lib/storage: reset MetricName->TSID cache after marking metricIDs as deleted

This is a follow-up commit after 12b16077c4 ,
which didn't reset the `tsidCache` in all the required places.
This could result in indefinite errors like:

    missing metricName by metricID ...; this could be the case after unclean shutdown; deleting the metricID, so it could be re-created next time

Fix this by resetting the cache inside deleteMetricIDs function.
This commit is contained in:
Aliaksandr Valialkin 2020-07-14 14:02:14 +03:00
parent 25f80d320b
commit e1107fec10
4 changed files with 44 additions and 19 deletions

View File

@ -128,6 +128,9 @@ type indexDB struct {
// Cache for fast MetricID -> MetricName lookup. // Cache for fast MetricID -> MetricName lookup.
metricNameCache *workingsetcache.Cache metricNameCache *workingsetcache.Cache
// Cache for fast MetricName -> TSID lookups.
tsidCache *workingsetcache.Cache
// Cache for useless TagFilters entries, which have no tag filters // Cache for useless TagFilters entries, which have no tag filters
// matching low number of metrics. // matching low number of metrics.
uselessTagFiltersCache *workingsetcache.Cache uselessTagFiltersCache *workingsetcache.Cache
@ -155,13 +158,16 @@ type indexDB struct {
} }
// openIndexDB opens index db from the given path with the given caches. // openIndexDB opens index db from the given path with the given caches.
func openIndexDB(path string, metricIDCache, metricNameCache *workingsetcache.Cache, currHourMetricIDs, prevHourMetricIDs *atomic.Value) (*indexDB, error) { func openIndexDB(path string, metricIDCache, metricNameCache, tsidCache *workingsetcache.Cache, currHourMetricIDs, prevHourMetricIDs *atomic.Value) (*indexDB, error) {
if metricIDCache == nil { if metricIDCache == nil {
logger.Panicf("BUG: metricIDCache must be non-nil") logger.Panicf("BUG: metricIDCache must be non-nil")
} }
if metricNameCache == nil { if metricNameCache == nil {
logger.Panicf("BUG: metricNameCache must be non-nil") logger.Panicf("BUG: metricNameCache must be non-nil")
} }
if tsidCache == nil {
logger.Panicf("BUG: tsidCache must be nin-nil")
}
if currHourMetricIDs == nil { if currHourMetricIDs == nil {
logger.Panicf("BUG: currHourMetricIDs must be non-nil") logger.Panicf("BUG: currHourMetricIDs must be non-nil")
} }
@ -187,6 +193,7 @@ func openIndexDB(path string, metricIDCache, metricNameCache *workingsetcache.Ca
tagCache: workingsetcache.New(mem/32, time.Hour), tagCache: workingsetcache.New(mem/32, time.Hour),
metricIDCache: metricIDCache, metricIDCache: metricIDCache,
metricNameCache: metricNameCache, metricNameCache: metricNameCache,
tsidCache: tsidCache,
uselessTagFiltersCache: workingsetcache.New(mem/128, time.Hour), uselessTagFiltersCache: workingsetcache.New(mem/128, time.Hour),
metricIDsPerDateTagFilterCache: workingsetcache.New(mem/128, time.Hour), metricIDsPerDateTagFilterCache: workingsetcache.New(mem/128, time.Hour),
@ -360,6 +367,7 @@ func (db *indexDB) decRef() {
db.tagCache = nil db.tagCache = nil
db.metricIDCache = nil db.metricIDCache = nil
db.metricNameCache = nil db.metricNameCache = nil
db.tsidCache = nil
db.uselessTagFiltersCache = nil db.uselessTagFiltersCache = nil
db.metricIDsPerDateTagFilterCache = nil db.metricIDsPerDateTagFilterCache = nil
@ -1218,6 +1226,9 @@ func (db *indexDB) deleteMetricIDs(metricIDs []uint64) error {
// Reset TagFilters -> TSIDS cache, since it may contain deleted TSIDs. // Reset TagFilters -> TSIDS cache, since it may contain deleted TSIDs.
invalidateTagCache() invalidateTagCache()
// Reset MetricName -> TSID cache, since it may contain deleted TSIDs.
db.tsidCache.Reset()
// Do not reset uselessTagFiltersCache, since the found metricIDs // Do not reset uselessTagFiltersCache, since the found metricIDs
// on cache miss are filtered out later with deletedMetricIDs. // on cache miss are filtered out later with deletedMetricIDs.
return nil return nil
@ -1488,7 +1499,7 @@ func (is *indexSearch) searchTSIDs(tfss []*TagFilters, tr TimeRange, maxMetrics
tsids := make([]TSID, len(metricIDs)) tsids := make([]TSID, len(metricIDs))
i := 0 i := 0
for _, metricID := range metricIDs { for _, metricID := range metricIDs {
// Try obtaining TSIDs from db.tsidCache. This is much faster // Try obtaining TSIDs from MetricID->TSID cache. This is much faster
// than scanning the mergeset if it contains a lot of metricIDs. // than scanning the mergeset if it contains a lot of metricIDs.
tsid := &tsids[i] tsid := &tsids[i]
err := is.db.getFromMetricIDCache(tsid, metricID) err := is.db.getFromMetricIDCache(tsid, metricID)

