Beispiel #1
0
// FIXME(msolomon) This validate presumes the master is up and running.
// Even when that isn't true, there are validation processes that might be valuable.
func (wr *Wrangler) validateShard(keyspace, shard string, pingTablets bool, wg *sync.WaitGroup, results chan<- error) {
	shardInfo, err := wr.ts.GetShard(keyspace, shard)
	if err != nil {
		results <- fmt.Errorf("TopologyServer.GetShard(%v, %v) failed: %v", keyspace, shard, err)
		return
	}

	aliases, err := topo.FindAllTabletAliasesInShard(wr.ctx, wr.ts, keyspace, shard)
	if err != nil {
		results <- fmt.Errorf("TopologyServer.FindAllTabletAliasesInShard(%v, %v) failed: %v", keyspace, shard, err)
		return
	}

	tabletMap, _ := topo.GetTabletMap(wr.ctx, wr.ts, aliases)

	var masterAlias topo.TabletAlias
	for _, alias := range aliases {
		tabletInfo, ok := tabletMap[alias]
		if !ok {
			results <- fmt.Errorf("tablet %v not found in map", alias)
			continue
		}
		if tabletInfo.Parent.Uid == topo.NO_TABLET {
			if masterAlias.Cell != "" {
				results <- fmt.Errorf("tablet %v already has a master %v", alias, masterAlias)
			} else {
				masterAlias = alias
			}
		}
	}

	if masterAlias.Cell == "" {
		results <- fmt.Errorf("no master for shard %v/%v", keyspace, shard)
	} else if shardInfo.MasterAlias != masterAlias {
		results <- fmt.Errorf("master mismatch for shard %v/%v: found %v, expected %v", keyspace, shard, masterAlias, shardInfo.MasterAlias)
	}

	for _, alias := range aliases {
		wg.Add(1)
		go func(alias topo.TabletAlias) {
			defer wg.Done()
			if err := topo.Validate(wr.ts, alias); err != nil {
				results <- fmt.Errorf("Validate(%v) failed: %v", alias, err)
			} else {
				wr.Logger().Infof("tablet %v is valid", alias)
			}
		}(alias)
	}

	if pingTablets {
		wr.validateReplication(shardInfo, tabletMap, results)
		wr.pingTablets(tabletMap, wg, results)
	}

	return
}
Beispiel #2
0
func (agent *ActionAgent) verifyTopology() error {
	tablet := agent.Tablet()
	if tablet == nil {
		return fmt.Errorf("agent._tablet is nil")
	}

	if err := topo.Validate(agent.TopoServer, agent.TabletAlias); err != nil {
		// Don't stop, it's not serious enough, this is likely transient.
		log.Warningf("tablet validate failed: %v %v", agent.TabletAlias, err)
	}

	return nil
}
Beispiel #3
0
// Validate all tablets in all discoverable cells, even if they are
// not in the replication graph.
func (wr *Wrangler) validateAllTablets(wg *sync.WaitGroup, results chan<- error) {
	cellSet := make(map[string]bool, 16)

	keyspaces, err := wr.ts.GetKeyspaces()
	if err != nil {
		results <- fmt.Errorf("TopologyServer.GetKeyspaces failed: %v", err)
		return
	}
	for _, keyspace := range keyspaces {
		shards, err := wr.ts.GetShardNames(keyspace)
		if err != nil {
			results <- fmt.Errorf("TopologyServer.GetShardNames(%v) failed: %v", keyspace, err)
			return
		}

		for _, shard := range shards {
			aliases, err := topo.FindAllTabletAliasesInShard(wr.ctx, wr.ts, keyspace, shard)
			if err != nil {
				results <- fmt.Errorf("TopologyServer.FindAllTabletAliasesInShard(%v, %v) failed: %v", keyspace, shard, err)
				return
			}
			for _, alias := range aliases {
				cellSet[alias.Cell] = true
			}
		}
	}

	for cell := range cellSet {
		aliases, err := wr.ts.GetTabletsByCell(cell)
		if err != nil {
			results <- fmt.Errorf("TopologyServer.GetTabletsByCell(%v) failed: %v", cell, err)
			continue
		}

		for _, alias := range aliases {
			wg.Add(1)
			go func(alias topo.TabletAlias) {
				defer wg.Done()
				if err := topo.Validate(wr.ts, alias); err != nil {
					results <- fmt.Errorf("Validate(%v) failed: %v", alias, err)
				} else {
					wr.Logger().Infof("tablet %v is valid", alias)
				}
			}(alias)
		}
	}
}