v0.90.0 — guest RAM resize (R-24) + fast-tick-until-convergence (R-28)

MinAgent coupling: felhom-controller v0.143.0 gates its guest-memory-resize UI on
this agent (FeatureGuestMemoryResize, MinAgent 0.90.0).

R-24 guest RAM resize (internal/localapi/guestmemory.go): self-scoped GET/POST
/guest/memory. Agent enforces every bound FRESH per request (min 2048, max
host_total-2048, shrink floor max(2048, usage+512)); applies via PVE SetConfig —
live cgroup apply, no reboot (Phase-0 proven on the nested demo box). Verify-after-apply
re-reads maxmem before claiming success. New narrow MemoryOps seam (GuestAPI untouched);
Options.Memory nil -> 503. Memory only.

R-28 fast-tick (internal/fasttick): while any desired-state item is unapplied -
including the pre-tunnel window a hub poke can't reach - pulse the shared out-of-band
trigger every 30s, self-disarm on convergence. Four cached sources (desired-gen==0,
reconcile Planned-Pending>0, pbsdr waiting_secret only, wgtunnel desired-not-operational);
LOUD pbsdr states + pending_signature excluded. Seams: reconcile.Engine.LastResult() +
wgtunnel.Manager.TunnelConvergence() (cached, no per-tick exec).

Guests-0/0: hypothesis REFUTED live (9201 IS a pool member; 0/0 was the pre-provision
window; PoolAddVMID re-assert already covers restore-over-existing). No code change; the
fast-tick mitigates the window.