View File

@ -450,8 +450,10 @@ func TestMarshalUnmarshalTSIDs(t *testing.T) {
func TestIndexDBOpenClose(t *testing.T) { func TestIndexDBOpenClose(t *testing.T) {
metricIDCache := workingsetcache.New(1234, time.Hour) metricIDCache := workingsetcache.New(1234, time.Hour)
metricNameCache := workingsetcache.New(1234, time.Hour) metricNameCache := workingsetcache.New(1234, time.Hour)
tsidCache := workingsetcache.New(1234, time.Hour)
defer metricIDCache.Stop() defer metricIDCache.Stop()
defer metricNameCache.Stop() defer metricNameCache.Stop()
defer tsidCache.Stop()
var hmCurr atomic.Value var hmCurr atomic.Value
hmCurr.Store(&hourMetricIDs{}) hmCurr.Store(&hourMetricIDs{})
@ -459,7 +461,7 @@ func TestIndexDBOpenClose(t *testing.T) {
hmPrev.Store(&hourMetricIDs{}) hmPrev.Store(&hourMetricIDs{})
for i := 0; i < 5; i++ { for i := 0; i < 5; i++ {
db, err := openIndexDB("test-index-db", metricIDCache, metricNameCache, &hmCurr, &hmPrev) db, err := openIndexDB("test-index-db", metricIDCache, metricNameCache, tsidCache, &hmCurr, &hmPrev)
if err != nil { if err != nil {
t.Fatalf("cannot open indexDB: %s", err) t.Fatalf("cannot open indexDB: %s", err)
} }
@ -476,8 +478,10 @@ func TestIndexDB(t *testing.T) {
t.Run("serial", func(t *testing.T) { t.Run("serial", func(t *testing.T) {
metricIDCache := workingsetcache.New(1234, time.Hour) metricIDCache := workingsetcache.New(1234, time.Hour)
metricNameCache := workingsetcache.New(1234, time.Hour) metricNameCache := workingsetcache.New(1234, time.Hour)
tsidCache := workingsetcache.New(1234, time.Hour)
defer metricIDCache.Stop() defer metricIDCache.Stop()
defer metricNameCache.Stop() defer metricNameCache.Stop()
defer tsidCache.Stop()
var hmCurr atomic.Value var hmCurr atomic.Value
hmCurr.Store(&hourMetricIDs{}) hmCurr.Store(&hourMetricIDs{})
@ -485,7 +489,7 @@ func TestIndexDB(t *testing.T) {
hmPrev.Store(&hourMetricIDs{}) hmPrev.Store(&hourMetricIDs{})
dbName := "test-index-db-serial" dbName := "test-index-db-serial"
db, err := openIndexDB(dbName, metricIDCache, metricNameCache, &hmCurr, &hmPrev) db, err := openIndexDB(dbName, metricIDCache, metricNameCache, tsidCache, &hmCurr, &hmPrev)
if err != nil { if err != nil {
t.Fatalf("cannot open indexDB: %s", err) t.Fatalf("cannot open indexDB: %s", err)
} }
@ -515,7 +519,7 @@ func TestIndexDB(t *testing.T) {
// Re-open the db and verify it works as expected. // Re-open the db and verify it works as expected.
db.MustClose() db.MustClose()
