14e0b7be9 · carlos/platform

diff --git a/internal/console/apiinstances.go b/internal/console/apiinstances.go

index 36f82cf3a..6a1a123d3 100644

--- a/internal/console/apiinstances.go

+++ b/internal/console/apiinstances.go

@@ -336,6 +336,7 @@ func (s *Server) healthByHost(ctx context.Context, routeFor map[string]registry.

// disagreement no reader could diagnose.

type boxJoin struct {

versions map[string]string

+ serving map[string]string

upstreams map[string][]status.UpstreamState

dbBytes map[string]int64

rollouts map[string]rolloutView

@@ -429,7 +430,7 @@ func loadBoxJoin(ctx context.Context, s release.Store, now time.Time) boxJoin {

// joinBoxes is loadBoxJoin's fold over snapshots already read, so a page that

// has its boxes reads them once.

func joinBoxes(boxes []status.Box, now time.Time) boxJoin {

- j := boxJoin{versions: map[string]string{}, upstreams: map[string][]status.UpstreamState{},

+ j := boxJoin{versions: map[string]string{}, serving: map[string]string{}, upstreams: map[string][]status.UpstreamState{},

dbBytes: map[string]int64{}, rollouts: map[string]rolloutView{}}

for _, b := range boxes {

if b.Metrics == nil || status.BoxStale(b, now, status.SnapshotPeriod) {

@@ -438,6 +439,9 @@ func joinBoxes(boxes []status.Box, now time.Time) boxJoin {

for host, v := range b.Metrics.RouteVersions {

j.versions[host] = v

}

+ for host, v := range b.Metrics.RouteServing {

+ j.serving[host] = v

+ }

for host, ups := range b.Metrics.Upstreams {

j.upstreams[host] = ups

}

diff --git a/internal/console/apisidecars.go b/internal/console/apisidecars.go

index e6a08b12a..67f9a491a 100644

--- a/internal/console/apisidecars.go

+++ b/internal/console/apisidecars.go

@@ -119,16 +119,14 @@ func (s *Server) sidecarManifest(ctx context.Context, w http.ResponseWriter, acc

// meant to touch.

type sidecarListRecord struct {

provision.SidecarRecord

- // RunningVersion is what the sidecar's WORKER last reported, read from

- // whichever box carries its route via that box's own

- // status/boxes/<label>.json (loadBoxJoin, which drops a box that has gone

- // stale rather than letting it answer as current). For a sidecar this is

- // specifically the version stamped by the worker's own /v1/ready call

- // (activator's SetVersions) — the adopt pass never stamps a sidecar row,

- // because release.Adoptable excludes them — so a sidecar that has never

- // reported reads EMPTY here and must be rendered "unknown", never as the

- // channel's desired version. A stamp that overclaims is worse than an

- // absent one.

+ // RunningVersion is the build the agent last started for the sidecar

+ // (registry ServingVersion, zero-downtime plan 5), read from whichever

+ // box carries its route via status/boxes/<label>.json's RouteServing,

+ // under loadBoxJoin's staleness rule. NOT RouteVersions: since plan 4

+ // that is root's install stamp, which can name a build nothing runs yet,

+ // and the worker's own /v1/ready report is display metadata only. A

+ // sidecar that has not started anywhere reads EMPTY and is rendered

+ // "unknown", never as the channel's desired version.

RunningVersion string `json:"running_version"`

}

@@ -151,7 +149,7 @@ func (s *Server) handleAPISidecarList(w http.ResponseWriter, r *http.Request) {

join := loadBoxJoin(ctx, s.cfg.Releases, time.Now())

out := make([]sidecarListRecord, 0, len(recs))

for _, rec := range recs {

- out = append(out, sidecarListRecord{SidecarRecord: rec, RunningVersion: join.versions[rec.Host]})

+ out = append(out, sidecarListRecord{SidecarRecord: rec, RunningVersion: join.serving[rec.Host]})

}

if s.cfg.HostClaim {

pending, err := s.claimPendingViews(ctx, acct.Sqid, app, "sidecar")

diff --git a/internal/console/apisidecars_test.go b/internal/console/apisidecars_test.go

index c1bfd088e..01a2111dd 100644

--- a/internal/console/apisidecars_test.go

+++ b/internal/console/apisidecars_test.go

@@ -698,20 +698,20 @@ func TestAPISidecarListAndGet(t *testing.T) {

}

- // A fresh box reporting alpha's version, and a STALE one reporting

- // beta's, which must drop out of the join rather than answer as current.

