[How to use regexp in @filter]: <regexp use>

[WSC998]

Question.

I had about 50million nodes，if I use ‘regexp’ like “{
resources(func:regexp(name, /abc/i) )@filter(eq(workspace_key, "def")) {
id
name
resource_key
}
}”，
it will be fast with in 1 second,however,
if I use it like "{
resources(func: eq(workspace_key, "def"))@filter(regexp(name, /abc/i)) {
id
name
resource_key
}
}",
it will be useless and overtime. Most of the time I have to spell out a complex logical combination, so it's impossible to put a regexp in "func" but filter.
So I really wanna know whether the index is invalid when I use 'regexp' in @filter？
And if somebody has a good idea to solve this problem?

[mangalaman93]

We don't have a query planner yet, Dgraph may not do a good job in figuring out in what order to execute the filters. Is it possible for you to share the data? It may still be worth looking into this.

[WSC998]

`package main

import (
"context"
"encoding/json"
"fmt"
"github.com/dgraph-io/dgo"
"github.com/dgraph-io/dgo/protos/api"
"google.golang.org/grpc"
"strconv"
"sync"
)

func nodeExists(dg *dgo.Dgraph, id string) (bool, error) {
query := fmt.Sprintf({ data(func: eq(id, "%s")) { uid } }, id)
resp, err := dg.NewTxn().Query(context.Background(), query)
if err != nil {
return false, err
}
var result struct {
Data []struct {
UID string json:"uid"
} json:"data"
}
if err := json.Unmarshal(resp.Json, &result); err != nil {
return false, err
}
return len(result.Data) > 0, nil
}

func saveNode(dg *dgo.Dgraph, j int) error {
id := "test_text" + strconv.Itoa(j)
name := name + id
node := map[string]interface{}{
"name": name,
"workspace_key": "_xGWV7",
"create_by": "xxw" + strconv.Itoa(j),
"qa": "wws",
"id": id,
}

ctx := context.Background()

nodeJSON, err := json.Marshal(node)
if err != nil {
    return err
}

mutation := &api.Mutation{
    CommitNow: true,

    SetJson: nodeJSON,
}

_, err = dg.NewTxn().Mutate(ctx, mutation)
if err != nil {
    return err
}
fmt.Printf("Created  node %s \n", id)
return nil

}

func saveRelation(dg *dgo.Dgraph, srcID, dstID string) error {
addRelationQuery := { left as var(func: eq(id, "%s")) right as var(func: eq(id, "%s")) }
addQuads := fmt.Sprintf(uid(left) <%s> uid(right) . , "test_parent_to_child")
addReq := &api.Request{
CommitNow: true,
Query: fmt.Sprintf(addRelationQuery, srcID, dstID),
Mutations: []*api.Mutation{
&api.Mutation{
SetNquads: []byte(addQuads),
},
},
}
_, err := dg.NewTxn().Do(context.Background(), addReq)
if err != nil {
return err
}

return nil

}

func main() {
conn, err := grpc.Dial("127.0.0.1:9080", grpc.WithInsecure())
if err != nil {
fmt.Println("Error connecting to Dgraph server:", err)
return
}
defer conn.Close()

dg := dgo.NewDgraphClient(api.NewDgraphClient(conn))

concurrency := 100
totalNodes := 1000000
startIndex := 0
nodeBatchSize := (totalNodes - startIndex) / concurrency

var wg sync.WaitGroup
wg.Add(concurrency)

for i := 0; i < concurrency; i++ {
    go func(workerID int) {
        defer wg.Done()

        start := startIndex + workerID*nodeBatchSize
        end := start + nodeBatchSize
        if workerID == concurrency-1 {
            end = totalNodes
        }

        for j := start; j < end; j++ {
            //id := "wsc_test_text" + strconv.Itoa(j)
            //if isExist, _ := nodeExists(dg, id); !isExist {
            err = saveNode(dg, j)
            if err != nil {
                fmt.Println("Error saving node:", err)
                return
            }
            //}
        }
    }(i)
}
//
wg.Wait()
fmt.Println("Node creation completed!")

}
`
It's my easy demo to create data, you can change "totalNodes " to decide how many nodes you want to generate.
And I set predicate "workspace_key" with "exact" and others set "trigram" and "term".

[damonfeldman]

Definitely a problem. Regexp in filter should use the index. See also this discuss thread for same issue: https://discuss.dgraph.io/t/how-to-use-regexp-in-filter/18901/8 .

My guess is that when regexp was allowed in the filter, someone mistakenly or naively thought it would not need to be indexed. So it's a feature that you can post-filter with regexp, but misleading that it does not use the filter. I suspect GraphQL will build this underlying (non-optimized) DQL structure under the hood, making this more impactful for those users.

[github-actions]

This issue has been stale for 60 days and will be closed automatically in 7 days. Comment to keep it open.