refactor: move the deep-analysis last-run record out of the cache

README.md

@@ -155,7 +155,7 @@ Pick the configuration that matches your Plex server:
 
 | Mount | Description |
 | --- | --- |
-| `/config` | Persistent cache (`profiles.json` learned language profiles, `tokens.json` encrypted shared-user tokens, `state.json` sync state). Mount a named volume or host path to preserve data across restarts. A corrupt file resets only its own section; a `cache.json` from an earlier version migrates automatically on first start. |
+| `/config` | Persistent cache (`profiles.json` learned language profiles, `tokens.json` encrypted shared-user tokens, `state.json` sync state, `.plex-language-sync-last-run` completion record for the deep-scan safety net). Mount a named volume or host path to preserve data across restarts. A corrupt file resets only its own section; a `cache.json` from an earlier version migrates automatically on first start. |
 
 ## Graceful shutdown
 

internal/cache/cache.go

@@ -1,13 +1,12 @@
 // Package cache is the on-disk persistence layer for processed-episode
-// deduplication, per-user language profiles, shared-user tokens, and the
-// scheduler's last-run marker.
+// deduplication, per-user language profiles, and shared-user tokens.
 //
 // On-disk layout: state is split across three files in the cache dir by
 // retention class, so one corrupt file never costs another class's state:
 //
-//	profiles.json  language_profiles                     irreplaceable learned state
-//	tokens.json    user_tokens                           re-fetchable encrypted secrets
-//	state.json     processed_episodes, last_scheduler_run disposable operational state
+//	profiles.json  language_profiles  irreplaceable learned state
+//	tokens.json    user_tokens        re-fetchable encrypted secrets
+//	state.json     processed_episodes disposable operational state
 //
 // Field names, types, and JSON tags within each file are an inviolate
 // read-forward / write-back contract across deploys — any change is a
@@ -55,8 +54,8 @@ const (
 )
 
 // Data is the in-memory state shape. It doubles as the decode target for
-// the legacy pre-split cache.json union schema, whose field names and JSON
-// tags it preserves verbatim (read-forward contract).
+// the legacy pre-split cache.json union schema; a legacy key with no field
+// here (last_scheduler_run) is ignored on decode.
 type Data struct {
 	// ProcessedEpisodes tracks recently processed episode keys to avoid
 	// re-processing the same episode on rapid successive events.
@@ -77,9 +76,6 @@ type Data struct {
 	// UserTokens maps userID → accessToken for shared users. Persisted in
 	// tokens.json, encrypted at the disk boundary when a key is set.
 	UserTokens map[string]string `json:"user_tokens"`
-	// LastSchedulerRun is the unix timestamp of the last scheduler run.
-	// Persisted in state.json.
-	LastSchedulerRun int64 `json:"last_scheduler_run"`
 }
 
 // profilesData is the profiles.json schema (irreplaceable learned state).
@@ -96,7 +92,6 @@ type tokensData struct {
 // stateData is the state.json schema (disposable operational state).
 type stateData struct {
 	ProcessedEpisodes map[string]int64 `json:"processed_episodes"`
-	LastSchedulerRun  int64            `json:"last_scheduler_run"`
 }
 
 // Cache is the concurrent-safe persistent cache. The zero value is usable;
@@ -242,7 +237,6 @@ func (c *Cache) applyLegacyLocked(legacy *Data) {
 		c.data.UserTokens = legacy.UserTokens
 		c.decryptTokensLocked()
 	}
-	c.data.LastSchedulerRun = legacy.LastSchedulerRun
 }
 
 // overlayProfilesLocked loads profiles.json over the baseline. Returns 1
@@ -319,7 +313,6 @@ func (c *Cache) overlayStateLocked(dir string, errs *[]error) int {
 	if c.data.ProcessedEpisodes == nil {
 		c.data.ProcessedEpisodes = make(map[string]int64)
 	}
-	c.data.LastSchedulerRun = sd.LastSchedulerRun
 	return 1
 }
 
