --stderr $"kubectl top pods failed: ($pods_res.stderr | str trim)"
|
||
|
|
[]
|
||
|
|
} else {
|
||
|
|
$pods_res.stdout | str replace --all "\r" "" | lines
|
||
|
|
| where { ($in | str trim | is-not-empty) }
|
||
|
|
| each {|ln|
|
||
|
|
let c = ($ln | str trim | split row --regex '\s+')
|
||
|
|
{
|
||
|
|
namespace: ($c | get 0? | default ""),
|
||
|
|
pod: ($c | get 1? | default ""),
|
||
|
|
cpu: ($c | get 2? | default ""),
|
||
|
|
mem: ($c | get 3? | default ""),
|
||
|
|
mem_bytes: (k8s-quantity-bytes ($c | get 3? | default "0")),
|
||
|
|
}
|
||
|
|
}
|
||
|
|
| sort-by mem_bytes --reverse
|
||
|
|
| first $limit
|
||
|
|
| reject mem_bytes
|
||
|
|
}
|
||
|
|
} else { [] }
|
||
|
|
|
||
|
|
let data = if $pods { { nodes: $node_rows, pods: $pod_rows } } else { $node_rows }
|
||
|
|
match ($out | str downcase) {
|
||
|
|
"json" => ($data | to json)
|
||
|
|
"yaml" | "yml" => ($data | to yaml)
|
||
|
|
"" | "table" => {
|
||
|
|
if $pods {
|
||
|
|
print "NODES"
|
||
|
|
print ($node_rows | table)
|
||
|
|
cluster-top-legend
|
||
|
|
print ""
|
||
|
|
print $"TOP ($limit) PODS BY MEMORY"
|
||
|
|
$pod_rows
|
||
|
|
} else {
|
||
|
|
print ($node_rows | table)
|
||
|
|
cluster-top-legend
|
||
|
|
null
|
||
|
|
}
|
||
|
|
}
|
||
|
|
_ => { cli-err $"unknown --out format: ($out) \(use json|yaml|table\)" }
|
||
|
|
}
|
||
|
|
}
|
||
|
|
|
||
|
|
# Interpretive legend for `cluster top` columns — printed under the node table so
|
||
|
|
# the numbers can be read without external context. Suppressed in json/yaml output.
|
||
|
|
def cluster-top-legend []: nothing -> nothing {
|
||
|
|
print ""
|
||
|
|
print "How to read it:"
|
||
|
|
print " nominal physical CPU/RAM of the Hetzner type (cax11=2c/4Gi, cax21=4c/8Gi)."
|
||
|
|
print " cpu /mem LIVE usage now (kubectl top). mem% is vs ALLOCATABLE (RAM minus"
|
||
|
|
print " kernel+kube reservations), so it reads higher than vs nominal."
|
||
|
|
print " req% sum of pod memory REQUESTS vs allocatable = what the scheduler"
|
||
|
|
print " reserved. Healthy < 80%; it is the real 'fullness' for placement."
|
||
|
|
print " lim% sum of pod memory LIMITS vs allocatable = overcommit. > 100% means"
|
||
|
|
print " pods may collectively exceed RAM → OOM-kill risk under pressure."
|
||
|
|
print " no-req containers with NO memory request — invisible to the scheduler,"
|
||
|
|
print " cause blind bin-packing and are first to be OOM-killed."
|
||
|
|
print " labels operator node labels (node_class, can_use_fip.*, custom) that drive"
|
||
|
|
print " placement; system/cloud domains (kubernetes.io, hetzner.cloud) hidden."
|
||
|
|
}
|
||
|
|
|
||
|
|
# Reconcile declared operator node labels onto the live Kubernetes nodes. Source
|
||
|
|
# of truth: servers.ncl node_labels (record) + node_class (list, projected as
|
||
|
|
# node-class.<class>=true). Closes the drift where declared labels never reach the
|
||
|
|
# nodes (no kubectl label runs at join time). Dry-run by default: prints the plan;
|
||
|
|
# --apply runs `kubectl label node ... --overwrite` over the bounded SSH path.
|
||
|
|
export def "main cluster label-nodes" [
|
||
|
|
--workspace (-w): string = ""
|
||
|
|
--infra (-i): string = ""
|
||
|
|
--apply (-a)
|
||
|
|
]: nothing -> any {
|
||
|
|
let ws_root = (comp-ws-path $workspace)
|
||
|
|
if ($ws_root | is-empty) { cli-err "no workspace resolved" }
|
||
|
|
let infra_name = if ($infra | is-not-empty) { $infra } else { comp-ws-infra $ws_root }
|
||
|
|
if ($infra_name | is-empty) { cli-err $"no infra found under ($ws_root)/infra" }
|
||
|
|
|
||
|
|
let settings = (comp-ncl-export $ws_root $"infra/($infra_name)/settings.ncl")
|
||
|
|
let servers = ($settings | get -o servers | default []
|
||
|
|
| where { ($in.not_use? | default false) == false })
|
||
|
|
|
||
|
|
# Desired operator labels per node: node_labels record + node_class projection.
|
||
|
|
let desired = ($servers | reduce --fold {} {|s, acc|
|
||
|
|
let base = ($s.node_labels? | default {})
|
||
|
|
let cls = ($s.node_class? | default [] | reduce --fold {} {|c, m| $m | insert $"node-class.($c)" "true" })
|
||
|
|
$acc | insert ($s.hostname) ($base | merge $cls)
|
||
|
|
})
|
||
|
|
|
||
|
|
let access = (comp-kubectl-access $ws_root $infra_name)
|
||
|
|
let k8s = (comp-ssh-k8s-data $ws_root $infra_name $access.host)
|
||
|
|
let live = ($k8s.nodes | reduce --fold {} {|n, acc|
|
||
|
|
$acc | insert ($n.metadata.name) ($n.metadata?.labels? | default {})
|
||
|
|
})
