shipper: Be strict about upload order unless it's specified so & cut v0.13.0-rc.2 (#2765)

* shipper: Be strict about upload order unless it's specified so.

Signed-off-by: Bartlomiej Plotka <bwplotka@gmail.com>

* Cut 0.13.0-rc.2

Signed-off-by: Bartlomiej Plotka <bwplotka@gmail.com>

Author: bwplotka

CHANGELOG.md

@@ -11,7 +11,7 @@ We use *breaking* word for marking changes that are not backward compatible (rel
 
 ## Unreleased
 
-## [v0.13.0](https://github.com/thanos-io/thanos/releases/tag/v0.13.0) - 2020.06.15
+## [v0.13.0-rc.2](https://github.com/thanos-io/thanos/releases/tag/v0.13.0-rc.2) - 2020.06.15
 
 ### Fixed
 
@@ -26,6 +26,7 @@ We use *breaking* word for marking changes that are not backward compatible (rel
 - [#2416](https://github.com/thanos-io/thanos/pull/2416) Bucket: Fixed issue #2416 bug in `inspect --sort-by` doesn't work correctly in all cases.
 - [#2719](https://github.com/thanos-io/thanos/pull/2719) Query: `irate` and `resets` use now counter downsampling aggregations.
 - [#2705](https://github.com/thanos-io/thanos/pull/2705) minio-go: Added support for `af-south-1` and `eu-south-1` regions.
+- [#2753](https://github.com/thanos-io/thanos/issues/2753) Sidecar, Receive, Rule: Fixed possibility of out of order uploads in error cases. This could potentially cause Compactor to create overlapping blocks.
 
 ### Added
 

VERSION

@@ -1 +1 @@
-0.13.0
+0.13.0-rc.2

cmd/thanos/config.go

@@ -112,17 +112,23 @@ func (rc *reloaderConfig) registerFlag(cmd *kingpin.CmdClause) *reloaderConfig {
 }
 
 type shipperConfig struct {
-	uploadCompacted bool
-	ignoreBlockSize bool
+	uploadCompacted       bool
+	ignoreBlockSize       bool
+	allowOutOfOrderUpload bool
 }
 
 func (sc *shipperConfig) registerFlag(cmd *kingpin.CmdClause) *shipperConfig {
 	cmd.Flag("shipper.upload-compacted",
-		"If true sidecar will try to upload compacted blocks as well. Useful for migration purposes. Works only if compaction is disabled on Prometheus. Do it once and then disable the flag when done.").
+		"If true shipper will try to upload compacted blocks as well. Useful for migration purposes. Works only if compaction is disabled on Prometheus. Do it once and then disable the flag when done.").
 		Default("false").BoolVar(&sc.uploadCompacted)
 	cmd.Flag("shipper.ignore-unequal-block-size",
-		"If true sidecar will not require prometheus min and max block size flags to be set to the same value. Only use this if you want to keep long retention and compaction enabled on your Prometheus instance, as in the worst case it can result in ~2h data loss for your Thanos bucket storage.").
+		"If true shipper will not require prometheus min and max block size flags to be set to the same value. Only use this if you want to keep long retention and compaction enabled on your Prometheus instance, as in the worst case it can result in ~2h data loss for your Thanos bucket storage.").
 		Default("false").Hidden().BoolVar(&sc.ignoreBlockSize)
+	cmd.Flag("shipper.allow-out-of-order-uploads",
+		"If true, shipper will skip failed block uploads in the given iteration and retry later. This means that some newer blocks might be uploaded sooner than older blocks."+
+			"This can trigger compaction without those blocks and as a result will create an overlap situation. Set it to true if you have vertical compaction enabled and wish to upload blocks as soon as possible without caring"+
+			"about order.").
+		Default("false").Hidden().BoolVar(&sc.allowOutOfOrderUpload)
 	return sc
 }
 

cmd/thanos/receive.go

@@ -89,6 +89,12 @@ func registerReceive(m map[string]setupFunc, app *kingpin.Application) {
 
 	walCompression := cmd.Flag("tsdb.wal-compression", "Compress the tsdb WAL.").Default("true").Bool()
 