@@ -475,7 +468,6 @@ func (c *Cache) encodeAllForSave() (profiles, tokens, state []byte, err error) {
 	}
 	if state, err = json.MarshalIndent(&stateData{
 		ProcessedEpisodes: c.data.ProcessedEpisodes,
-		LastSchedulerRun:  c.data.LastSchedulerRun,
 	}, "", "  "); err != nil {
 		return nil, nil, nil, fmt.Errorf("marshal %s: %w", stateFile, err)
 	}

internal/cache/cache_test.go

@@ -57,7 +57,6 @@ func TestCacheSaveLoadRoundTrip(t *testing.T) {
 		"1": {"jpn": "eng", "eng": ""},
 	}
 	orig.data.UserTokens = map[string]string{"2": "t2"}
-	orig.data.LastSchedulerRun = 1700000000
 
 	if err := orig.Save(dir); err != nil {
 		t.Fatalf("Save() error = %v", err)
@@ -83,8 +82,27 @@ func TestCacheSaveLoadRoundTrip(t *testing.T) {
 	if loaded.data.UserTokens["2"] != "t2" {
 		t.Errorf("UserTokens[2] = %q, want t2", loaded.data.UserTokens["2"])
 	}
-	if loaded.data.LastSchedulerRun != 1700000000 {
-		t.Errorf("LastSchedulerRun = %d, want 1700000000", loaded.data.LastSchedulerRun)
+}
+
+// TestCacheLoadIgnoresRetiredMarkerKey pins the read-forward contract for
+// files written by earlier versions: state.json (and the legacy cache.json)
+// used to carry a last_scheduler_run key, and a file still holding it must
+// load its live sections cleanly. Fails if the load path ever opts into
+// strict decoding (DisallowUnknownFields).
+func TestCacheLoadIgnoresRetiredMarkerKey(t *testing.T) {
+	t.Parallel()
+	dir := t.TempDir()
+	state := `{"processed_episodes":{"timeline:9":1700000000},"last_scheduler_run":1700000000}`
+	if err := os.WriteFile(filepath.Join(dir, stateFile), []byte(state), 0o600); err != nil {
+		t.Fatal(err)
+	}
+
+	c := New()
+	if err := c.Load(dir); err != nil {
+		t.Errorf("Load() with a retired key in state.json = %v, want nil", err)
+	}
+	if len(c.data.ProcessedEpisodes) != 1 {
+		t.Errorf("ProcessedEpisodes = %v, want the seeded entry loaded beside the retired key", c.data.ProcessedEpisodes)
 	}
 }
 
@@ -211,7 +229,6 @@ func TestCacheLoadCorruptSectionIsolation(t *testing.T) {
 		orig.data.ProcessedEpisodes = map[string]int64{"streams:1:100:1:2": time.Now().Unix()}
 		orig.data.LanguageProfiles = map[string]map[string]string{"1": {"jpn": "eng"}}
 		orig.data.UserTokens = map[string]string{"2": "t2"}
-		orig.data.LastSchedulerRun = 1700000000
 		if err := orig.Save(dir); err != nil {
 			t.Fatalf("seed Save() error = %v", err)
 		}
@@ -242,8 +259,7 @@ func TestCacheLoadCorruptSectionIsolation(t *testing.T) {
 
 			gotProfiles := loaded.data.LanguageProfiles["1"]["jpn"] == "eng"
 			gotTokens := loaded.data.UserTokens["2"] == "t2"
-			gotState := loaded.data.LastSchedulerRun == 1700000000 &&
-				len(loaded.data.ProcessedEpisodes) == 1
+			gotState := len(loaded.data.ProcessedEpisodes) == 1
 