|
||
|
|
|
||
|
|
# Diff via transpose (label keys carry dots/slashes — unsafe for cell-path get).
|
||
|
|
let plan = ($desired | transpose host want | each {|e|
|
||
|
|
let have = (($live | get -o $e.host | default {}) | transpose key value)
|
||
|
|
$e.want | transpose key value | each {|w|
|
||
|
|
let m = ($have | where key == $w.key)
|
||
|
|
let cur = if ($m | is-empty) { null } else { ($m | first | get value) }
|
||
|
|
if $cur != $w.value {
|
||
|
|
{ node: $e.host, label: $w.key, current: ($cur | default "—"), desired: $w.value }
|
||
|
|
} else { null }
|
||
|
|
} | compact
|
||
|
|
} | flatten)
|
||
|
|
|
||
|
|
if ($plan | is-empty) {
|
||
|
|
print "All declared node labels already applied — nothing to do."
|
||
|
|
return []
|
||
|
|
}
|
||
|
|
if not $apply {
|
||
|
|
print $"Planned label changes \(($plan | length)\) — dry-run, use --apply to apply:"
|
||
|
|
return $plan
|
||
|
|
}
|
||
|
|
$plan | each {|c|
|
||
|
|
let r = (comp-kubectl-ssh $ws_root $infra_name $"label node ($c.node) ($c.label)=($c.desired) --overwrite")
|
||
|
|
{ node: $c.node, label: $c.label, desired: $c.desired, status: (if $r.exit_code == 0 { "applied" } else { ($r.stderr | str trim) }) }
|
||
|
|
}
|
||
|
|
}
|
||
|
|
|
||
|
|
# Placement simulator: project the per-node impact of (re)deploying a service.
|
||
|
|
# Reads the candidate's declared resources_effective (memory request) + placement
|
||
|
|
# from settings, the live node snapshot (allocatable, current pod requests, live
|
||
|
|
# usage) via the bounded SSH path, then computes projected requests% and
|
||
|
|
# working-set% per eligible node (eligibility = node-class label ∩ placement.node_class).
|
||
|
|
# Answers "does the next deploy fit, and where?" — the dry-run the ops loop lacked.
|
||
|
|
export def "main cluster plan" [
|
||
|
|
service: string
|
||
|
|
--workspace (-w): string = ""
|
||
|
|
--infra (-i): string = ""
|
||
|
|
--out (-o): string = ""
|
||
|
|
--req-threshold: int = 80 # % allocatable requests treated as "full"
|
||
|
|
--use-threshold: int = 90 # % allocatable working-set treated as risky
|
||
|
|
]: nothing -> any {
|
||
|
|
let ws_root = (comp-ws-path $workspace)
|
||
|
|
if ($ws_root | is-empty) { cli-err "no workspace resolved" }
|
||
|
|
let infra_name = if ($infra | is-not-empty) { $infra } else { comp-ws-infra $ws_root }
|
||
|
|
if ($infra_name | is-empty) { cli-err $"no infra found under ($ws_root)/infra" }
|
||
|
|
|
||
|
|
# Candidate spec from declared NCL (settings.components.<key>).
|
||
|
|
let settings = (comp-ncl-export $ws_root $"infra/($infra_name)/settings.ncl")
|
||
|
|
let comps = ($settings | get -o components | default {})
|
||
|
|
let key = if (($comps | get -o $service) != null) { $service } else { ($service | str replace --all "-" "_") }
|
||
|
|
let comp = ($comps | get -o $key | default null)
|
||
|
|
if ($comp == null) { cli-err $"component '($service)' not found in settings.components" }
|
||
|
|
let re = ($comp | get -o resources_effective | default {})
|
||
|
|
if ($re | is-empty) { cli-err $"component '($service)' has no resources_effective — declare class/resources first \(step 2c\)" }
|
||
|
|
let cand_req = (k8s-quantity-bytes ($re | get -o memory | default {} | get -o request | default "0"))
|
||
|
|
let cand_lim = (k8s-quantity-bytes ($re | get -o memory | default {} | get -o limit | default "0"))
|
||
|
|
let want_nc = ($comp | get -o placement | default {} | get -o node_class | default [] | each {|c| $c | into string})
|
||
|
|
let svc_class = ($comp | get -o class | default "")
|
||
|
|
|
||
|
|
let access = (comp-kubectl-access $ws_root $infra_name)
|
||
|
|
let k8s = (comp-ssh-k8s-data $ws_root $infra_name $access.host)
|
||
|
|
