Files
mesh-controller/internal/broker/composed.go
T
jschoubben bfb271a91b Take done keys in from a durable pull consumer, a batch at a time (issue 1945, ADR 0544)
The take-in watched the builder's and the indexer's buckets, started a follower
per module that retried on a 30 s timer, and at start took every catalogued
module in alphabetically: on 2026-10-25 a restart took in about 2 modules a
minute, about 190 followers queued behind 16 slots, and every take-in composed
two or three times even with nothing new, so a new build waited 30 minutes and
more for its credential.

Now two durable consumers named mesh-controller-take-in read the indexer's
modules bucket (every module's last trunk value when the consumer is made, then
each write) and the builder's builds bucket (each write), at most 16 keys
unacknowledged each. The bus keeps what is not taken in: the start's look is the
consumers' own catch-up, and a restart costs only what was left unacknowledged.
A batch reads the catalogue once, takes each module in once, and asks the
composer twice: once for its modules' nodes, once for the bus's node with the
grants. A module registered at its trunk whose compositions and grant exist, or
that this process took in at the same trunk, composes nothing; a key delivered
again is taken in whole. A wait for the builder's or indexer's next write is
acknowledged, a wait on a read or a failure refused with a 30 s delay. The
backlog is in status, and a backlog standing still for 5 minutes raises
take-in-stalled. The controller's account is granted the two consumers' acks.

The follower tests go with the followers: twelve tests become three.
Research 072: the take-ins on a bounded pull consumer (milestone 4, done keys
alone) and the start's look composing only what changed (milestone 2).
2026-10-25 17:59:20 +01:00

