WikiData Import 2020-07-15

From BITPlan Wiki
Jump to navigation Jump to search

see also Get_your_own_copy_of_WikiData - this was the fifth attempt after four failures. The main success factor was to use a 4 TB SSD disk which was kindly supplied by the ConfIDent project

Environment

  1. Mac Pro Mid 2010
  2. 12 core 3.46 GHz
  3. 64 GB RAM
  4. macOS High Sierra 10.13.6
  5. Source Disk: 4 TB 7200 rpm hard disk WD Gold WDC WD4002FYYZ Blackmagic speed rating: 175 MB/s write 175 MB/s read
  6. Target Disk: 4 TB SSD Samsung 860 EVO Blackmagic speed rating: 257 MB/s write 270 MB/s read
  7. java -version
    openjdk version "11.0.5" 2019-10-15
    OpenJDK Runtime Environment AdoptOpenJDK (build 11.0.5+10)
    OpenJDK 64-Bit Server VM AdoptOpenJDK (build 11.0.5+10, mixed mode)
    

Summary

  1. trying to replicate success story of https://issues.apache.org/jira/projects/JENA/issues/JENA-1909
  2. download of 24 GB took some 90 min
  3. unzipping to to some 670 GB took some 4 h 30 min
  4. Import took 4 d 14 h
  5. resulting TDB store is 700 GB

Import Timing

The diagram shows the number of triples imported versus number of triples per batch/avg. The x and y scales are both logarithmic. Even with an SSD harddisk there is a logarithmic degradation of import speed. WikidataimportStatistics2020-07-26.png

Test Query

SELECT (COUNT(*) as ?Triples) WHERE { ?s ?p ?o}

result

"5250681892"^^xsd:integer

log for query

14:51:11 INFO  Fuseki          :: [7] POST http://localhost:3030/wikidata
14:51:11 INFO  Fuseki          :: [7] Query = SELECT (COUNT(*) as ?Triples) WHERE { ?s ?p ?o}
16:23:07 INFO  Fuseki          :: [7] 200 OK (5.516,224 s)

same query on a copy / rotating disk

06:34:25 INFO  Server          :: Started 2020/07/30 06:34:25 CEST on port 3030
06:34:41 INFO  Fuseki          :: [5] POST http://jena.bitplan.com/wikidata
06:34:41 INFO  Fuseki          :: [5] Query = SELECT (COUNT(*) as ?Triples) WHERE { ?s ?p ?o}
15:19:43 INFO  Fuseki          :: [5] 200 OK (31,501.769 s)

Download and unpack

This download was done with the "truthy" dataset.