+ // A fresh box reporting alpha's serving version and gamma's installed

+ // version, and a stale box reporting beta's serving version.

fresh := status.CollectBox(nil, status.Probes{MachineID: "box1"}, time.Now())

- fresh.Metrics = &status.BoxMetrics{RouteVersions: map[string]string{

- "alpha.titogo.bab.sidecar.internal": "abc1234"}}

+ fresh.Metrics = &status.BoxMetrics{

+ RouteServing: map[string]string{"alpha.titogo.bab.sidecar.internal": "abc1234"},

+ // gamma is installed but not started: root's stamp is not a running build.

+ RouteVersions: map[string]string{"gamma.titogo.bab.sidecar.internal": "f00d123"},

+ }

if err := status.WriteBox(context.Background(), fx.rel, fresh); err != nil {

t.Fatal(err)

}

stale := status.CollectBox(nil, status.Probes{MachineID: "box2"}, time.Now().Add(-time.Hour))

- stale.Metrics = &status.BoxMetrics{RouteVersions: map[string]string{

+ stale.Metrics = &status.BoxMetrics{RouteServing: map[string]string{

"beta.titogo.bab.sidecar.internal": "deadbeef"}}

- if err := status.WriteBox(context.Background(), fx.rel, stale); err != nil {

- t.Fatal(err)

- }

code, _, out := rawReq(t, http.MethodGet, u, token, "")

if code != http.StatusOK {

@@ -752,7 +752,7 @@ func TestAPISidecarListAndGet(t *testing.T) {

t.Errorf("beta running_version = %q, want empty — its only reporting box is stale", got["beta"])

}

if got["gamma"] != "" {

- t.Errorf("gamma running_version = %q, want empty — no box has reported it", got["gamma"])

+ t.Errorf("gamma running_version = %q, want empty — an install stamp is not a running build", got["gamma"])

}

// GET one.

diff --git a/internal/status/snapshot.go b/internal/status/snapshot.go

index 45788606c..b0ba61923 100644

--- a/internal/status/snapshot.go

+++ b/internal/status/snapshot.go

@@ -454,6 +454,12 @@ type BoxMetrics struct {

// no versioned route, same "absent, not empty" rule as the rest of this

// block.

RouteVersions map[string]string `json:"route_versions,omitempty"`

+ // RouteServing is host -> the build the agent last started for that

+ // route (registry ServingVersion, zero-downtime plan 5), for this box's

+ // own non-platform routes that have one. RouteVersions is root's install

+ // stamp and can name a build nothing runs yet; this is what runs. Private

+ // and omitted when empty, for RouteVersions' reasons.

+ RouteServing map[string]string `json:"route_serving,omitempty"`

// Upstreams is the box's per-upstream probe table, route host -> the

// UpstreamState of each addr behind it (multi-upstream routes only; a

// single-addr route with no health config is never probed and never

@@ -765,6 +771,10 @@ func CollectBox(routes []registry.Route, p Probes, now time.Time) Box {

// box's Metrics attachment check below sees exactly "nothing to report"

// rather than a present-but-empty map.

var routeVersions map[string]string

+ // routeServing is plan 5's ServingVersion: host -> the build the agent

+ // last started, for this box's own non-platform routes that have one.

+ // Same nil-until-first-entry convention as routeVersions.

+ var routeServing map[string]string

// aliasReadings is issue #151's Metrics.Aliases source. Declared here,

// above the route loop, so the route loop below (a first-route claim's

// ROUTE readings) and the p.Aliases loop after it (genuine alias rows)

@@ -793,6 +803,12 @@ func CollectBox(routes []registry.Route, p Probes, now time.Time) Box {

}

routeVersions[r.Host] = r.Version

}

+ if r.ServingVersion != "" {

+ if routeServing == nil {

+ routeServing = map[string]string{}

+ }

+ routeServing[r.Host] = r.ServingVersion

+ }

if r.Kind != "static" && r.DBPath != "" && p.DBBytes != nil {

if n := p.DBBytes(r.DBPath); n != nil {

if routeDBBytes == nil {

@@ -914,7 +930,7 @@ func CollectBox(routes []registry.Route, p Probes, now time.Time) Box {

if p.MemTotalBytes != nil || p.MemAvailBytes != nil || p.Load1 != nil ||

p.DiskFreeBytes != nil || p.Release != "" || p.Tunnel != "" ||

p.InstanceID != "" || p.Platform != "" || p.SystemChannel != "" || p.System != nil ||