 			if want := tc.corrupt != profilesFile; gotProfiles != want {
 				t.Errorf("profiles survived = %v, want %v", gotProfiles, want)
@@ -283,7 +299,6 @@ func TestCacheLoadMigratesLegacyCacheJSON(t *testing.T) {
 		ProcessedEpisodes: map[string]int64{"timeline:9": time.Now().Unix()},
 		LanguageProfiles:  map[string]map[string]string{"1": {"jpn": "eng"}},
 		UserTokens:        map[string]string{"2": "plain-tok"},
-		LastSchedulerRun:  1700000000,
 	})
 
 	key, err := DeriveKey("admin-token")
@@ -297,7 +312,7 @@ func TestCacheLoadMigratesLegacyCacheJSON(t *testing.T) {
 	}
 
 	if c.data.LanguageProfiles["1"]["jpn"] != "eng" || c.data.UserTokens["2"] != "plain-tok" ||
-		c.data.LastSchedulerRun != 1700000000 || len(c.data.ProcessedEpisodes) != 1 {
+		len(c.data.ProcessedEpisodes) != 1 {
 		t.Errorf("migrated state incomplete: %+v", c.data)
 	}
 	for _, name := range splitFiles {
@@ -343,14 +358,12 @@ func TestCacheLoadSplitWinsOverStaleLegacy(t *testing.T) {
 	split := New()
 	split.data.LanguageProfiles = map[string]map[string]string{"1": {"jpn": "eng"}}
 	split.data.UserTokens = map[string]string{"2": "new-tok"}
-	split.data.LastSchedulerRun = 2000000000
 	if err := split.Save(dir); err != nil {
 		t.Fatalf("Save() error = %v", err)
 	}
 	legacyWrite(t, dir, Data{
 		LanguageProfiles: map[string]map[string]string{"1": {"jpn": "STALE"}},
 		UserTokens:       map[string]string{"2": "STALE"},
-		LastSchedulerRun: 1,
 	})
 
 	loaded := New()
@@ -365,9 +378,6 @@ func TestCacheLoadSplitWinsOverStaleLegacy(t *testing.T) {
 		t.Errorf("tokens = %q, want the split file's value to win over stale legacy",
 			loaded.data.UserTokens["2"])
 	}
-	if loaded.data.LastSchedulerRun != 2000000000 {
-		t.Errorf("LastSchedulerRun = %d, want the split file's value", loaded.data.LastSchedulerRun)
-	}
 	if _, err := os.Stat(filepath.Join(dir, legacyCacheFile)); !os.IsNotExist(err) {
 		t.Errorf("stale legacy cache.json not removed (stat err = %v)", err)
 	}
@@ -383,7 +393,6 @@ func TestCacheLoadLegacyFillsMissingSection(t *testing.T) {
 	legacyWrite(t, dir, Data{
 		LanguageProfiles: map[string]map[string]string{"1": {"jpn": "LEGACY"}},
 		UserTokens:       map[string]string{"2": "legacy-tok"},
-		LastSchedulerRun: 1700000000,
 	})
 	pd, err := json.Marshal(&profilesData{
 		LanguageProfiles: map[string]map[string]string{"1": {"jpn": "SPLIT"}},
@@ -405,9 +414,6 @@ func TestCacheLoadLegacyFillsMissingSection(t *testing.T) {
 	if loaded.data.UserTokens["2"] != "legacy-tok" {
 		t.Errorf("tokens = %q, want the legacy value for the missing section", loaded.data.UserTokens["2"])
 	}
-	if loaded.data.LastSchedulerRun != 1700000000 {
-		t.Errorf("LastSchedulerRun = %d, want the legacy value", loaded.data.LastSchedulerRun)
-	}
 	if _, err := os.Stat(filepath.Join(dir, legacyCacheFile)); !os.IsNotExist(err) {
 		t.Errorf("legacy cache.json not removed after completing migration (stat err = %v)", err)
 	}
@@ -535,7 +541,7 @@ func TestCacheLoadLeavesAuthoritativeSplitFilesUntouched(t *testing.T) {
 			LanguageProfiles: map[string]map[string]string{"1": {"jpn": "eng"}},
 		}),
 		tokensFile: mustCompactJSON(t, &tokensData{UserTokens: map[string]string{"2": "new-tok"}}),
-		stateFile:  mustCompactJSON(t, &stateData{LastSchedulerRun: 2000000000}),
+		stateFile:  mustCompactJSON(t, &stateData{ProcessedEpisodes: map[string]int64{"timeline:9": 2000000000}}),
 	}
 	for name, raw := range written {
 		if err := os.WriteFile(filepath.Join(dir, name), raw, 0o600); err != nil {
@@ -581,7 +587,7 @@ func mustCompactJSON(t *testing.T, v any) []byte {
 	return raw
 }
 