db, err = openIndexDB(dbName, metricIDCache, metricNameCache, &hmCurr, &hmPrev) db, err = openIndexDB(dbName, metricIDCache, metricNameCache, tsidCache, &hmCurr, &hmPrev)
if err != nil { if err != nil {
t.Fatalf("cannot open indexDB: %s", err) t.Fatalf("cannot open indexDB: %s", err)
} }
@ -533,8 +537,10 @@ func TestIndexDB(t *testing.T) {
t.Run("concurrent", func(t *testing.T) { t.Run("concurrent", func(t *testing.T) {
metricIDCache := workingsetcache.New(1234, time.Hour) metricIDCache := workingsetcache.New(1234, time.Hour)
metricNameCache := workingsetcache.New(1234, time.Hour) metricNameCache := workingsetcache.New(1234, time.Hour)
tsidCache := workingsetcache.New(1234, time.Hour)
defer metricIDCache.Stop() defer metricIDCache.Stop()
defer metricNameCache.Stop() defer metricNameCache.Stop()
defer tsidCache.Stop()
var hmCurr atomic.Value var hmCurr atomic.Value
hmCurr.Store(&hourMetricIDs{}) hmCurr.Store(&hourMetricIDs{})
@ -542,7 +548,7 @@ func TestIndexDB(t *testing.T) {
hmPrev.Store(&hourMetricIDs{}) hmPrev.Store(&hourMetricIDs{})
dbName := "test-index-db-concurrent" dbName := "test-index-db-concurrent"
db, err := openIndexDB(dbName, metricIDCache, metricNameCache, &hmCurr, &hmPrev) db, err := openIndexDB(dbName, metricIDCache, metricNameCache, tsidCache, &hmCurr, &hmPrev)
if err != nil { if err != nil {
t.Fatalf("cannot open indexDB: %s", err) t.Fatalf("cannot open indexDB: %s", err)
} }
@ -1463,8 +1469,10 @@ func TestMatchTagFilters(t *testing.T) {
func TestSearchTSIDWithTimeRange(t *testing.T) { func TestSearchTSIDWithTimeRange(t *testing.T) {
metricIDCache := workingsetcache.New(1234, time.Hour) metricIDCache := workingsetcache.New(1234, time.Hour)
metricNameCache := workingsetcache.New(1234, time.Hour) metricNameCache := workingsetcache.New(1234, time.Hour)
tsidCache := workingsetcache.New(1234, time.Hour)
defer metricIDCache.Stop() defer metricIDCache.Stop()
defer metricNameCache.Stop() defer metricNameCache.Stop()
defer tsidCache.Stop()
currMetricIDs := &hourMetricIDs{ currMetricIDs := &hourMetricIDs{
isFull: true, isFull: true,
@ -1482,7 +1490,7 @@ func TestSearchTSIDWithTimeRange(t *testing.T) {
hmPrev.Store(prevMetricIDs) hmPrev.Store(prevMetricIDs)
dbName := "test-index-db-ts-range" dbName := "test-index-db-ts-range"
db, err := openIndexDB(dbName, metricIDCache, metricNameCache, &hmCurr, &hmPrev) db, err := openIndexDB(dbName, metricIDCache, metricNameCache, tsidCache, &hmCurr, &hmPrev)
if err != nil { if err != nil {
t.Fatalf("cannot open indexDB: %s", err) t.Fatalf("cannot open indexDB: %s", err)
} }