193 lines
8.8 KiB
Go

package broker
import (
"context"
"fmt"
"git.novox.be/novox/mesh-sdk/go/delivery"
"github.com/nats-io/nats.go/jetstream"
)
// The compositions' buckets (novox/hq ADR 0539 clauses 2 and 4): one per node, `composed-<node>`, holding that node's
// compositions, which only that node's node-engine reads, since a composition can carry a sealed value (ADR 0488 §2's
// reason); and `mesh-controller_composed`, holding each composition's digest and the judgement of its move, which the
// assigner reads. The controller writes both, by compare-and-set while it holds the lease; each keeps five revisions
// of a key. The names and layouts are mesh-sdk's (delivery.ComposedBucketFor, ControllerComposedBucket, ComposedKey).
// ControllerComposed is the local name of the compositions' digests, as a reader names it in `reads` after
// `mesh-controller.`: delivery.ControllerComposedBucket.
const ControllerComposed = "composed"
// ComposedValueBytes is the most a composition may hold: a file over 64 KiB travels by digest, so a module's
// composition is far smaller; just under the server's largest message, which also carries the subject and headers.
const ComposedValueBytes = 1<<20 - 8<<10
// ComposedBucketBytes is a node's composition bucket's size: five revisions of every module's compositions of the node.
const ComposedBucketBytes = 256 << 20
// ComposedDigestsBytes is the size of the bucket of digests: a value of a few hundred bytes per node, module and
// fingerprint kept.
const ComposedDigestsBytes = 64 << 20
// ComposedConfig is a node's composition bucket as the controller asserts it.
func ComposedConfig(node string) (jetstream.KeyValueConfig, error) {
bucket, err := delivery.ComposedBucketFor(node)
if err != nil {
return jetstream.KeyValueConfig{}, err
}
return jetstream.KeyValueConfig{
Bucket: bucket,
Description: "the compositions of " + node + " (novox/hq ADR 0539): each module's resources for this node at one " +
"fingerprint, and the node's own, signed; written by the controller alone, read by this node's node-engine " +
"alone; deleted when outside the module's five newest fingerprints and named by no want",
History: delivery.ComposedHistory,
MaxValueSize: ComposedValueBytes,
MaxBytes: ComposedBucketBytes,
Storage: jetstream.FileStorage,
}, nil
}
// ComposedDigestsConfig is the bucket of the compositions' digests as the controller asserts it.
func ComposedDigestsConfig() jetstream.KeyValueConfig {
return jetstream.KeyValueConfig{
Bucket: delivery.ControllerComposedBucket,
Description: "each composition's digest and the judgement of its move from what the node runs (novox/hq ADR 0539 " +
"clauses 2 and 14): written by the controller alone, read by the assigner",
History: delivery.ControllerComposedHistory,
MaxValueSize: 4 << 10,
MaxBytes: ComposedDigestsBytes,
Storage: jetstream.FileStorage,
}
}
// EnsureComposed creates the bucket of digests and each named node's composition bucket if absent and brings their
// options to match. An update, never a delete.
func EnsureComposed(ctx context.Context, js jetstream.JetStream, nodes ...string) error {
if _, err := js.CreateOrUpdateKeyValue(ctx, ComposedDigestsConfig()); err != nil {
return fmt.Errorf("asserting bucket %s: %w", delivery.ControllerComposedBucket, err)
}
for _, node := range nodes {
cfg, err := ComposedConfig(node)
if err != nil {
return err
}
if _, err := js.CreateOrUpdateKeyValue(ctx, cfg); err != nil {
return fmt.Errorf("asserting bucket %s: %w", cfg.Bucket, err)
}
}
return nil
}
// ComposedWrites is what the controller publishes to write a node's compositions: the bucket's own subjects.
func ComposedWrites(node string) (string, error) {
bucket, err := delivery.ComposedBucketFor(node)
if err != nil {
return "", err
}
return "$KV." + bucket + ".>", nil
}
// ComposedReads is what a node's node-engine publishes to follow its wants and compositions (ADR 0539 clauses 1, 2 and
// 9): every key of its own composition bucket, and its own wants (`want.<node>.<module>` and `want.<node>.node`) in the
// assigner's bucket; nothing of another node's, nor the bucket of digests.
func ComposedReads(node string) ([]string, error) {
bucket, err := delivery.ComposedBucketFor(node)
if err != nil {
return nil, err
}
if !safeSubject.MatchString(node) {
return nil, fmt.Errorf("%q cannot be part of a subject: a permission is a subject pattern, and this would widen it", node)
}
return append(readGrants(bucket, ""), readGrants(delivery.AssignerWantsBucket, "want."+node+".>")...), nil
}
// VerbsTheNodeAgentAsks are the seat verbs a node's node-engine calls of its own machine while it applies a want (ADR
// 0539 clause 16, ADR 0429 §14): its backup holder backs a data module up before it moves, and says when it was. Asked
// of its own machine only, `mesh.seat.<seat>.tool.<verb>.<node>`.
var VerbsTheNodeAgentAsks = []SeatVerb{{Seat: "node-backup", Verb: "now"}, {Seat: "node-backup", Verb: "backed-up"}}
// NodeAgentAsks are the subjects of VerbsTheNodeAgentAsks for one node.
func NodeAgentAsks(node string) ([]string, error) {
if !safeSubject.MatchString(node) {
return nil, fmt.Errorf("%q cannot be part of a subject: a permission is a subject pattern, and this would widen it", node)
}
var out []string
for _, v := range VerbsTheNodeAgentAsks {
out = append(out, "mesh.seat."+v.Seat+".tool."+v.Verb+"."+node)
}
return out, nil
}
// NodeAgentHears are the events a node's node-engine subscribes while it applies a want: the backup holder's word that a
// backup is taken (`backed-up`), from whichever module holds node-backup on its machine (restic today).
var NodeAgentHears = []string{"mesh.mod." + BackupHolderModule + ".event.backed-up"}
// BackupHolderModule is the module that holds node-backup on every node, restic: the one whose word a node-engine hears.
// A second holder of the seat is a change of this grant.
const BackupHolderModule = "restic"
// administers says a person may invoke every tool (`*`): the operator.
func administers(invokes []string) bool {
for _, i := range invokes {
if i == "*" {
return true
}
}
return false
}
// HandPathWrites is what the operator's account publishes for release 6's hand path (ADR 0539 clauses 2 and 10): a want
// of any node, written and read back, in the assigner's bucket; each named node's compositions; the digests; and each
// node's object bucket for a file over 64 KiB, whose put reads an object's metadata and purges a replaced one's chunks.
func HandPathWrites(nodes []string) ([]string, error) {
out := append([]string{"$KV." + delivery.AssignerWantsBucket + ".want.>"}, readGrants(delivery.AssignerWantsBucket, "want.>")...)
out = append(out, "$KV."+delivery.ControllerComposedBucket+".>")
out = append(out, readGrants(delivery.ControllerComposedBucket, "")...)
for _, node := range nodes {
w, err := ComposedWrites(node)
if err != nil {
return nil, err
}
bucket, _ := delivery.ComposedBucketFor(node)
out = append(out, w)
out = append(out, readGrants(bucket, "")...)
files, err := DeclarationFilesWrites(node)
if err != nil {
return nil, err
}
stream := "OBJ_" + DeclarationFilesBucket + node
out = append(out, files, "$JS.API.STREAM.INFO."+stream, "$JS.API.DIRECT.GET."+stream+".>",
"$JS.API.STREAM.MSG.GET."+stream, "$JS.API.STREAM.PURGE."+stream)
}
return out, nil
}
// DeclarationsStream is the stream of each node's last declaration, `mesh.node.<node>.declare` (issue 107). Release 6
// sends none (ADR 0539 clauses 5 and 11): the stream is kept, with each node's consumer, only so a node away during the
// switch finds its last declaration and with it the new node-engine; the controller deletes it once every node's have
// names the new node-engine, and from then on neither asserts it nor any node's consumer on it; nor makes it again once
// it is gone from the bus for any reason, since a stream made anew would hold no node's last declaration (issue 1945).
const DeclarationsStream = "NODES"
// RaiseAfterDeclarations is Raise once the declarations' stream is retired: the mesh's streams and consumers but the
// declarations' stream, and no node's consumer of a declaration.
func RaiseAfterDeclarations(r Raiser) error {
for _, s := range MeshStreams() {
if s.Name == DeclarationsStream {
continue
}
if err := r.EnsureStream(s); err != nil {
return fmt.Errorf("asserting stream %s: %w", s.Name, err)
}
}
return AssertMeshConsumers(r)
}
// TakeInConsumer is the durable consumer the controller takes done keys in from, on the stream of each of TakeInBuckets
// (novox/hq ADR 0544 §1).
const TakeInConsumer = "mesh-controller-take-in"
// TakeInBuckets are the buckets the take-in consumes: the indexer's modules (each module's trunk value) and the builder's
// builds (each build's key).
var TakeInBuckets = []string{delivery.ModulesBucket, delivery.BuilderBuildsBucket}