[release-15.0] [bugfix] Allow VTExplain to handle shards that are not active during resharding (#11640) (#11652)

* VTexplain topology only uses serving shards

This addresses isse #11632 , which
causes vtexplain to sometimes give bad results if the keyspace is being
resharded, because sometimes it picks source shards and other times target
shards, for routing the query.

The issue is that the `VTExplain.buildTopolog()` adds both source and
destination shards to the map that holds shards per keyspace, when only one of
them is actually serving traffic at any point in time. Later on, vtexplain
loops over this map. Because looping over the map gives a non-deterministic
order, sometimes the results are correct, and sometimes incorrect - that is,
sometimes it gives the result of the shard that is serving, and other times,
the shard that is not serving.

This change ensures that only the serving shards are added to the shards per
keyspace map, thus avoiding the incorrect vtexplain.

Signed-off-by: Eduardo J. Ortega U <5791035+ejortegau@users.noreply.github.com>

* This addresses issue #11632 , which
causes vtexplain to sometimes give bad results if the keyspace is being
resharded, because sometimes it picks source shards and other times target
shards, for routing the query.

The issue is that the VTExplain.buildTopolog() adds both source and
destination shards to the map that holds shards per keyspace, when only one of
them is actually serving traffic at any point in time. Later on, vtexplain
loops over this map. Because looping over the map gives a non-deterministic
order, sometimes the results are correct, and sometimes incorrect - that is,
sometimes it gives the result of the shard that is serving, and other times,
the shard that is not serving.

This change ensures that only the serving shards are added to the shards per
keyspace map, thus avoiding the incorrect vtexplain.

Signed-off-by: Eduardo J. Ortega U <5791035+ejortegau@users.noreply.github.com>

* Fix check_make_vtadmin_authz_testgen

Signed-off-by: Eduardo J. Ortega U <5791035+ejortegau@users.noreply.github.com>

* empty commit to trigger CI

Signed-off-by: Andres Taylor <andres@planetscale.com>

Signed-off-by: Eduardo J. Ortega U <5791035+ejortegau@users.noreply.github.com>
Signed-off-by: Andres Taylor <andres@planetscale.com>
Co-authored-by: Eduardo J. Ortega U <5791035+ejortegau@users.noreply.github.com>
Co-authored-by: Andres Taylor <andres@planetscale.com>

go/vt/vtadmin/api_authz_test.go

@@ -3210,6 +3210,7 @@ func testClusters(t testing.TB) []*cluster.Cluster {
 									Name:     "-",
 									Shard: &topodatapb.Shard{
 										KeyRange:         &topodatapb.KeyRange{},
+										IsPrimaryServing: true,
 									},
 								},
 							},

go/vt/vtadmin/testutil/authztestgen/config.json

@@ -23,7 +23,7 @@
                 {
                     "field": "FindAllShardsInKeyspaceResults",
                     "type": "map[string]struct{\nResponse *vtctldatapb.FindAllShardsInKeyspaceResponse\nError error}",
-                    "value": "\"test\": {\nResponse: &vtctldatapb.FindAllShardsInKeyspaceResponse{\nShards: map[string]*vtctldatapb.Shard{\n\"-\": {\nKeyspace: \"test\",\nName: \"-\",\nShard: &topodatapb.Shard{\nKeyRange: &topodatapb.KeyRange{},\n},\n},\n},\n},\n},"
+                    "value": "\"test\": {\nResponse: &vtctldatapb.FindAllShardsInKeyspaceResponse{\nShards: map[string]*vtctldatapb.Shard{\n\"-\": {\nKeyspace: \"test\",\nName: \"-\",\nShard: &topodatapb.Shard{\nKeyRange: &topodatapb.KeyRange{},\nIsPrimaryServing: true,\n},\n},\n},\n},\n},"
                 },
                 {
                     "field": "GetBackupsResults",

go/vt/vtexplain/vtexplain_test.go

@@ -283,14 +283,14 @@ func TestJSONOutput(t *testing.T) {
 	}
 }
 
-func testShardInfo(ks, start, end string, t *testing.T) *topo.ShardInfo {
+func testShardInfo(ks, start, end string, primaryServing bool, t *testing.T) *topo.ShardInfo {
 	kr, err := key.ParseKeyRangeParts(start, end)
 	require.NoError(t, err)
 