Tests + red-proofs (i floor guard, ii max guard, iii always-pulse) all restored green.
This commit is contained in:
2026-07-17 19:09:40 +02:00
parent 9127f547f9
commit ac112c956e
10 changed files with 958 additions and 1 deletions
+259
View File
@@ -0,0 +1,259 @@
package localapi
import (
"context"
"fmt"
"net/http"
"gitea.dooplex.hu/admin/felhom-agent/internal/proxmox"
)
// Guest RAM resize (v0.90.0, R-24 controller-direct). The customer sees the guest's current memory
// on the controller's system-settings page and resizes it; the controller calls these self-scoped
// endpoints; the AGENT enforces every bound FRESH per request (the UI's numbers are decoration) and
// applies via the PVE API's SetConfig — a live cgroup apply, no reboot (Phase-0 proved it: maxmem
// moves with the guest running, /proc/meminfo ripples via lxcfs). Memory only; cores stay observation.
//
// Ruled bounds (Viktor, 2026-07-17):
// - min = minGuestMemoryMB (2048)
// - max = host_total hostReserveMB (2048 reserved for the host)
// - shrink floor = max(minGuestMemoryMB, current_usage + shrinkUsageMarginMB) — a customer wanting
// less must stop applications first (the refusal says so, in the controller's Hungarian).
//
// The floor gates SHRINK only; a grow is bounded by min/max alone.
const (
// minGuestMemoryMB is the ruled floor for any guest allocation (a Felhom box needs headroom).
minGuestMemoryMB int64 = 2048
// hostReserveMB is held back from the host total so a resize can never starve the hypervisor.
hostReserveMB int64 = 2048
// shrinkUsageMarginMB is the safety gap kept above live usage on a SHRINK (the "never below
// current usage" ruling, with headroom). One named constant — trivially re-ruled.
shrinkUsageMarginMB int64 = 512
// mib is one Proxmox "memory" unit (MiB) in bytes. PVE config `memory` is MiB; status/node
// memory fields are bytes — convert at this boundary (the §8 units trap).
mib int64 = 1 << 20
)
// MemoryOps is the guest-RAM-resize Proxmox surface (v0.90.0). Satisfied by *proxmox.Client. Kept
// separate from GuestAPI so adding it breaks no existing fake. Every method is invoked ONLY with the
// token-resolved VMID (never a caller-supplied id).
type MemoryOps interface {
GuestStatus(ctx context.Context, vmid int) (proxmox.Guest, error)
GuestConfig(ctx context.Context, vmid int) (proxmox.GuestConfig, error)
NodeStatus(ctx context.Context) (proxmox.NodeStatus, error)
SetConfig(ctx context.Context, vmid int, params map[string]string) (string, error)
WaitTask(ctx context.Context, upid string, opts proxmox.WaitOptions) (proxmox.TaskStatus, error)
}
// MemoryInfo is GET /guest/memory — every field in MB, computed agent-side (Scenario C). min/max/floor
// are the CURRENT enforced bounds, so a stale UI re-renders honestly on every read/refusal.
type MemoryInfo struct {
VMID int `json:"vmid"`
AllocatedMB int64 `json:"allocated_mb"`
UsageMB int64 `json:"usage_mb"`
HostTotalMB int64 `json:"host_total_mb"`
MinMB int64 `json:"min_mb"`
MaxMB int64 `json:"max_mb"`
FloorMB int64 `json:"floor_mb"`
Running bool `json:"running"`
}
// memoryResizeRequest is the POST /guest/memory body.
type memoryResizeRequest struct {
VMID int `json:"vmid,omitempty"` // optional; if set must equal the token's guest (self-scope)
MemoryMB int64 `json:"memory_mb"`
}
// memoryRefusal is the data field of a 412 refusal: the machine code the controller maps to Hungarian,
// plus the fresh bounds so the UI re-renders without a second round-trip.
type memoryRefusal struct {
Code string `json:"code"` // below_min | above_max | below_usage_floor
AllocatedMB int64 `json:"allocated_mb"`
UsageMB int64 `json:"usage_mb"`
HostTotalMB int64 `json:"host_total_mb"`
MinMB int64 `json:"min_mb"`
MaxMB int64 `json:"max_mb"`
FloorMB int64 `json:"floor_mb"`
}
// memoryResizeResult is the success data field.
type memoryResizeResult struct {
VMID int `json:"vmid"`
OldMB int64 `json:"old_mb"`
NewMB int64 `json:"new_mb"`
Unchanged bool `json:"unchanged"`
}
// memoryBounds is one fresh snapshot of everything the validation needs.
type memoryBounds struct {
allocatedMB int64
usageMB int64
hostTotalMB int64
minMB int64
maxMB int64
floorMB int64
running bool
}
func (b memoryBounds) refusal(code string) memoryRefusal {
return memoryRefusal{
Code: code, AllocatedMB: b.allocatedMB, UsageMB: b.usageMB, HostTotalMB: b.hostTotalMB,
MinMB: b.minMB, MaxMB: b.maxMB, FloorMB: b.floorMB,
}
}
func (b memoryBounds) info(vmid int) MemoryInfo {
return MemoryInfo{
VMID: vmid, AllocatedMB: b.allocatedMB, UsageMB: b.usageMB, HostTotalMB: b.hostTotalMB,
MinMB: b.minMB, MaxMB: b.maxMB, FloorMB: b.floorMB, Running: b.running,
}
}
// bytesToMBUp converts bytes → MB rounding UP (usage must never be under-reported for the floor).
func bytesToMBUp(b int64) int64 {
if b <= 0 {
return 0
}
return (b + mib - 1) / mib
}
// readMemoryBounds computes the enforced bounds from a FRESH read of guest config/status + host total.
// Called by BOTH the GET and the POST so a stale UI can never smuggle an old max/floor.
func (s *Server) readMemoryBounds(ctx context.Context, vmid int) (memoryBounds, error) {
cfg, err := s.mem.GuestConfig(ctx, vmid)
if err != nil {
return memoryBounds{}, fmt.Errorf("guest config: %w", err)
}
st, err := s.mem.GuestStatus(ctx, vmid)
if err != nil {
return memoryBounds{}, fmt.Errorf("guest status: %w", err)
}
node, err := s.mem.NodeStatus(ctx)
if err != nil {