-// --- PBT: JSON round-trip preserves LastSchedulerRun + map lengths ---
+// --- PBT: JSON round-trip preserves the map lengths ---
 
 func TestCacheDataJSONRoundTrip(t *testing.T) {
 	rapid.Check(t, func(t *rapid.T) {
@@ -596,7 +602,6 @@ func TestCacheDataJSONRoundTrip(t *testing.T) {
 			ProcessedEpisodes: processed,
 			LanguageProfiles:  make(map[string]map[string]string),
 			UserTokens:        make(map[string]string),
-			LastSchedulerRun:  int64(rapid.IntRange(0, 2000000000).Draw(t, "last_run")),
 		}
 
 		data, err := json.Marshal(&original)
@@ -613,10 +618,6 @@ func TestCacheDataJSONRoundTrip(t *testing.T) {
 			t.Errorf("ProcessedEpisodes length: got %d, want %d",
 				len(decoded.ProcessedEpisodes), len(original.ProcessedEpisodes))
 		}
-		if decoded.LastSchedulerRun != original.LastSchedulerRun {
-			t.Errorf("LastSchedulerRun: got %d, want %d",
-				decoded.LastSchedulerRun, original.LastSchedulerRun)
-		}
 	})
 }
 

internal/cache/contract.go

@@ -3,7 +3,6 @@ package cache
 import (
 	"sync"
 	"testing"
-	"time"
 
 	"github.com/cplieger/plex-language-sync/internal/streams"
 )
@@ -24,8 +23,6 @@ type Contract interface {
 	IntentFor(userID, showKey string) (streams.Intent, bool)
 	UserTokens() map[string]string
 	SetUserTokens(tokens map[string]string)
-	LastSchedulerRun() time.Time
-	SetLastSchedulerRun(t time.Time)
 }
 
 // RunContract exercises the persisted-cache contract against any
@@ -80,36 +77,9 @@ func RunContract(t *testing.T, c Contract) {
 		wg.Wait()
 	})
 
-	schedulerRunContract(t, c)
 	checkAndMarkContract(t, c)
 }
 
-// schedulerRunContract exercises the scheduler-run portion of the Contract
-// contract: the last-scheduler-run marker round-trips a whole-second value and
-// resets to the zero time. Split out of RunCacheContract to keep that
-// function's cognitive complexity under the gate.
-func schedulerRunContract(t *testing.T, c Contract) {
-	t.Helper()
-
-	t.Run("scheduler_run_roundtrip", func(t *testing.T) {
-		// Whole-second value: internal/cache persists the marker as a unix
-		// int64 (time.Unix truncation), so the shared contract is pinned at
-		// second granularity that both implementations honour.
-		want := time.Unix(1700000000, 0)
-		c.SetLastSchedulerRun(want)
-		if got := c.LastSchedulerRun(); !got.Equal(want) {
-			t.Errorf("LastSchedulerRun = %v, want %v", got, want)
-		}
-	})
-
-	t.Run("scheduler_run_zero", func(t *testing.T) {
-		c.SetLastSchedulerRun(time.Time{})
-		if got := c.LastSchedulerRun(); !got.IsZero() {
-			t.Errorf("LastSchedulerRun after zero set = %v, want zero", got)
-		}
-	})
-}
-
 // Language-code literals shared by the contract subtests.
 const (
 	langJPN = "jpn"