View File

@ -45,8 +45,10 @@ func BenchmarkIndexDBAddTSIDs(b *testing.B) {
metricIDCache := workingsetcache.New(1234, time.Hour) metricIDCache := workingsetcache.New(1234, time.Hour)
metricNameCache := workingsetcache.New(1234, time.Hour) metricNameCache := workingsetcache.New(1234, time.Hour)
tsidCache := workingsetcache.New(1234, time.Hour)
defer metricIDCache.Stop() defer metricIDCache.Stop()
defer metricNameCache.Stop() defer metricNameCache.Stop()
defer tsidCache.Stop()
var hmCurr atomic.Value var hmCurr atomic.Value
hmCurr.Store(&hourMetricIDs{}) hmCurr.Store(&hourMetricIDs{})
@ -54,7 +56,7 @@ func BenchmarkIndexDBAddTSIDs(b *testing.B) {
hmPrev.Store(&hourMetricIDs{}) hmPrev.Store(&hourMetricIDs{})
const dbName = "bench-index-db-add-tsids" const dbName = "bench-index-db-add-tsids"
db, err := openIndexDB(dbName, metricIDCache, metricNameCache, &hmCurr, &hmPrev) db, err := openIndexDB(dbName, metricIDCache, metricNameCache, tsidCache, &hmCurr, &hmPrev)
if err != nil { if err != nil {
b.Fatalf("cannot open indexDB: %s", err) b.Fatalf("cannot open indexDB: %s", err)
} }
@ -113,8 +115,10 @@ func BenchmarkHeadPostingForMatchers(b *testing.B) {
// See https://www.robustperception.io/evaluating-performance-and-correctness for more details. // See https://www.robustperception.io/evaluating-performance-and-correctness for more details.
metricIDCache := workingsetcache.New(1234, time.Hour) metricIDCache := workingsetcache.New(1234, time.Hour)
metricNameCache := workingsetcache.New(1234, time.Hour) metricNameCache := workingsetcache.New(1234, time.Hour)
tsidCache := workingsetcache.New(1234, time.Hour)
defer metricIDCache.Stop() defer metricIDCache.Stop()
defer metricNameCache.Stop() defer metricNameCache.Stop()
defer tsidCache.Stop()
var hmCurr atomic.Value var hmCurr atomic.Value
hmCurr.Store(&hourMetricIDs{}) hmCurr.Store(&hourMetricIDs{})
@ -122,7 +126,7 @@ func BenchmarkHeadPostingForMatchers(b *testing.B) {
hmPrev.Store(&hourMetricIDs{}) hmPrev.Store(&hourMetricIDs{})
const dbName = "bench-head-posting-for-matchers" const dbName = "bench-head-posting-for-matchers"
db, err := openIndexDB(dbName, metricIDCache, metricNameCache, &hmCurr, &hmPrev) db, err := openIndexDB(dbName, metricIDCache, metricNameCache, tsidCache, &hmCurr, &hmPrev)
if err != nil { if err != nil {
b.Fatalf("cannot open indexDB: %s", err) b.Fatalf("cannot open indexDB: %s", err)
} }
@ -295,8 +299,10 @@ func BenchmarkHeadPostingForMatchers(b *testing.B) {
func BenchmarkIndexDBGetTSIDs(b *testing.B) { func BenchmarkIndexDBGetTSIDs(b *testing.B) {
metricIDCache := workingsetcache.New(1234, time.Hour) metricIDCache := workingsetcache.New(1234, time.Hour)
metricNameCache := workingsetcache.New(1234, time.Hour) metricNameCache := workingsetcache.New(1234, time.Hour)
tsidCache := workingsetcache.New(1234, time.Hour)
defer metricIDCache.Stop() defer metricIDCache.Stop()
defer metricNameCache.Stop() defer metricNameCache.Stop()
defer tsidCache.Stop()
var hmCurr atomic.Value var hmCurr atomic.Value
hmCurr.Store(&hourMetricIDs{}) hmCurr.Store(&hourMetricIDs{})
@ -304,7 +310,7 @@ func BenchmarkIndexDBGetTSIDs(b *testing.B) {
hmPrev.Store(&hourMetricIDs{}) hmPrev.Store(&hourMetricIDs{})
const dbName = "bench-index-db-get-tsids" const dbName = "bench-index-db-get-tsids"
db, err := openIndexDB(dbName, metricIDCache, metricNameCache, &hmCurr, &hmPrev) db, err := openIndexDB(dbName, metricIDCache, metricNameCache, tsidCache, &hmCurr, &hmPrev)
if err != nil { if err != nil {
b.Fatalf("cannot open indexDB: %s", err) b.Fatalf("cannot open indexDB: %s", err)
} }

