invalid memory address or nil pointer dereference
SIGSEGV: segmentation violation code=0x1 addr=0x0 pc=0x14fb4be
upsert{
query{
v as var(func: eq(id,12)) # this one has 1 000 000 followers
}
mutation {
delete{
uid(v) <followers> * .
}
}
}
followers has count and reverse index.
code at this line
Dgraph version : v20.07.0-12-g681fe9116
Dgraph codename : shuri-mod
Go version : go1.14.1
@chewxy we use twitter data:
wget http://an.kaist.ac.kr/\~haewoon/release/twitter_social_graph/twitter_rv.tar.gz
# split by line
split -l 54535107 twitter_rv.net
then use the go code to convert csv data to rdf
package main
import (
"path/filepath"
"os"
"fmt"
"flag"
"bufio"
"io"
"strings"
)
func getFilelist(path string) []string {
files := []string{}
err := filepath.Walk(path, func(path string, f os.FileInfo, err error) error {
if f == nil {
return err
}
if f.IsDir() {
return nil
}
files = append(files,path)
return nil
})
if err != nil {
fmt.Printf("filepath.Walk() returned %v\n", err)
}
return files
}
func rdf(f string, separator string, ch chan string) {
rv, err := os.Open(f)
if err != nil {
fmt.Println("open file err=", err)
rv.Close()
return
}
defer rv.Close()
out, _ := os.OpenFile(f+".rdf", os.O_RDWR|os.O_CREATE|os.O_APPEND, 0644)
reader := bufio.NewReader(rv)
wirter := bufio.NewWriter(out)
nodes := make(map[string]bool)
//循环的读取文件的内容
errcount := 0
for {
str, err := reader.ReadString('\n') // 读到一个换行就结束
if err == io.EOF { // io.EOF表示文件的末尾
break
}
//输出内容
line := strings.Split(str, separator)
if len(line) != 2 {
errcount++
fmt.Printf("from now on find %d errors in %s\n", errcount, f)
continue
}
src := strings.Trim(line[0], "\n")
dst := strings.Trim(line[1], "\n")
tpl := "_:v%s <id> \"%s\" . \n_:v%s <dgraph.type> \"twitter_user\" .\n"
if !nodes[src] {
lout := fmt.Sprintf(tpl, src, src, src)
wirter.Write([]byte(lout))
nodes[src] = true
}
if !nodes[dst] {
lout := fmt.Sprintf(tpl, dst, dst, dst)
wirter.Write([]byte(lout))
nodes[dst] = true
}
lout := fmt.Sprintf("_:v%s <followers> _:v%s . \n", src, dst)
wirter.Write([]byte(lout))
}
wirter.Flush()
out.Close()
ch <- f + ".rdf is ok!"
}
func main(){
flag.Parse()
separatorKey := flag.Arg(0)
root := flag.Arg(1)
separatorMap := map[string]string{"s": " ", "t": "\t", "c": ","}
separator := separatorMap[separatorKey]
if 0 == len(separator) {
fmt.Printf("Please specify the separator！\n t for Tab \n s for Space \n c for comma \n such as: ./gocsv2rdf s ./test \n")
return
}
if 0 == len(root) {
fmt.Printf("Please specify the CSV folder！\n such as: ./gocsv2rdf s ./test \n")
return
}
files :=getFilelist(root)
fmt.Printf("%v\n", files)
count := len(files)
if 0 == count {
fmt.Printf("Please specify the correct CSV folder！\n")
return
}
ch := make(chan string, count)
for i:= 0;i<count;i++{
name := files[i]
fmt.Printf("Runninng for %s\n", name)
//rdf(name, separator, ch)
go rdf(name, separator, ch) //内存足够大就用这行代码,并行处理
}
i := 0
for x := range ch {
fmt.Println(x)
i++
if i == int(count) {
break
}
}
fmt.Println("Over!")
}
then bulk load or live load into dgraph.
The schema is
id: int @index(int) .
followers: [uid] .
type twitter_user {
id
followers
}
Hey @BlankRain, can you show me the complete stack trace of the crash? I’m trying to figure out what sequence of function calls led to this crash.
@BlankRain It looks like you’re running a modified version of dgraph
Dgraph version : v20.07.0-12-g681fe9116
Dgraph codename : shuri-mod
Go version : go1.14.1
Can you try running on a released version of dgraph? I don’t have the code you’re running and I cannot debug it 
.
Hi
The code based on the release of v20.07.0 .
I modified the code to fix some bugs .
for example:
Bulk loader crashes during reduce phase - #26 by BlankRain
The git log ,I based is this one.
commit 7431be0dce0ffc42e3ff3e31b39fcdd5505e2bb3
Author: Martin Martinez Rivera <martinmr@dgraph.io>
Date: Mon Aug 3 11:54:25 2020 -0700
add cluster lables to the jaeger containers (#5951) (#6009)
Thanks @BlankRain. So the call to posting() is failing because uidPosting is nil on line 273.