if ($k8s.nodes | is-empty) { cli-err $"could not fetch live nodes \(CP: ($access.host | default 'unresolved'))" }
|
||
|
|
let top = (comp-kubectl-ssh $ws_root $infra_name "top nodes --no-headers")
|
||
|
|
let usage = if $top.exit_code == 0 {
|
||
|
|
$top.stdout | str replace --all "\r" "" | lines | where { ($in | str trim | is-not-empty) }
|
||
|
|
| reduce --fold {} {|ln, acc|
|
||
|
|
let c = ($ln | str trim | split row --regex '\s+')
|
||
|
|
$acc | insert ($c | get 0? | default "") (k8s-quantity-bytes ($c | get 3? | default "0"))
|
||
|
|
}
|
||
|
|
} else { {} }
|
||
|
|
|
||
|
|
let live_pods = ($k8s.pods | where { (($in.status?.phase? | default "") not-in ["Succeeded", "Failed"]) and (($in.spec?.nodeName? | default "") | is-not-empty) })
|
||
|
|
let rows = ($k8s.nodes | each {|n|
|
||
|
|
let name = ($n.metadata.name)
|
||
|
|
let alloc = (k8s-quantity-bytes ($n.status?.allocatable?.memory? | default "0"))
|
||
|
|
let labels = ($n.metadata?.labels? | default {} | transpose k v
|
||
|
|
| where {|r| $r.k | str starts-with "node-class." } | each {|r| $r.k | str replace "node-class." "" })
|
||
|
|
let eligible = if ($want_nc | is-empty) { true } else { ($want_nc | any {|c| $c in $labels }) }
|
||
|
|
let cur_req = ($live_pods | where { ($in.spec.nodeName) == $name }
|
||
|
|
| reduce --fold 0 {|p, s| $s + ($p.spec?.containers? | default [] | reduce --fold 0 {|c, s2| $s2 + (k8s-quantity-bytes ($c.resources?.requests?.memory? | default "0")) }) })
|
||
|
|
let cur_use = ($usage | get -o $name | default 0)
|
||
|
|
let proj_req = ($cur_req + $cand_req)
|
||
|
|
let proj_use = ($cur_use + $cand_req)
|
||
|
|
let req_pct = (if $alloc > 0 { ($proj_req * 100 / $alloc) | math round } else { 0 })
|
||
|
|
let use_pct = (if $alloc > 0 { ($proj_use * 100 / $alloc) | math round } else { 0 })
|
||
|
|
{
|
||
|
|
node: $name,
|
||
|
|
eligible: $eligible,
|
||
|
|
"req now": (if $alloc > 0 { $"(($cur_req * 100 / $alloc) | math round)%" } else { "—" }),
|
||
|
|
"req→": (if $alloc > 0 { $"($req_pct)%" } else { "—" }),
|
||
|
|
"ws→": (if $alloc > 0 { $"($use_pct)%" } else { "—" }),
|
||
|
|
fits: ($eligible and ($req_pct <= $req_threshold) and ($use_pct <= $use_threshold)),
|
||
|
|
_req_pct: $req_pct,
|
||
|
|
}
|
||
|
|
})
|
||
|
|
|
||
|
|
let eligible_fit = ($rows | where {|r| $r.eligible and $r.fits } | sort-by _req_pct)
|
||
|
|
let best = ($eligible_fit | get 0? | get -o node | default "")
|
||
|
|
let mem_h = ($re | get -o memory | default {})
|
||
|
|
let summary = {
|
||
|
|
service: $key, class: $svc_class,
|
||
|
|
request: ($mem_h | get -o request | default "?"), limit: ($mem_h | get -o limit | default "?"),
|
||
|
|
node_class: ($want_nc | str join ","),
|
||
|
|
recommendation: (if ($best | is-not-empty) {
|
||
|
|
$"schedules on ($best) \(lowest projected requests among eligible nodes that fit\)"
|
||
|
|
} else {
|
||
|
|
"NO eligible node fits within thresholds — activate another worker (wrk-1) or resize a node"
|
||
|
|
}),
|
||
|
|
}
|
||
|
|
|
||
|
|
let out_rows = ($rows | reject _req_pct)
|
||
|
|
match ($out | str downcase) {
|
||
|
|
"json" => ({ summary: $summary, nodes: $out_rows } | to json)
|
||
|
|
"yaml" | "yml" => ({ summary: $summary, nodes: $out_rows } | to yaml)
|
||
|
|
_ => {
|
||
|
|
print $"plan: ($key) class=($svc_class) mem req/lim=($summary.request)/($summary.limit) prefers node-class=($summary.node_class)"
|
||
|
|
print ($out_rows | table)
|
||
|
|
print $"→ ($summary.recommendation)"
|
||
|
|
null
|
||
|
|
}
|
||
|
|
}
|
||
|
|
}
|
||
|
|
|
||
|
|
# Format a byte count as a compact Gi/Mi string.
|
||
|
|
def fmt-gib [b: int]: nothing -> string {
|
||
|
|
if $b >= 1073741824 { $"(($b / 1073741824 * 10 | math round) / 10)Gi" } else { $"(($b / 1048576) | math round)Mi" }