 	return topo.NewShardInfo(
 		ks,
 		fmt.Sprintf("%s-%s", start, end),
-		&topodata.Shard{KeyRange: kr},
+		&topodata.Shard{KeyRange: kr, IsPrimaryServing: primaryServing},
 		&vtexplainTestTopoVersion{},
 	)
 }
@@ -304,14 +304,17 @@ func TestUsingKeyspaceShardMap(t *testing.T) {
 			testcase: "select-sharded-8",
 			ShardRangeMap: map[string]map[string]*topo.ShardInfo{
 				"ks_sharded": {
-					"-20":   testShardInfo("ks_sharded", "", "20", t),
-					"20-40": testShardInfo("ks_sharded", "20", "40", t),
-					"40-60": testShardInfo("ks_sharded", "40", "60", t),
-					"60-80": testShardInfo("ks_sharded", "60", "80", t),
-					"80-a0": testShardInfo("ks_sharded", "80", "a0", t),
-					"a0-c0": testShardInfo("ks_sharded", "a0", "c0", t),
-					"c0-e0": testShardInfo("ks_sharded", "c0", "e0", t),
-					"e0-":   testShardInfo("ks_sharded", "e0", "", t),
+					"-20":   testShardInfo("ks_sharded", "", "20", true, t),
+					"20-40": testShardInfo("ks_sharded", "20", "40", true, t),
+					"40-60": testShardInfo("ks_sharded", "40", "60", true, t),
+					"60-80": testShardInfo("ks_sharded", "60", "80", true, t),
+					"80-a0": testShardInfo("ks_sharded", "80", "a0", true, t),
+					"a0-c0": testShardInfo("ks_sharded", "a0", "c0", true, t),
+					"c0-e0": testShardInfo("ks_sharded", "c0", "e0", true, t),
+					"e0-":   testShardInfo("ks_sharded", "e0", "", true, t),
+					// Some non-serving shards below - these should never be in the output of vtexplain
+					"-80": testShardInfo("ks_sharded", "", "80", false, t),
+					"80-": testShardInfo("ks_sharded", "80", "", false, t),
 				},
 			},
 		},
@@ -321,11 +324,15 @@ func TestUsingKeyspaceShardMap(t *testing.T) {
 				// Have mercy on the poor soul that has this keyspace sharding.
 				// But, hey, vtexplain still works so they have that going for them.
 				"ks_sharded": {
-					"-80":   testShardInfo("ks_sharded", "", "80", t),
-					"80-90": testShardInfo("ks_sharded", "80", "90", t),
-					"90-a0": testShardInfo("ks_sharded", "90", "a0", t),
-					"a0-e8": testShardInfo("ks_sharded", "a0", "e8", t),
-					"e8-":   testShardInfo("ks_sharded", "e8", "", t),
+					"-80":   testShardInfo("ks_sharded", "", "80", true, t),
+					"80-90": testShardInfo("ks_sharded", "80", "90", true, t),
+					"90-a0": testShardInfo("ks_sharded", "90", "a0", true, t),
+					"a0-e8": testShardInfo("ks_sharded", "a0", "e8", true, t),
+					"e8-":   testShardInfo("ks_sharded", "e8", "", true, t),
+					// Plus some un-even shards that are not serving and which should never be in the output of vtexplain
+					"80-a0": testShardInfo("ks_sharded", "80", "a0", false, t),
+					"a0-a5": testShardInfo("ks_sharded", "a0", "a5", false, t),
+					"a5-":   testShardInfo("ks_sharded", "a5", "", false, t),
 				},
 			},
 		},

go/vt/vtexplain/vtexplain_vtgate.go

@@ -131,6 +131,14 @@ func (vte *VTExplain) buildTopology(opts *Options, vschemaStr string, ksShardMap
 		vte.explainTopo.KeyspaceShards[ks] = make(map[string]*topodatapb.ShardReference)
 
 		for _, shard := range shards {
+			// If the topology is in the middle of a reshard, there can be two shards covering the same key range (e.g.
+			// both source shard 80- and target shard 80-c0 cover the keyrange 80-c0). For the purposes of explain, we
+			// should only consider the one that is serving, hence we skip the ones not serving. Otherwise, vtexplain
+			// gives inconsistent results - sometimes it will route the query being explained to the source shard, and
+			// sometimes to the destination shard. See https://github.com/vitessio/vitess/issues/11632 .
+			if shardInfo, ok := ksShardMap[ks][shard.Name]; ok && !shardInfo.IsPrimaryServing {
+				continue
+			}
 			hostname := fmt.Sprintf("%s/%s", ks, shard.Name)
 			log.Infof("registering test tablet %s for keyspace %s shard %s", hostname, ks, shard.Name)