+	allowOutOfOrderUpload := cmd.Flag("shipper.allow-out-of-order-uploads",
+		"If true, shipper will skip failed block uploads in the given iteration and retry later. This means that some newer blocks might be uploaded sooner than older blocks."+
+			"This can trigger compaction without those blocks and as a result will create an overlap situation. Set it to true if you have vertical compaction enabled and wish to upload blocks as soon as possible without caring"+
+			"about order.").
+		Default("false").Hidden().Bool()
+
 	m[comp.String()] = func(g *run.Group, logger log.Logger, reg *prometheus.Registry, tracer opentracing.Tracer, _ <-chan struct{}, _ bool) error {
 		lset, err := parseFlagLabels(*labelStrs)
 		if err != nil {
@@ -157,6 +163,7 @@ func registerReceive(m map[string]setupFunc, app *kingpin.Application) {
 			*replicationFactor,
 			time.Duration(*forwardTimeout),
 			comp,
+			*allowOutOfOrderUpload,
 		)
 	}
 }
@@ -195,6 +202,7 @@ func runReceive(
 	replicationFactor uint64,
 	forwardTimeout time.Duration,
 	comp component.SourceStoreAPI,
+	allowOutOfOrderUpload bool,
 ) error {
 	logger = log.With(logger, "component", "receive")
 	level.Warn(logger).Log("msg", "setting up receive; the Thanos receive component is EXPERIMENTAL, it may break significantly without notice")
@@ -246,6 +254,7 @@ func runReceive(
 		lset,
 		tenantLabelName,
 		bkt,
+		allowOutOfOrderUpload,
 	)
 	writer := receive.NewWriter(log.With(logger, "component", "receive-writer"), dbs)
 	webHandler := receive.NewHandler(log.With(logger, "component", "receive-handler"), &receive.Options{

cmd/thanos/rule.go

@@ -114,6 +114,12 @@ func registerRule(m map[string]setupFunc, app *kingpin.Application) {
 	dnsSDResolver := cmd.Flag("query.sd-dns-resolver", "Resolver to use. Possible options: [golang, miekgdns]").
 		Default("golang").Hidden().String()
 
+	allowOutOfOrderUpload := cmd.Flag("shipper.allow-out-of-order-uploads",
+		"If true, shipper will skip failed block uploads in the given iteration and retry later. This means that some newer blocks might be uploaded sooner than older blocks."+
+			"This can trigger compaction without those blocks and as a result will create an overlap situation. Set it to true if you have vertical compaction enabled and wish to upload blocks as soon as possible without caring"+
+			"about order.").
+		Default("false").Hidden().Bool()
+
 	m[comp.String()] = func(g *run.Group, logger log.Logger, reg *prometheus.Registry, tracer opentracing.Tracer, reload <-chan struct{}, _ bool) error {
 		lset, err := parseFlagLabels(*labelStrs)
 		if err != nil {
@@ -197,6 +203,7 @@ func registerRule(m map[string]setupFunc, app *kingpin.Application) {
 			time.Duration(*dnsSDInterval),
 			*dnsSDResolver,
 			comp,
+			*allowOutOfOrderUpload,
 		)
 	}
 }
@@ -283,6 +290,7 @@ func runRule(
 	dnsSDInterval time.Duration,
 	dnsSDResolver string,
 	comp component.Component,
+	allowOutOfOrderUpload bool,
 ) error {
 	metrics := newRuleMetrics(reg)
 
@@ -615,7 +623,7 @@ func runRule(
 			}
 		}()
 
-		s := shipper.New(logger, reg, dataDir, bkt, func() labels.Labels { return lset }, metadata.RulerSource)
+		s := shipper.New(logger, reg, dataDir, bkt, func() labels.Labels { return lset }, metadata.RulerSource, allowOutOfOrderUpload)
 
 		ctx, cancel := context.WithCancel(context.Background())
 

cmd/thanos/sidecar.go

@@ -273,9 +273,9 @@ func runSidecar(
 
 			var s *shipper.Shipper
 			if conf.shipper.uploadCompacted {
-				s = shipper.NewWithCompacted(logger, reg, conf.tsdb.path, bkt, m.Labels, metadata.SidecarSource)
+				s = shipper.NewWithCompacted(logger, reg, conf.tsdb.path, bkt, m.Labels, metadata.SidecarSource, conf.shipper.allowOutOfOrderUpload)
 			} else {
-				s = shipper.New(logger, reg, conf.tsdb.path, bkt, m.Labels, metadata.SidecarSource)
+				s = shipper.New(logger, reg, conf.tsdb.path, bkt, m.Labels, metadata.SidecarSource, conf.shipper.allowOutOfOrderUpload)
 			}
 
