A scheduled step may hold its module's own containers still (hq ADR 0189)

while-stopped names resource ids of the same module's containers; the host
stops them before the run and starts them again after it, in reverse order,
whatever the step did. The restart is deferred before the first stop and runs
on its own context, because the one real risk of this field is a window that
never closes.

Scheduled steps only: at apply the declaration is applied in order and a
run-once step already gates what follows.
This commit is contained in:
2026-10-04 02:34:33 +02:00
parent 8c2e76f4c1
commit e8c4824ae2
3 changed files with 350 additions and 1 deletions
+68
View File
@@ -1037,6 +1037,24 @@ type Container struct {
// rather than stacked. It is exclusive with RunOnce and with restart-on: a container runs once
// and gates, runs on a cadence, or stays up — never two of these.
Schedule string `json:"schedule,omitempty"`
// WhileStopped names resources of the same module — containers — that must be held still for
// the duration of this step's run (novox/hq ADR 0189). The host stops each before the run and
// starts each again after it, **whatever the step did**: a step that failed must leave the
// service running, because the one real risk of this field is a window that never closes.
//
// **For the work a service cannot have done underneath it.** The artifact store's collector
// walks the storage and requires every writer stopped; a run-once step runs beside containers
// and a scheduled one is the same container again, so until this there was no way for a module
// to say it. The predecessor said it with a shell script, which is how the mesh inherited a
// store that has never collected anything.
//
// **Its own module's containers, and only on a schedule.** A module that could quiesce a
// neighbour could stop the mesh. And at apply time the host already has a window — the
// declaration is applied in order and a run-once step gates what follows — so a one-time
// offline job says *before*, not *instead of*; a recurring window is the case order cannot
// express, and the only one this serves.
WhileStopped []string `json:"while-stopped,omitempty"`
}
func (c *Container) Identity() string { return c.ID }
@@ -1068,6 +1086,20 @@ func (c *Container) validate(where string, _ bool) []string {
problems = append(problems, where+": "+err.Error())
}
}
// A maintenance window belongs to a recurring step (novox/hq ADR 0189). Refused on anything
// else here, where the field is; that it names containers of the same module, and not itself,
// is judged against the whole declaration (see whileStoppedNames).
if len(c.WhileStopped) > 0 && c.Schedule == "" {
problems = append(problems, where+": while-stopped needs a schedule; at apply the host "+
"already has a window — the declaration is applied in order and a run-once step gates "+
"what follows — so a one-time offline job is declared before what it works on")
}
for _, id := range c.WhileStopped {
if id == c.ID {
problems = append(problems, where+": while-stopped names "+strconv.Quote(id)+
", which is this step itself")
}
}
// The runtime's flags take addresses, and a name here would be handed to it verbatim and
// refused at create — after the old container was already removed. Refused on arrival instead.
for _, d := range c.Dns {
@@ -1482,6 +1514,7 @@ func parse(raw []byte, allowActions bool) (*Declaration, error) {
problems = append(problems, resource.validate(where, allowActions)...)
d.Resources = append(d.Resources, resource)
}
problems = append(problems, checkWhileStopped(d.Resources)...)
problems = append(problems, checkAdoption(env.Adoption, d.Resources, allowActions)...)
problems = append(problems, checkReplaces(d.Resources)...)
if env.Adoption == nil {
@@ -1511,6 +1544,41 @@ func parse(raw []byte, allowActions bool) (*Declaration, error) {
return d, nil
}
// checkWhileStopped judges a maintenance window against the whole declaration (novox/hq ADR 0189).
//
// A step may hold still only a container that is **here** — in this same declaration, which is to
// say on this machine and placed by the mesh. That is what makes it the module's own: a node's
// declaration carries one module's resources beside another's, so the id must also be a container
// and not a file or a directory, which there would be nothing to stop.
//
// Refused on arrival rather than discovered at the first fire. A window that names something the
// host cannot stop is a window that opens at 03:00 and reports nothing until somebody reads a log.
func checkWhileStopped(resources []Resource) []string {
containers := map[string]bool{}
for _, r := range resources {
if r.Kind() == TypeContainer {
containers[r.Identity()] = true
}
}
var problems []string
for _, r := range resources {
c, ok := r.(*Container)
if !ok {
continue
}
for _, id := range c.WhileStopped {
if containers[id] {
continue
}
problems = append(problems, fmt.Sprintf(
"resource %q: while-stopped names %q, and this declaration has no container by "+
"that id. A step may hold still only a container placed on this machine "+
"beside it", c.ID, id))
}
}
return problems
}
func strictDecode(raw []byte, into any) error {
// DisallowUnknownFields is the whole point rather than strictness for its own sake: a
// field the host does not know is a thing the control plane believes it asked for.