But it looks like we initialize the it.uidPosting in init which is called by iterate
@BlankRain would you be able to share your dataset? We can try to reproduce it but I doubt if we can easily reproduce it on our end. If you can share your dataset, we can look at it to figure out what’s causing this.
@animesh2049 Do you see any reason pb.Posting can be nil here? The cluster is running in ludicrous mode.
I can share my dataset. It may be too huge to copy.
My dataset is based on the twitter data.
I can show you how to produce the data.
wget http://an.kaist.ac.kr/\~haewoon/release/twitter_social_graph/twitter_rv.tar.gz
tar xvf twitter_rv.tar.gz
split -l 54535107 twitter_rv.net
mkdir test
mv x* test
rm test/*.rdf
./gocsv2rdf t test
# then copy test/*.rdf out for bulk load and start the dgraph cluster
the source code of gocsv2rdf is here
package main
import (
"path/filepath"
"os"
"fmt"
"flag"
"bufio"
"io"
"strings"
)
func getFilelist(path string) []string {
files := []string{}
err := filepath.Walk(path, func(path string, f os.FileInfo, err error) error {
if f == nil {
return err
}
if f.IsDir() {
return nil
}
files = append(files,path)
return nil
})
if err != nil {
fmt.Printf("filepath.Walk() returned %v\n", err)
}
return files
}
func rdf(f string, separator string, ch chan string) {
rv, err := os.Open(f)
if err != nil {
fmt.Println("open file err=", err)
rv.Close()
return
}
defer rv.Close()
out, _ := os.OpenFile(f+".rdf", os.O_RDWR|os.O_CREATE|os.O_APPEND, 0644)
reader := bufio.NewReader(rv)
wirter := bufio.NewWriter(out)
nodes := make(map[string]bool)
//循环的读取文件的内容
errcount := 0
for {
str, err := reader.ReadString('\n') // 读到一个换行就结束
if err == io.EOF { // io.EOF表示文件的末尾
break
}
//输出内容
line := strings.Split(str, separator)
if len(line) != 2 {
errcount++
fmt.Printf("from now on find %d errors in %s\n", errcount, f)
continue
}
src := strings.Trim(line[0], "\n")
dst := strings.Trim(line[1], "\n")
tpl := "_:v%s <id> \"%s\" . \n_:v%s <dgraph.type> \"twitter_user\" .\n"
if !nodes[src] {
lout := fmt.Sprintf(tpl, src, src, src)
wirter.Write([]byte(lout))
nodes[src] = true
}
if !nodes[dst] {
lout := fmt.Sprintf(tpl, dst, dst, dst)
wirter.Write([]byte(lout))
nodes[dst] = true
}
lout := fmt.Sprintf("_:v%s <followers> _:v%s . \n", src, dst)
wirter.Write([]byte(lout))
}
wirter.Flush()
out.Close()
ch <- f + ".rdf is ok!"
}
func main(){
flag.Parse()
separatorKey := flag.Arg(0)
root := flag.Arg(1)
separatorMap := map[string]string{"s": " ", "t": "\t", "c": ","}
separator := separatorMap[separatorKey]
if 0 == len(separator) {
fmt.Printf("Please specify the separator！\n t for Tab \n s for Space \n c for comma \n such as: ./gocsv2rdf s ./test \n")
return
}
if 0 == len(root) {
fmt.Printf("Please specify the CSV folder！\n such as: ./gocsv2rdf s ./test \n")
return
}
files :=getFilelist(root)
fmt.Printf("%v\n", files)
count := len(files)
if 0 == count {
fmt.Printf("Please specify the correct CSV folder！\n")
return
}
ch := make(chan string, count)
for i:= 0;i<count;i++{
name := files[i]
fmt.Printf("Runninng for %s\n", name)
//rdf(name, separator, ch)
go rdf(name, separator, ch) //内存足够大就用这行代码,并行处理
}
i := 0
for x := range ch {
fmt.Println(x)
i++
if i == int(count) {
break
}
}
fmt.Println("Over!")
}
go build it then run it.
all this may need at least 30min or more.
upsert{
query{
v as var(func: eq(id,12)) # this one has 1 000 000 followers
}
mutation {
delete{
uid(v) <followers> * .
}
}
}
I do some delete in upsert block.
the reason pb.Posting can be nil may around here.
Thanks @BlankRain. I’ve accepted this as a bug. @animesh2049 will run some tests locally on your dataset and get back to you.
Hey @BlankRain I have pushed a change, commit id 7873f04087e671b5e45d87aaa1875680324a0f7b. Can you please apply this commit and see if you are still getting the error.
hi, which branch ?
you can just git checkout 7873f040
ok,thanks ,let me try
just post the related pr here: Initialize posting list in moveToNextPart by animesh2049 · Pull Request #6560 · dgraph-io/dgraph · GitHub
Sorry , we didn’t finish the test yet. Our cluster is broken. I will let you know when we finish the test.