return memoryBounds{}, fmt.Errorf("node status: %w", err)
}
b := memoryBounds{
allocatedMB: cfg.Memory, // PVE config memory is already MB
usageMB: bytesToMBUp(st.Mem), // bytes → MB, rounded up
hostTotalMB: node.Memory.Total / mib, // bytes → MB, floor (conservative for max)
minMB: minGuestMemoryMB,
running: st.Status == "running",
}
b.maxMB = b.hostTotalMB - hostReserveMB
b.floorMB = b.usageMB + shrinkUsageMarginMB
if b.floorMB < minGuestMemoryMB {
b.floorMB = minGuestMemoryMB
}
return b, nil
}
// handleGuestMemory serves GET /guest/memory — the current allocation, live usage, and the enforced
// bounds, all agent-computed (Scenario C).
func (s *Server) handleGuestMemory(w http.ResponseWriter, r *http.Request, vmid int) {
if s.mem == nil {
writeErr(w, http.StatusServiceUnavailable, "guest memory resize not configured on this host")
return
}
b, err := s.readMemoryBounds(r.Context(), vmid)
if err != nil {
s.logger.Error("local-api: guest-memory read", "vmid", vmid, "err", err)
writeErr(w, http.StatusBadGateway, "could not read guest memory: "+err.Error())
return
}
writeOK(w, b.info(vmid))
}
// handleGuestMemoryResize serves POST /guest/memory — validate FRESH against the ruled bounds (the UI's
// numbers are decoration), then apply via SetConfig (live cgroup apply) and verify the new maxmem before
// claiming success. Single-flight (one customer per host). SetConfig is NEVER called on a refusal path.
func (s *Server) handleGuestMemoryResize(w http.ResponseWriter, r *http.Request, vmid int) {
if s.mem == nil {
writeErr(w, http.StatusServiceUnavailable, "guest memory resize not configured on this host")
return
}
var req memoryResizeRequest
if !decodeBody(w, r, &req) {
return
}
if !s.scopedFromBody(w, req.VMID, vmid, r.URL.Path) {
return
}
if req.MemoryMB <= 0 {
writeErr(w, http.StatusBadRequest, "memory_mb must be a positive integer")
return
}
// Single-flight: one customer per host; last-write-wins at PVE is not a UX we want.
s.memMu.Lock()
defer s.memMu.Unlock()
b, err := s.readMemoryBounds(r.Context(), vmid)
if err != nil {
s.logger.Error("local-api: guest-memory resize precheck", "vmid", vmid, "err", err)
writeErr(w, http.StatusBadGateway, "could not read guest memory: "+err.Error())
return
}
target := req.MemoryMB
// No-op: target == current allocation → success, no SetConfig call.
if target == b.allocatedMB {
s.logger.Info("local-api: guest-memory resize no-op (unchanged)", "vmid", vmid, "memory_mb", target)
writeOK(w, memoryResizeResult{VMID: vmid, OldMB: b.allocatedMB, NewMB: b.allocatedMB, Unchanged: true})
return
}
// Ruled refusals — SetConfig MUST NOT run on any of these.
switch {
case target < b.minMB:
s.logger.Warn("local-api: guest-memory resize refused (below_min)", "vmid", vmid, "target_mb", target, "min_mb", b.minMB)
writeStatus(w, http.StatusPreconditionFailed, false, b.refusal("below_min"),
fmt.Sprintf("requested %d MB is below the minimum %d MB", target, b.minMB))
return
case target > b.maxMB:
s.logger.Warn("local-api: guest-memory resize refused (above_max)", "vmid", vmid, "target_mb", target, "max_mb", b.maxMB)
writeStatus(w, http.StatusPreconditionFailed, false, b.refusal("above_max"),
fmt.Sprintf("requested %d MB exceeds the maximum %d MB (host reserve %d MB)", target, b.maxMB, hostReserveMB))
return
case target < b.allocatedMB && target < b.floorMB:
// SHRINK below the usage floor — the customer must stop applications first.
s.logger.Warn("local-api: guest-memory resize refused (below_usage_floor)",
"vmid", vmid, "target_mb", target, "usage_mb", b.usageMB, "floor_mb", b.floorMB)
writeStatus(w, http.StatusPreconditionFailed, false, b.refusal("below_usage_floor"),
fmt.Sprintf("requested %d MB is too close to current usage %d MB (floor %d MB) — stop applications first", target, b.usageMB, b.floorMB))
return
}
// Apply. SetConfig may be synchronous (empty UPID) or return a task to wait on.
upid, err := s.mem.SetConfig(r.Context(), vmid, map[string]string{"memory": fmt.Sprintf("%d", target)})
if err != nil {
s.logger.Error("local-api: guest-memory SetConfig", "vmid", vmid, "target_mb", target, "err", err)
writeErr(w, http.StatusBadGateway, "memory resize failed: "+err.Error())
return
}
if upid != "" {
if _, err := s.mem.WaitTask(r.Context(), upid, proxmox.WaitOptions{}); err != nil {
s.logger.Error("local-api: guest-memory WaitTask", "vmid", vmid, "upid", upid, "err", err)
writeErr(w, http.StatusBadGateway, "memory resize task failed: "+err.Error())
return
}
}
// Verify: re-read and confirm maxmem reflects the target before claiming success (never trust the
// POST 200 — the config could have been rejected at task execution, or applied as pending).
st, err := s.mem.GuestStatus(r.Context(), vmid)
if err != nil {
s.logger.Error("local-api: guest-memory post-apply status", "vmid", vmid, "err", err)
writeErr(w, http.StatusBadGateway, "resize applied but verification read failed: "+err.Error())
return
}
if st.MaxMem != target*mib {
s.logger.Error("local-api: guest-memory resize NOT reflected after apply",
"vmid", vmid, "target_mb", target, "observed_maxmem_bytes", st.MaxMem)
writeErr(w, http.StatusBadGateway,
fmt.Sprintf("resize did not take effect (guest still reports %d MB) — a reboot may be required", st.MaxMem/mib))
return
}
s.logger.Info("local-api: guest-memory resized", "vmid", vmid, "old_mb", b.allocatedMB, "new_mb", target)
writeOK(w, memoryResizeResult{VMID: vmid, OldMB: b.allocatedMB, NewMB: target})
}