mesh-controller/internal/inventory/chain_links.go

package inventory

import (
	"context"
	"errors"
	"fmt"
	"os"
	"sort"
	"time"

	"github.com/jackc/pgx/v5"

	"github.com/novox/mesh-controller/internal/chain"
)

// The links of the chain (novox/hq ADR 0408 §1, design 47 acceptance 10, issue 1095): a send that puts a build on a machine
// writes, in the assignment of the module there, the build, what it published, its commit and the rollout that sent it, and
// since when; the rollout lists the assignments it changed with the build before and after (the controller's send, in
// release_plan.go). One read of an assignment then names the commit, the build and the rollout that put it there, and one read
// of the rollout lists its assignments.

type rolloutKey struct{}

// RolloutModule is what a rollout sends of one module: the commit and the build of it.
type RolloutModule struct{ Commit, Build string }

type rolloutOf struct {
	id      string
	modules map[string]RolloutModule
}

// WithRollout is a context for a send a rollout makes: what RecordSent writes in the assignments of the modules it carries
// names this rollout and the build it sent. A module the send carries beside these, that another walk is moving, is named
// by the build the mesh holds of its commit and by no rollout.
func WithRollout(ctx context.Context, id string, modules map[string]RolloutModule) context.Context {
	return context.WithValue(ctx, rolloutKey{}, rolloutOf{id: id, modules: modules})
}

// AssignmentRuns is what each module assigned to a machine runs there, as its assignment names it: the build and the
// commit. Empty when the assignments are not on the bus, and for a module not sent yet.
func (i *Inventory) AssignmentRuns(ctx context.Context, node string) (map[string]chain.Ref, error) {
	c, on, err := i.ChainReads(ctx, KindAssignments)
	if err != nil || !on {
		return nil, err
	}
	all, err := c.Assignments.All(ctx)
	if err != nil {
		return nil, err
	}
	out := map[string]chain.Ref{}
	for _, k := range all {
		a := k.Value
		if a.Node == node && !a.Assigned.IsZero() && a.Stands() && (a.Build != "" || a.Commit != "") {
			out[a.Module] = chain.Ref{Build: a.Build, Commit: a.Commit}
		}
	}
	return out, nil
}

// AssignmentsOn are the assignments of a machine as the chain names them, by module: the standing ones, with their links.
func (i *Inventory) AssignmentsOn(ctx context.Context, node string) ([]chain.Assignment, error) {
	c, on, err := i.ChainReads(ctx, KindAssignments)
	if err != nil || !on {
		return nil, err
	}
	all, err := c.Assignments.All(ctx)
	if err != nil {
		return nil, err
	}
	var out []chain.Assignment
	for _, k := range all {
		if a := k.Value; a.Node == node && !a.Assigned.IsZero() && a.Stands() {
			out = append(out, a)
		}
	}
	sort.Slice(out, func(a, b int) bool { return out[a].Module < out[b].Module })
	return out, nil
}

// recordRuns writes what a send put on a machine into the assignments, once the send is recorded: for each module it carried
// a build of — by the commit — the build, what it published, the commit, the rollout and since when. **A module that runs the
// commit already keeps what it names**, filled where it names nothing and never blanked, so a send that carries it again, or a
// push by hand, changes nothing it knows. A module the machine is not assigned is no assignment to write.
func (i *Inventory) recordRuns(ctx context.Context, nodeID string, builds map[string]string, at time.Time) error {
	c, on, err := i.ChainReads(ctx, KindAssignments)
	if err != nil || !on || len(builds) == 0 {
		return err
	}
	var node string
	if err := i.store.Pool().QueryRow(ctx, `select name from node where id = $1`, nodeID).Scan(&node); errors.Is(err, pgx.ErrNoRows) {
		return nil
	} else if err != nil {
		return err
	}
	rollout, _ := ctx.Value(rolloutKey{}).(rolloutOf)
	modules := make([]string, 0, len(builds))
	for m := range builds {
		modules = append(modules, m)
	}
	sort.Strings(modules)
	var failed []error
	for _, m := range modules {
		commit := builds[m]
		if commit == "" {
			continue
		}
		if err := i.recordRun(ctx, c, node, m, commit, rollout, at); err != nil {
			failed = append(failed, fmt.Errorf("%s on %s: %w", m, node, err))
		}
	}
	return errors.Join(failed...)
}

func (i *Inventory) recordRun(ctx context.Context, c *chain.Chain, node, module, commit string, rollout rolloutOf, at time.Time) error {
	key := assignmentKey(node, module)
	var build, rolloutID string
	if sent, ours := rollout.modules[module]; ours && sent.Commit == commit {
		build, rolloutID = sent.Build, rollout.id
	}
	var published *chain.Published
	resolved := false
	for try := 0; try < casTries; try++ {
		cur, found, err := c.Assignments.Get(ctx, key)
		if err != nil {
			return err
		}
		if !found || cur.Value.Assigned.IsZero() || !cur.Value.Stands() {
			return nil // nothing assigned there to say what it runs
		}
		same := cur.Value.Commit == commit
		if same && cur.Value.Build != "" && (rolloutID == "" || cur.Value.Rollout != "") {
			return nil // a send that carries what the machine runs already says nothing new
		}
		if !resolved {
			// The build of the commit is looked up once, and only for a module that moved or names no build.
			if build == "" {
				if build, err = i.BuildOf(ctx, module, commit); err != nil {
					return err
				}
			}
			if build != "" {
				if b, found, err := i.BuildByID(ctx, build); err != nil {
					return err
				} else if found {
					published = publishedOf(b.Made)
				}
			}
			resolved = true
		}
		next := cur.Value
		if same {
			// Already running it: fill what is missing, blank nothing.
			if next.Build == "" {
				next.Build = build
			}
			if next.Rollout == "" {
				next.Rollout = rolloutID
			}
			if next.Since.IsZero() {
				next.Since = at.UTC()
			}
		} else {
			next.Commit, next.Build, next.Rollout, next.Since = commit, build, rolloutID, at.UTC()
			next.Artifact, next.Package = "", nil
		}
		if published != nil && next.Artifact == "" && next.Package == nil && next.Build == build {
			next.Artifact, next.Package = published.Artifact, published.Package
		}
		if next.Build == cur.Value.Build && next.Commit == cur.Value.Commit && next.Rollout == cur.Value.Rollout &&
			next.Artifact == cur.Value.Artifact && (next.Package == nil) == (cur.Value.Package == nil) &&
			next.Since.Equal(cur.Value.Since) {
			return nil
		}
		if _, err := c.Assignments.Update(ctx, key, next, cur.Revision); errors.Is(err, chain.ErrMoved) {
			continue
		} else {
			return err
		}
	}
	return fmt.Errorf("the assignment kept moving under the write")
}

// saidRunsNot says a send's links could not be written, to the log of whoever sent: the send is away and is recorded, and
// a record of what it put where that could not be written is said, never a reason to call the send failed.
func saidRunsNot(node string, err error) {
	if err != nil {
		fmt.Fprintf(os.Stderr, "mesh-controller: %s was sent its declaration, and the build, commit and rollout of each module "+
			"could not be written in its assignments: %v\n", node, err)
	}
}