 			return runutil.Repeat(30*time.Second, ctx.Done(), func() error {

docs/components/sidecar.md

@@ -151,7 +151,7 @@ Flags:
                                  details:
                                  https://thanos.io/storage.md/#configuration
       --shipper.upload-compacted
-                                 If true sidecar will try to upload compacted
+                                 If true shipper will try to upload compacted
                                  blocks as well. Useful for migration purposes.
                                  Works only if compaction is disabled on
                                  Prometheus. Do it once and then disable the

docs/operating/troubleshooting.md

@@ -7,7 +7,6 @@ slug: /troubleshooting.md
 
 # Troubleshooting; Common cases
 
-
 ## Overlaps
 
 **Block overlap**: Set of blocks with exactly the same external labels in meta.json and for the same time or overlapping time period.
@@ -29,13 +28,15 @@ Checking producers log for such ULID, and checking meta.json (e.g if sample stat
 
 ### Reasons
 
+- You are running Thanos (sidecar, ruler or receive) older than 0.13.0. During transient upload errors there is a possibility to have overlaps caused by the compactor not being aware of all blocks See: [this](https://github.com/thanos-io/thanos/issues/2753)
 - Misconfiguraiton of sidecar/ruler: Same external labels or no external labels across many block producers.
 - Running multiple compactors for single block "stream", even for short duration.
 - Manually uploading blocks to the bucket.
 - Eventually consistent block storage until we fully implement [RW for bucket](https://thanos.io/proposals/201901-read-write-operations-bucket.md)
 
 ### Solutions
 
+- Upgrade sidecar, ruler and receive to 0.13.0+
 - Compactor can be blocked for some time, but if it is urgent. Mitigate by removing overlap or better: Backing up somewhere else (you can rename block ULID to non-ulid).
 - Who uploaded the block? Search for logs with this ULID across all sidecars/rulers. Check access logs to object storage. Check debug/metas or meta.json of problematic block to see how blocks looks like and what is the `source`.
 - Determine what you misconfigured.

pkg/receive/multitsdb.go

@@ -38,8 +38,9 @@ type MultiTSDB struct {
 	labels          labels.Labels
 	bucket          objstore.Bucket
 
-	mtx     *sync.RWMutex
-	tenants map[string]*tenant
+	mtx                   *sync.RWMutex
+	tenants               map[string]*tenant
+	allowOutOfOrderUpload bool
 }
 
 func NewMultiTSDB(
@@ -50,21 +51,23 @@ func NewMultiTSDB(
 	labels labels.Labels,
 	tenantLabelName string,
 	bucket objstore.Bucket,
+	allowOutOfOrderUpload bool,
 ) *MultiTSDB {
 	if l == nil {
 		l = log.NewNopLogger()
 	}
 
 	return &MultiTSDB{
-		dataDir:         dataDir,
-		logger:          l,
-		reg:             reg,
-		tsdbOpts:        tsdbOpts,
-		mtx:             &sync.RWMutex{},
-		tenants:         map[string]*tenant{},
-		labels:          labels,
-		tenantLabelName: tenantLabelName,
-		bucket:          bucket,
+		dataDir:               dataDir,
+		logger:                l,
+		reg:                   reg,
+		tsdbOpts:              tsdbOpts,
+		mtx:                   &sync.RWMutex{},
+		tenants:               map[string]*tenant{},
+		labels:                labels,
+		tenantLabelName:       tenantLabelName,
+		bucket:                bucket,
+		allowOutOfOrderUpload: allowOutOfOrderUpload,
 	}
 }
 
@@ -256,6 +259,7 @@ func (t *MultiTSDB) getOrLoadTenant(tenantID string, blockingStart bool) (*tenan
 				t.bucket,
 				func() labels.Labels { return lbls },
 				metadata.ReceiveSource,
+				t.allowOutOfOrderUpload,
 			)
 		}
 

pkg/receive/multitsdb_test.go

@@ -43,6 +43,7 @@ func TestMultiTSDB(t *testing.T) {
 			labels.FromStrings("replica", "01"),
 			"tenant_id",
 			nil,
+			false,
 		)
 		defer testutil.Ok(t, m.Flush())
 
@@ -109,6 +110,7 @@ func TestMultiTSDB(t *testing.T) {
 			labels.FromStrings("replica", "01"),
 			"tenant_id",
 			nil,
+			false,
 		)
 		defer testutil.Ok(t, m.Flush())