- len(routeVersions) > 0 || len(aliasReadings) > 0 || len(p.Upstreams) > 0 ||

+ len(routeVersions) > 0 || len(routeServing) > 0 || len(aliasReadings) > 0 || len(p.Upstreams) > 0 ||

len(routeDBBytes) > 0 || p.Layout != nil || len(p.ConfigRefusals) > 0 || len(p.ReleaseRefusals) > 0 || len(p.ProofWaits) > 0 || p.Rollout != nil || p.ConfigGuard != "" ||

p.Alerts != nil || p.GeoIP != nil {

box.Metrics = &BoxMetrics{

@@ -937,6 +953,7 @@ func CollectBox(routes []registry.Route, p Probes, now time.Time) Box {

InstanceID: p.InstanceID,

Alerts: p.Alerts,

RouteVersions: routeVersions,

+ RouteServing: routeServing,

Aliases: aliasReadings,

// Carried through as the prober handed it over: proberSet.State

// already returns a fresh copy per call, so there is nothing here

diff --git a/internal/status/snapshot_test.go b/internal/status/snapshot_test.go

index 1afd4c3b7..286e2dfea 100644

--- a/internal/status/snapshot_test.go

+++ b/internal/status/snapshot_test.go

@@ -7,6 +7,7 @@ import (

"testing"

"time"

+ "github.com/carlosframework/platform/internal/registry"

"github.com/carlosframework/platform/internal/release"

)

@@ -168,3 +169,23 @@ func TestTheRolloutReportRidesThePrivateMetrics(t *testing.T) {

t.Fatal("the flag off publishes no rollout block and no Metrics because of it")

}

+

+// RouteServing is the build the agent started for each route (plan 5's

+// ServingVersion), carried beside RouteVersions, root's install stamp.

+func TestCollectBoxCarriesTheServingBuild(t *testing.T) {

+ routes := []registry.Route{

+ {Host: "side.titogo.bab.sidecar.internal", Kind: "sidecar", Version: "v2", ServingVersion: "v1"},

+ {Host: "never.titogo.bab.sidecar.internal", Kind: "sidecar", Version: "v2"},

+ {Host: "alias.example.com", Platform: true, ServingVersion: "v9"},

+ }

+ b := CollectBox(routes, Probes{MachineID: "m"}, time.Now())

+ if b.Metrics == nil || b.Metrics.RouteServing["side.titogo.bab.sidecar.internal"] != "v1" {

+ t.Fatalf("RouteServing = %+v", b.Metrics)

+ }

+ if _, ok := b.Metrics.RouteServing["never.titogo.bab.sidecar.internal"]; ok {

+ t.Error("a route that never started has a serving build")

+ }

+ if _, ok := b.Metrics.RouteServing["alias.example.com"]; ok {

+ t.Error("a platform alias carries a serving build")

+ }

+}

diff --git a/sidecars_cli.go b/sidecars_cli.go

index ec7dc5e0c..652528eb3 100644

--- a/sidecars_cli.go

+++ b/sidecars_cli.go

@@ -366,13 +366,12 @@ func runSidecarsCreate(args []string) {

mutedf(stdout, " an always-on one (hibernate: false in the app's sidecar settings) starts as soon as the row lands")

}

-// sidecarVersionCell renders `ls`'s VERSION column: the raw running version

-// when the sidecar has reported one via /v1/ready, or the literal

-// "unknown" when it never has — the honesty rule this branch was built

-// around (design §9.2, binding on this task): a stamp that overclaims is

-// worse than an absent one, so this must NEVER render the record's own

-// Channel (the desired version) as a stand-in for a version nobody has

-// confirmed is actually running.

+// sidecarVersionCell renders `ls`'s VERSION column: the build the agent

+// last started for the sidecar, or the literal "unknown" when it never has

+// — the honesty rule this branch was built around (design §9.2, binding on

+// this task): a stamp that overclaims is worse than an absent one, so this

+// must NEVER render the record's own Channel (the desired version) as a

+// stand-in for a version nobody has confirmed is actually running.

func sidecarVersionCell(running string) string {

if running == "" {

return "unknown"

carlos / platform Public

zdd-6 task 13: carlos sidecars ls shows the serving build