|
||
|
|
}
|
||
|
|
|
||
|
|
# Cluster capacity analysis: per-node working-set broken down by category
|
||
|
|
# (movable = general app classes, pinned = datastore/storage_engine, infra =
|
||
|
|
# kube-system/longhorn/cilium/daemonsets), plus an aggregate capacity check by
|
||
|
|
# node-class with what-if projections. Answers "horizontal vs vertical": movable
|
||
|
|
# load dominating storage nodes → add a general worker; pinned dominating → resize.
|
||
|
|
export def "main cluster capacity" [
|
||
|
|
--workspace (-w): string = ""
|
||
|
|
--infra (-i): string = ""
|
||
|
|
--out (-o): string = ""
|
||
|
|
--add-worker # what-if: add one cax11 general worker (~3.4Gi alloc)
|
||
|
|
--resize: string = "" # what-if: resize <node> cax11→cax21 (+4Gi alloc)
|
||
|
|
]: nothing -> any {
|
||
|
|
let ws_root = (comp-ws-path $workspace)
|
||
|
|
if ($ws_root | is-empty) { cli-err "no workspace resolved" }
|
||
|
|
let infra_name = if ($infra | is-not-empty) { $infra } else { comp-ws-infra $ws_root }
|
||
|
|
if ($infra_name | is-empty) { cli-err $"no infra found under ($ws_root)/infra" }
|
||
|
|
|
||
|
|
let settings = (comp-ncl-export $ws_root $"infra/($infra_name)/settings.ncl")
|
||
|
|
let comps = ($settings | get -o components | default {})
|
||
|
|
# pod-name-prefix → pinned? (component name with _→- is the usual pod prefix)
|
||
|
|
let prefixes = ($comps | items {|k v|
|
||
|
|
{ prefix: ($k | str replace --all "_" "-"), pinned: (($v.class? | default "") in ["datastore", "storage_engine"]) }
|
||
|
|
} | sort-by {|c| $c.prefix | str length } --reverse)
|
||
|
|
let infra_ns = ["kube-system", "longhorn-system", "cilium", "cert-manager"]
|
||
|
|
|
||
|
|
let access = (comp-kubectl-access $ws_root $infra_name)
|
||
|
|
let k8s = (comp-ssh-k8s-data $ws_root $infra_name $access.host)
|
||
|
|
if ($k8s.nodes | is-empty) { cli-err $"could not fetch live nodes \(CP: ($access.host | default 'unresolved'))" }
|
||
|
|
let top = (comp-kubectl-ssh $ws_root $infra_name "top pods --all-namespaces --no-headers")
|
||
|
|
let usage = if $top.exit_code == 0 {
|
||
|
|
$top.stdout | str replace --all "\r" "" | lines | where { ($in | str trim | is-not-empty) }
|
||
|
|
| reduce --fold {} {|ln, acc|
|
||
|
|
let c = ($ln | str trim | split row --regex '\s+')
|
||
|
|
$acc | insert $"($c | get 0? | default '')/($c | get 1? | default '')" (k8s-quantity-bytes ($c | get 3? | default "0"))
|
||
|
|
}
|
||
|
|
} else { {} }
|
||
|
|
|
||
|
|
let live = ($k8s.pods | where { (($in.status?.phase? | default "") not-in ["Succeeded", "Failed"]) and (($in.spec?.nodeName? | default "") | is-not-empty) })
|
||
|
|
let per_node = ($k8s.nodes | each {|n|
|
||
|
|
let name = ($n.metadata.name)
|
||
|
|
let alloc = (k8s-quantity-bytes ($n.status?.allocatable?.memory? | default "0"))
|
||
|
|
let labels = ($n.metadata?.labels? | default {} | transpose k v
|
||
|
|
| where {|r| $r.k | str starts-with "node-class." } | each {|r| $r.k | str replace "node-class." "" })
|
||
|
|
let pods = ($live | where { ($in.spec.nodeName) == $name })
|
||
|
|
let cats = ($pods | reduce --fold { movable: 0, pinned: 0, infra: 0 } {|p, acc|
|
||
|
|
let ns = ($p.metadata.namespace)
|
||
|
|
let pn = ($p.metadata.name)
|
||
|
|
let use = ($usage | get -o $"($ns)/($pn)" | default 0)
|
||
|
|
let cat = if ($ns in $infra_ns) { "infra" } else {
|
||
|
|
let m = ($prefixes | where {|c| $pn | str starts-with $c.prefix } | get 0? | default null)
|
||
|
|
if ($m == null) { "movable" } else if $m.pinned { "pinned" } else { "movable" }
|
||
|
|
}
|
||
|
|
$acc | upsert $cat (($acc | get $cat) + $use)
|
||
|
|
})
|
||
|
|
{ node: $name, class: ($labels | str join ","), alloc: $alloc } | merge $cats
|
||
|
|
})
|
||
|
|
|
||
|
|
# Aggregate by node-class.
|
||
|
|
let general_alloc = ($per_node | where {|r| "general" in ($r.class | split row ",")} | reduce --fold 0 {|r, s| $s + $r.alloc})
|
||
|
|
let storage_alloc = ($per_node | where {|r| "storage" in ($r.class | split row ",")} | reduce --fold 0 {|r, s| $s + $r.alloc})
|
||
|
|
let movable_total = ($per_node | reduce --fold 0 {|r, s| $s + $r.movable})
|
||
|
|
let pinned_total = ($per_node | reduce --fold 0 {|r, s| $s + $r.pinned})
|
||
|
|
|
||
|
|
let cax11_alloc = 3650722201 # ~3.4Gi typical allocatable on a fresh cax11
|
||
|
|
let resize_node = ($per_node | where node == $resize | get 0? | default null)
|
||
|
|
let resize_is_general = (if ($resize_node != null) { "general" in ($resize_node.class | split row ",") } else { false })
|
||
|
|
let resize_is_storage = (if ($resize_node != null) { "storage" in ($resize_node.class | split row ",") } else { false })
|
||
|
|
let gen_alloc_wi = ($general_alloc
|
||
|
|
+ (if $add_worker { $cax11_alloc } else { 0 })
|
||
|
|
+ (if $resize_is_general { 4294967296 } else { 0 }))
|
||
|
|
let stg_alloc_wi = ($storage_alloc + (if $resize_is_storage { 4294967296 } else { 0 }))
|
||
|
|
|
||
|
|
let gen_util_now = (if $general_alloc > 0 { ($movable_total * 100 / $general_alloc | math round) } else { 0 })
|
||
|
|
let gen_util_wi = (if $gen_alloc_wi > 0 { ($movable_total * 100 / $gen_alloc_wi | math round) } else { 0 })