View File

@ -166,7 +166,7 @@ func OpenStorage(path string, retentionMonths int) (*Storage, error) {
if err := fs.MkdirAllIfNotExist(idbSnapshotsPath); err != nil { if err := fs.MkdirAllIfNotExist(idbSnapshotsPath); err != nil {
return nil, fmt.Errorf("cannot create %q: %w", idbSnapshotsPath, err) return nil, fmt.Errorf("cannot create %q: %w", idbSnapshotsPath, err)
} }
idbCurr, idbPrev, err := openIndexDBTables(idbPath, s.metricIDCache, s.metricNameCache, &s.currHourMetricIDs, &s.prevHourMetricIDs) idbCurr, idbPrev, err := openIndexDBTables(idbPath, s.metricIDCache, s.metricNameCache, s.tsidCache, &s.currHourMetricIDs, &s.prevHourMetricIDs)
if err != nil { if err != nil {
return nil, fmt.Errorf("cannot open indexdb tables at %q: %w", idbPath, err) return nil, fmt.Errorf("cannot open indexdb tables at %q: %w", idbPath, err)
} }
@ -537,7 +537,7 @@ func (s *Storage) mustRotateIndexDB() {
// Create new indexdb table. // Create new indexdb table.
newTableName := nextIndexDBTableName() newTableName := nextIndexDBTableName()
idbNewPath := s.path + "/indexdb/" + newTableName idbNewPath := s.path + "/indexdb/" + newTableName
idbNew, err := openIndexDB(idbNewPath, s.metricIDCache, s.metricNameCache, &s.currHourMetricIDs, &s.prevHourMetricIDs) idbNew, err := openIndexDB(idbNewPath, s.metricIDCache, s.metricNameCache, s.tsidCache, &s.currHourMetricIDs, &s.prevHourMetricIDs)
if err != nil { if err != nil {
logger.Panicf("FATAL: cannot create new indexDB at %q: %s", idbNewPath, err) logger.Panicf("FATAL: cannot create new indexDB at %q: %s", idbNewPath, err)
} }
@ -876,11 +876,10 @@ func (s *Storage) DeleteMetrics(tfss []*TagFilters) (int, error) {
if err != nil { if err != nil {
return deletedCount, fmt.Errorf("cannot delete tsids: %w", err) return deletedCount, fmt.Errorf("cannot delete tsids: %w", err)
} }
// Reset MetricName->TSID cache in order to prevent from adding new data points // Do not reset MetricName->TSID cache in order to prevent from adding new data points
// to deleted time series in Storage.add. // to deleted time series in Storage.add, since it is already reset inside DeleteTSIDs.
s.tsidCache.Reset()
// Do not reset MetricID -> MetricName cache, since it must be used only // Do not reset MetricID->MetricName cache, since it must be used only
// after filtering out deleted metricIDs. // after filtering out deleted metricIDs.
return deletedCount, nil return deletedCount, nil
@ -1617,7 +1616,8 @@ func (s *Storage) putTSIDToCache(tsid *TSID, metricName []byte) {
s.tsidCache.Set(metricName, buf) s.tsidCache.Set(metricName, buf)
} }
func openIndexDBTables(path string, metricIDCache, metricNameCache *workingsetcache.Cache, currHourMetricIDs, prevHourMetricIDs *atomic.Value) (curr, prev *indexDB, err error) { func openIndexDBTables(path string, metricIDCache, metricNameCache, tsidCache *workingsetcache.Cache,
currHourMetricIDs, prevHourMetricIDs *atomic.Value) (curr, prev *indexDB, err error) {
if err := fs.MkdirAllIfNotExist(path); err != nil { if err := fs.MkdirAllIfNotExist(path); err != nil {
return nil, nil, fmt.Errorf("cannot create directory %q: %w", path, err) return nil, nil, fmt.Errorf("cannot create directory %q: %w", path, err)
} }
@ -1676,12 +1676,12 @@ func openIndexDBTables(path string, metricIDCache, metricNameCache *workingsetca
// Open the last two tables. // Open the last two tables.
currPath := path + "/" + tableNames[len(tableNames)-1] currPath := path + "/" + tableNames[len(tableNames)-1]
curr, err = openIndexDB(currPath, metricIDCache, metricNameCache, currHourMetricIDs, prevHourMetricIDs) curr, err = openIndexDB(currPath, metricIDCache, metricNameCache, tsidCache, currHourMetricIDs, prevHourMetricIDs)
if err != nil { if err != nil {
return nil, nil, fmt.Errorf("cannot open curr indexdb table at %q: %w", currPath, err) return nil, nil, fmt.Errorf("cannot open curr indexdb table at %q: %w", currPath, err)
} }
prevPath := path + "/" + tableNames[len(tableNames)-2] prevPath := path + "/" + tableNames[len(tableNames)-2]
prev, err = openIndexDB(prevPath, metricIDCache, metricNameCache, currHourMetricIDs, prevHourMetricIDs) prev, err = openIndexDB(prevPath, metricIDCache, metricNameCache, tsidCache, currHourMetricIDs, prevHourMetricIDs)
if err != nil { if err != nil {
curr.MustClose() curr.MustClose()
return nil, nil, fmt.Errorf("cannot open prev indexdb table at %q: %w", prevPath, err) return nil, nil, fmt.Errorf("cannot open prev indexdb table at %q: %w", prevPath, err)