|
||
|
|
|
||
|
|
let scenario = if $add_worker { "+1 general worker (cax11)" } else if ($resize | is-not-empty) { $"resize ($resize) → cax21" } else { "current" }
|
||
|
|
let mv_s = (fmt-gib $movable_total)
|
||
|
|
let pn_s = (fmt-gib $pinned_total)
|
||
|
|
let recommendation = if ($movable_total > $pinned_total) {
|
||
|
|
$"movable load ($mv_s) dominates pinned ($pn_s) — HORIZONTAL: add a general worker to relieve the storage nodes; vertical resize would keep apps mislocated"
|
||
|
|
} else {
|
||
|
|
$"pinned/datastore load ($pn_s) dominates movable ($mv_s) — VERTICAL: resize the pressured storage node rather than adding a general worker"
|
||
|
|
}
|
||
|
|
|
||
|
|
let rows = ($per_node | each {|r|
|
||
|
|
{
|
||
|
|
node: $r.node, class: $r.class, alloc: (fmt-gib $r.alloc),
|
||
|
|
movable: (fmt-gib $r.movable), pinned: (fmt-gib $r.pinned), infra: (fmt-gib $r.infra),
|
||
|
|
}
|
||
|
|
})
|
||
|
|
let summary = {
|
||
|
|
movable_total: (fmt-gib $movable_total), pinned_total: (fmt-gib $pinned_total),
|
||
|
|
general_alloc: (fmt-gib $general_alloc), storage_alloc: (fmt-gib $storage_alloc),
|
||
|
|
general_util_now: $"($gen_util_now)%",
|
||
|
|
scenario: $scenario,
|
||
|
|
general_util_scenario: $"($gen_util_wi)%",
|
||
|
|
recommendation: $recommendation,
|
||
|
|
}
|
||
|
|
|
||
|
|
match ($out | str downcase) {
|
||
|
|
"json" => ({ nodes: $rows, summary: $summary } | to json)
|
||
|
|
"yaml" | "yml" => ({ nodes: $rows, summary: $summary } | to yaml)
|
||
|
|
_ => {
|
||
|
|
print "PER-NODE working-set by category (movable=general apps · pinned=datastore · infra=k8s/longhorn)"
|
||
|
|
print ($rows | table)
|
||
|
|
print ""
|
||
|
|
print $"movable total: ($summary.movable_total) pinned total: ($summary.pinned_total)"
|
||
|
|
print $"general node-class alloc: ($summary.general_alloc) storage: ($summary.storage_alloc)"
|
||
|
|
print $"general utilization \(movable/general-alloc\): now ($summary.general_util_now) → scenario [($scenario)] ($summary.general_util_scenario)"
|
||
|
|
print $"→ ($recommendation)"
|
||
|
|
null
|
||
|
|
}
|
||
|
|
}
|
||
|
|
}
|
||
|
|
|
||
|
|
# List all Kubernetes services via SSH to the cluster control plane.
|
||
|
|
# Matches CLI output of `kubectl get svc --all-namespaces`.
|
||
|
|
export def "main cluster services" [
|
||
|
|
--workspace (-w): string = ""
|
||
|
|
--infra (-i): string = ""
|
||
|
|
--namespace (-n): string = "" # filter to single namespace (default: all)
|
||
|
|
--out (-o): string = "" # json | yaml | table (default terminal view)
|
||
|
|
]: nothing -> any {
|
||
|
|
let ws_root = (comp-ws-path $workspace)
|
||
|
|
if ($ws_root | is-empty) { cli-err "no workspace resolved" }
|
||
|
|
let infra_name = if ($infra | is-not-empty) { $infra } else { comp-ws-infra $ws_root }
|
||
|
|
if ($infra_name | is-empty) { cli-err $"no infra found under ($ws_root)/infra" }
|
||
|
|
|
||
|
|
let ns_flag = if ($namespace | is-not-empty) { $"-n ($namespace)" } else { "--all-namespaces" }
|
||
|
|
let r = (comp-kubectl-ssh $ws_root $infra_name $"get svc ($ns_flag) -o json")
|
||
|
|
if $r.exit_code != 0 {
|
||
|
|
cli-err $"kubectl get svc failed: ($r.stderr | str trim)"
|
||
|
|
}
|
||
|
|
|
||
|
|
let items = (try { $r.stdout | from json | get -o items | default [] } catch { [] })
|
||
|
|
let rows = ($items | each {|svc|
|
||
|
|
let ns = ($svc.metadata.namespace? | default "")
|
||
|
|
let name = ($svc.metadata.name? | default "")
|
||
|
|
let svc_type = ($svc.spec.type? | default "ClusterIP")
|
||
|
|
let cip = ($svc.spec.clusterIP? | default "")
|
||
|
|
let ext_ips = ($svc.status.loadBalancer?.ingress? | default []
|
||
|
|
| each {|ing| $ing.ip? | default ($ing.hostname? | default "") }
|
||
|
|
| where { $in | is-not-empty } | str join ",")
|
||
|
|
let ports = ($svc.spec.ports? | default []
|
||
|
|
| each {|p| $"($p.port?):($p.protocol? | default 'TCP')" } | str join " ")
|
||
|
|
let has_https = ($svc.spec.ports? | default [] | any {|p|
|
||
|
|
($p.port? | default 0) == 443 or (($p.name? | default "") | str downcase | str contains "https")
|
||
|
|
})
|
||
|
|
{ namespace: $ns, name: $name, type: $svc_type, cluster_ip: $cip, external_ips: $ext_ips, ports: $ports, https: $has_https }
|
||
|
|
})
|
||
|
|
|
||
|
|
match $out {
|
||
|
|
"json" => { $rows | to json }
|
||
|
|
"yaml" => { $rows | to yaml }
|
||
|
|
_ => { $rows }
|
||
|
|
}
|
||
|
|
}
|
||
|
|
|
||
|
|
# Oldest last_sync across .servers-state.json and .cluster-state.json (= most stale data shown).
|
||
|
|
export def comp-cache-last-sync [ws_root: string, infra: string]: nothing -> string {
|
||
|
|
let srv_ts = do {
|
||
|
|
let p = ($ws_root | path join "infra" $infra ".servers-state.json")
|
||
|
|
if ($p | path exists) {
|
||
|
|
open $p | values | each { $in.last_sync? | default "" } | where { $in | is-not-empty }
|
||
|
|
} else { [] }
|
||
|
|
}
|
||
|
|
let cl_ts = do {
|
||
|
|
let p = (cluster-cache-path $ws_root $infra)
|
||
|
|
if ($p | path exists) {
|
||
|
|
let v = ((open $p).last_sync? | default "")
|
||
|
|
if ($v | is-not-empty) { [$v] } else { [] }
|
||
|
|
} else { [] }
|
||
|
|
}
|
||
|
|
let all = ($srv_ts | append $cl_ts) | where { $in | is-not-empty }
|
||
|
|
if ($all | is-empty) { "" } else { $all | sort | first }
|
||
|
|
}
|
||
|
|
|
||
|
|
# Single SSH session to CP: fetch pods + nodes as JSON in one connection.
|
||
|
|
# Returns {pods: list<record>, nodes: list<record>} — parsed from kubectl -o json output.
|
||
|
|
# Pods have metadata.namespace/name and status.phase/containerStatuses fields.
|
||
|
|
# Nodes have metadata.name and status.conditions fields.
|
||
|
|
export def comp-ssh-k8s-data [
|
||
|
|
ws_root: string,
|
||
|
|
infra: string,
|
||
|
|
server: string,
|
||
|
|
--systemd-services: list<string> = [], # service names to check via systemctl is-active
|
||
|
|
]: nothing -> record {
|
||
|
|
let empty = { pods: [], nodes: [], systemd: {} }
|
||
|
|
if ($server | is-empty) {
|
||
|
|
print --stderr "[live] no server resolved — cannot fetch k8s data"
|
||
|
|
return $empty
|
||
|
|
}
|
||
|
|
let info = (comp-server-ssh-info $ws_root $infra $server)
|
||
|
|
if (is-debug-level 2) {
|
||
|
|
let key_part = if ($info.key | is-not-empty) { $" -i ($info.key)" } else { "" }
|
||
|
|
print --stderr $"[live] ssh($key_part) ($info.user)@($info.host) kubectl get pods+nodes -o json"
|
||
|
|
}
|
||
|
|
# PATH: SSH non-interactive sessions may not have /usr/local/bin (where k0s and kubeadm kubectl live).
|
||
|
|
# KUBECONFIG: kubeadm clusters set this in ~/.bashrc (not sourced in BatchMode SSH).
|
||
|
|
# Probe /root/.kube/config then /etc/kubernetes/admin.conf; k0s kubectl ignores KUBECONFIG entirely.
|
||
|
|
let systemd_snippet = if ($systemd_services | is-not-empty) {
|
||
|
|
let checks = ($systemd_services | each {|svc| $"printf '($svc):'; systemctl is-active ($svc) 2>/dev/null || true"} | str join "; ")
|
||
|
|
$"echo '===SYSTEMD==='; ($checks)"
|
||
|
|
} else { "" }
|
||
|
|
let cmd_base = "export PATH=/usr/local/bin:/usr/bin:/bin:$PATH; for _kc in /root/.kube/config /etc/kubernetes/admin.conf; do [ -f \"$_kc\" ] && { export KUBECONFIG=\"$_kc\"; break; }; done; kc() { if command -v kubectl >/dev/null 2>&1; then kubectl \"$@\"; else k0s kubectl \"$@\"; fi; }; echo '===PODS==='; kc get pods --all-namespaces -o json 2>/dev/null || echo '{\"items\":[]}'; echo '===NODES==='; kc get nodes -o json 2>/dev/null || echo '{\"items\":[]}'"
|
||
|
|
let cmd = if ($systemd_snippet | is-not-empty) { $"($cmd_base); ($systemd_snippet)" } else { $cmd_base }
|
||
|
|
let result = (comp-ssh-run $info.user $info.key $info.host $cmd)
|
||
|
|
if (is-debug-level 2) {
|
||
|
|
print --stderr $"[live] SSH exit=($result.exit_code) stdout_len=($result.stdout | str length)"
|
||
|
|
}
|
||
|
|
if $result.exit_code != 0 { return $empty }
|
||
|
|
let all_lines = ($result.stdout | str replace --all "\r" "" | lines)
|
||
|
|
let pods_idx = ($all_lines | enumerate | where { ($in.item | str trim) == "===PODS===" } | get 0? | get -o index | default (-1))
|
||
|
|
let nodes_idx = ($all_lines | enumerate | where { ($in.item | str trim) == "===NODES===" } | get 0? | get -o index | default (-1))
|
||
|
|
let systemd_idx = ($all_lines | enumerate | where { ($in.item | str trim) == "===SYSTEMD===" } | get 0? | get -o index | default (-1))
|
||
|
|
if (is-debug-level 3) {
|
||
|
|
print --stderr $"[live] pods_idx=($pods_idx) nodes_idx=($nodes_idx) systemd_idx=($systemd_idx) total_lines=($all_lines | length)"
|
||
|
|
}
|
||
|
|
let pods = if $pods_idx >= 0 {
|
||
|
|
let from = ($pods_idx + 1)
|
||
|
|
let raw = if $nodes_idx > $pods_idx { $all_lines | skip $from | first ($nodes_idx - $from) } else { $all_lines | skip $from }
|
||
|
|
try { $raw | str join "\n" | from json | get -o items | default [] } catch {|e|
|
||
|
|
if (is-debug-level 1) {
|
||
|
|
print --stderr $"[live] pods JSON parse error: ($e.msg | str substring 0..120)"
|
||
|
|
print --stderr $"[live] pods first line: ($raw | get 0? | default '')"
|
||
|
|
}
|
||
|
|
[]
|
||
|
|
}
|
||
|
|
} else { [] }
|
||
|
|
let nodes = if $nodes_idx >= 0 {
|
||
|
|
let from = ($nodes_idx + 1)
|
||
|
|
let raw = if $systemd_idx > $nodes_idx { $all_lines | skip $from | first ($systemd_idx - $from) } else { $all_lines | skip $from }
|
||
|
|
try { $raw | str join "\n" | from json | get -o items | default [] } catch {|e|
|
||
|
|
if (is-debug-level 1) {
|
||
|
|
print --stderr $"[live] nodes JSON parse error: ($e.msg | str substring 0..120)"
|
||
|
|
}
|
||
|
|
[]
|
||
|
|
}
|
||
|
|
} else { [] }
|
||
|
|
let systemd = if $systemd_idx >= 0 {
|
||
|
|
$all_lines | skip ($systemd_idx + 1)
|
||
|
|
| each {|ln|
|
||
|
|
let parts = ($ln | str trim | split column ":" name status)
|
||
|
|
if ($parts | length) >= 1 and ($parts | get 0 | get name | is-not-empty) {
|
||
|
|
{ ($parts | get 0 | get name): ($parts | get 0 | get status | str trim) }
|
||
|
|
} else { {} }
|
||
|
|
}
|
||
|
|
| reduce -f {} {|it, acc| $acc | merge $it }
|
||
|
|
} else { {} }
|
||
|
|
if (is-debug-level 2) {
|
||
|
|
print --stderr $"[live] parsed pods=($pods | length) nodes=($nodes | length) systemd_keys=($systemd | columns | length)"
|
||
|
|
}
|
||
|
|
{ pods: $pods, nodes: $nodes, systemd: $systemd }
|
||
|
|
}
|
||
|
|
|
||
|
|
# Public API: return pod records for a server (used by describe-pod and external callers).
|
||
|
|
export def comp-ssh-pods [ws_root: string, infra: string, server: string]: nothing -> list {
|
||
|
|
(comp-ssh-k8s-data $ws_root $infra $server).pods
|
||
|
|
}
|
||
|
|
|
||
|
|
# Run kubectl describe pod -n <namespace> <pod> via SSH. Returns describe output as string.
|
||
|
|
export def comp-ssh-describe-pod [ws_root: string, infra: string, server: string, namespace: string, pod: string]: nothing -> string {
|
||
|
|
let info = (comp-server-ssh-info $ws_root $infra $server)
|
||
|
|
let cmd = $"if command -v kubectl >/dev/null 2>&1; then kubectl describe pod -n ($namespace) ($pod); else k0s kubectl describe pod -n ($namespace) ($pod); fi"
|
||
|
|
let result = (comp-ssh-run $info.user $info.key $info.host $cmd)
|
||
|
|
if $result.exit_code != 0 { $result.stderr | str trim } else { $result.stdout | str trim }
|
||
|
|
}
|
||
|
|
|
||
|
|
# Summarise pod records (from kubectl -o json) into {live, detail}.
|
||
|
|
def comp-kubectl-summarise [pods: list]: nothing -> record {
|
||
|
|
if ($pods | is-empty) { return { live: "deployed", detail: "no pods" } }
|
||
|
|
# Exclude Succeeded pods (completed jobs/migrations) — they never return to Running
|
||
|
|
# and would distort the ratio. Failed phase and error-waiting pods are kept and
|
||
|
|
# classified as degraded.
|
||
|
|
let active = ($pods | where { ($in | get -o status | default {} | get -o phase | default "") != "Succeeded" })
|
||
|
|
let total = ($active | length)
|
||
|
|
let running = ($active | where { ($in | get -o status | default {} | get -o phase | default "") == "Running" } | length)
|
||
|
|
let pending = ($active | where { ($in | get -o status | default {} | get -o phase | default "") == "Pending" } | length)
|
||
|
|
|
||
|
|
# Any pod in Failed phase OR any container waiting with an error reason → degraded.
|
||
|
|
let error_reasons = ["CrashLoopBackOff", "ImagePullBackOff", "ErrImagePull",
|
||
|
|
"Error", "OOMKilled", "CreateContainerError",
|
||
|
|
"CreateContainerConfigError", "RunContainerError",
|
||
|
|
"InvalidImageName", "PostStartHookError"]
|
||
|
|
let failing = ($active | where { |p|
|
||
|
|
let phase = ($p | get -o status | default {} | get -o phase | default "")
|
||
|
|
if $phase == "Failed" { true } else {
|
||
|
|
let cs = ($p | get -o status | default {} | get -o containerStatuses | default [])
|
||
|
|
let ics = ($p | get -o status | default {} | get -o initContainerStatuses | default [])
|
||
|
|
($cs | append $ics) | any { |c|
|
||
|
|
($c | get -o state | default {} | get -o waiting | default {} | get -o reason | default "") in $error_reasons
|
||
|
|
}
|
||
|
|
}
|
||
|
|
} | length)
|
||
|
|
|
||
|
|
let live = if $failing > 0 { "degraded"
|
||
|
|
} else if $pending > 0 { "starting"
|
||
|
|
} else if $total == 0 { "deployed"
|
||
|
|
} else if $running == $total { "healthy"
|
||
|
|
} else { "partial" }
|
||
|
|
|
||
|
|
let detail = if $failing > 0 {
|
||
|
|
$"($running)/($total) pods running ($failing) failing"
|
||
|
|
} else {
|
||
|
|
$"($running)/($total) pods running"
|
||
|
|
}
|
||
|
|
{ live: $live, detail: $detail }
|
||
|
|
}
|
||
|
|
|
||
|
|
# Filter pod records by namespace (exact match) and name substring.
|
||
|
|
def comp-filter-pods [namespace: string, search: string, pods: list]: nothing -> list {
|
||
|
|
let ns_filtered = if ($namespace | is-not-empty) {
|
||
|
|
$pods | where { ($in | get -o metadata | default {} | get -o namespace | default "") == $namespace }
|
||
|
|
} else { $pods }
|
||
|
|
$ns_filtered | where { ($in | get -o metadata | default {} | get -o name | default "") | str contains $search }
|
||
|
|
}
|
||
|
|
|
||
|
|
# Summarise node records (from kubectl -o json) into {live, detail}.
|
||
|
|
def comp-kubectl-nodes-summarise [nodes: list]: nothing -> record {
|
||
|
|
if ($nodes | is-empty) { return { live: "—", detail: "no node data" } }
|
||
|
|
let total = ($nodes | length)
|
||
|
|
let ready = ($nodes | where { |n|
|
||
|
|
$n | get -o status | default {} | get -o conditions | default []
|
||
|
|
| any { ($in | get -o type | default "") == "Ready" and ($in | get -o status | default "") == "True" }
|
||
|
|
} | length)
|
||
|
|
let live = if $ready == $total { "healthy"
|
||
|
|
} else if $ready == 0 { "degraded"
|
||
|
|
} else { "partial" }
|
||
|
|
{ live: $live, detail: $"($ready)/($total) nodes Ready" }
|
||
|
|
}
|
||
|
|
|
||
|
|
# Live check for a component using pre-fetched k8s_data {pods, nodes}.
|
||
|
|
# Resolves strategy from comp.live_check; falls back to namespace/pod_selector on the component.
|
||
|
|
# 'systemd strategy always returns "—" in live (fast) mode — requires prvng c health.
|
||
|
|
# Returns {live: string, detail: string}.
|
||
|
|
export def comp-live-check [name: string, comp: record, k8s_data: record]: nothing -> record {
|
||
|
|
let lc = ($comp | get -o live_check | default {})
|
||
|
|
let strategy = ($lc | get -o strategy | default "none" | into string)
|
||
|
|
match $strategy {
|
||
|
|
"k8s_pods" => {
|
||
|
|
if ($k8s_data.pods | is-empty) { return { live: "—", detail: "no pod data" } }
|
||
|
|
let ns = do {
|
||
|
|
let explicit = ($lc | get -o namespace | default "")
|
||
|
|
if ($explicit | is-not-empty) { $explicit } else { $comp | get -o namespace | default "" }
|
||
|
|
}
|
||
|
|
let sel = do {
|
||
|
|
let explicit = ($lc | get -o selector | default "")
|
||
|
|
if ($explicit | is-not-empty) { $explicit } else {
|
||
|
|
let ps = ($comp | get -o pod_selector | default "")
|
||
|
|
if ($ps | is-not-empty) { $ps } else { $name | str replace --all "_" "-" }
|
||
|
|
}
|
||
|
|
}
|
||
|
|
let matching = (comp-filter-pods $ns $sel $k8s_data.pods)
|
||
|
|
if ($matching | is-empty) { return { live: "—", detail: $"no pods: ($sel)" } }
|
||
|
|
comp-kubectl-summarise $matching
|
||
|
|
}
|
||
|
|
"k8s_nodes" => {
|
||
|
|
if ($k8s_data.nodes | is-empty) { return { live: "—", detail: "no node data" } }
|
||
|
|
let sel = ($lc | get -o selector | default "")
|
||
|
|
let filtered = if ($sel | is-not-empty) {
|
||
|
|
$k8s_data.nodes | where { ($in | get -o metadata | default {} | get -o name | default "") | str contains $sel }
|
||
|
|
} else {
|
||
|
|
$k8s_data.nodes
|
||
|
|
}
|
||
|
|
comp-kubectl-nodes-summarise $filtered
|
||
|
|
}
|
||
|
|
"k8s_api" => {
|
||
|
|
# apiserver healthy if kubectl returned any node data
|
||
|
|
if ($k8s_data.nodes | is-not-empty) or ($k8s_data.pods | is-not-empty) {
|
||
|
|
{ live: "healthy", detail: "apiserver ok" }
|
||
|
|
} else {
|
||
|
|
{ live: "—", detail: "apiserver unreachable" }
|
||
|
|
}
|
||
|
|
}
|
||
|
|
"systemd" => {
|
||
|
|
let svc = ($lc | get -o service | default "")
|
||
|
|
if ($svc | is-empty) { return { live: "—", detail: "health: systemd" } }
|
||
|
|
let units = ($k8s_data | get -o systemd | default {})
|
||
|
|
if ($units | is-empty) { return { live: "—", detail: "health: systemd" } }
|
||
|
|
let active = ($units | get -o $svc | default "")
|
||
|
|
match $active {
|
||
|
|
"active" => { live: "healthy", detail: $"systemd: ($svc) active" },
|
||
|
|
"inactive" => { live: "degraded", detail: $"systemd: ($svc) inactive" },
|
||
|
|
"failed" => { live: "degraded", detail: $"systemd: ($svc) failed" },
|
||
|
|
_ => { live: "—", detail: $"systemd: ($svc) ($active)" },
|
||
|
|
}
|
||
|
|
}
|
||
|
|
"on_demand" => { live: "—", detail: "on-demand" }
|
||
|
|
_ => { live: "—", detail: "no live check" }
|
||
|
|
}
|
||
|
|
}
|
||
|
|
|
||
|
|
# Extract pod summary records for a component from pre-fetched pod JSON records.
|
||
|
|
# Returns list of {namespace, pod, ready, status}.
|
||
|
|
export def comp-pods-for [namespace: string, name: string, pods: list, pod_selector: string = ""]: nothing -> list {
|
||
|
|
let search = if ($pod_selector | is-not-empty) { $pod_selector } else { $name | str replace --all "_" "-" }
|
||
|
|
(comp-filter-pods $namespace $search $pods)
|
||
|
|
| each {|p|
|
||
|
|
let meta = ($p | get -o metadata | default {})
|
||
|
|
let status = ($p | get -o status | default {})
|
||
|
|
let cs = ($status | get -o containerStatuses | default [])
|
||
|
|
let ready = ($cs | where { $in.ready? | default false } | length)
|
||
|
|
{
|
||
|
|
namespace: ($meta | get -o namespace | default ""),
|
||
|
|
pod: ($meta | get -o name | default ""),
|
||
|
|
ready: $"($ready)/($cs | length)",
|
||
|
|
status: ($status | get -o phase | default ""),
|
||
|
|
}
|
||
|
|
}
|
||
|
|
}
|
||
|
|
|