wget https://dumps.wikimedia.org/wikidatawiki/entities/latest-truthy.nt.bz2
--2020-07-15 15:24:25--  https://dumps.wikimedia.org/wikidatawiki/entities/latest-truthy.nt.bz2
Resolving dumps.wikimedia.org (dumps.wikimedia.org)... 2620::861:1:208:80:154:7, 208.80.154.7
Connecting to dumps.wikimedia.org (dumps.wikimedia.org)|2620::861:1:208:80:154:7|:443... connected.
HTTP request sent, awaiting response... 200 OK
Length: 25868964531 (24G) [application/octet-stream]
Saving to: ‘latest-truthy.nt.bz2’
latest-truthy.nt.bz   2%[                    ] 546.01M  4.66MB/s    eta 85m 57s
ls -l latest-truthy.nt.bz2 
-rw-r--r--  1 wf  admin  25868964531 Jul 11 23:33 latest-truthy.nt.bz2
bzip2 -dk latest-truthy.nt.bz2
ls -l latest-truthy.nt
-rw-------  1 wf  admin  671598919680 Jul 15 21:15 latest-truthy.nt
zeus:wikidata wf$ls -l latest-truthy.nt
-rw-r--r--  1 wf  admin  671749317281 Jul 11 23:33 latest-truthy.nt
start and progress
nohup ./wikidata2jena&
tail -f tdb2-err.log 
21:16:57 INFO  loader          :: Loader = LoaderParallel
21:16:57 INFO  loader          :: Start: latest-truthy.nt
21:17:00 INFO  loader          :: Add: 500.000 latest-truthy.nt (Batch: 151.883 / Avg: 151.883)
21:17:02 INFO  loader          :: Add: 1.000.000 latest-truthy.nt (Batch: 243.309 / Avg: 187.020)
...
21:33:21 INFO  loader          :: Add: 100.000.000 latest-truthy.nt (Batch: 209.292 / Avg: 101.592)
21:33:21 INFO  loader          ::   Elapsed: 984,33 seconds [2020/07/15 21:33:21 MESZ]
...
22:41:36 INFO  loader          :: Add: 500.000.000 latest-truthy.nt (Batch: 54.153 / Avg: 98.446)
22:41:36 INFO  loader          ::   Elapsed: 5.078,89 seconds [2020/07/15 22:41:36 MESZ]
...
02:55:36 INFO  loader          :: Add: 1.000.000.000 latest-truthy.nt (Batch: 21.504 / Avg: 49.215)
02:55:36 INFO  loader          ::   Elapsed: 20.318,94 seconds [2020/07/16 02:55:36 MESZ]
...
13:47:17 INFO  loader          :: Add: 2.000.000.000 latest-truthy.nt (Batch: 32.036 / Avg: 33.658)
13:47:17 INFO  loader          ::   Elapsed: 59.420,03 seconds [2020/07/16 13:47:17 MESZ]
...
06:10:12 INFO  loader          :: Add: 3.000.000.000 latest-truthy.nt (Batch: 10.900 / Avg: 25.338)
06:10:13 INFO  loader          ::   Elapsed: 118.395,31 seconds [2020/07/17 06:10:12 MESZ]
...
09:33:14 INFO  loader          :: Add: 4.000.000.000 latest-truthy.nt (Batch: 11.790 / Avg: 18.435)
09:33:14 INFO  loader          ::   Elapsed: 216.976,77 seconds [2020/07/18 09:33:14 MESZ]
...
00:55:21 INFO  loader          :: Add: 5.000.000.000 latest-truthy.nt (Batch: 4.551 / Avg: 13.939)
00:55:21 INFO  loader          ::   Elapsed: 358.703,75 seconds [2020/07/20 00:55:21 MESZ]
...
11:02:06 INFO  loader          :: Add: 5.253.500.000 latest-truthy.nt (Batch: 10.555 / Avg: 13.296)
11:02:38 INFO  loader          :: Finished: latest-truthy.nt: 5.253.753.313 tuples in 395140,91s (Avg: 13.295)
11:05:27 INFO  loader          :: Finish - index SPO
11:08:24 INFO  loader          :: Finish - index POS
11:08:59 INFO  loader          :: Finish - index OSP
11:08:59 INFO  loader          :: Time = 395.522,378 seconds : Triples = 5.253.753.313 : Rate = 13.283 /s
tail -f nohup.out 
x apache-jena-3.16.0/lib-src/jena-arq-3.16.0-sources.jar
x apache-jena-3.16.0/lib-src/jena-core-3.16.0-sources.jar
x apache-jena-3.16.0/lib-src/jena-tdb-3.16.0-sources.jar
x apache-jena-3.16.0/lib-src/jena-rdfconnection-3.16.0-sources.jar
x apache-jena-3.16.0/lib-src/jena-shacl-3.16.0-sources.jar
x apache-jena-3.16.0/lib-src/jena-cmds-3.16.0-sources.jar
creating /Volumes/Torterra/wikidata/data directory
creating temporary directory /Volumes/Torterra/wikidata/tmp
start loading latest-truthy.nt to /Volumes/Torterra/wikidata/data at 2020-07-15T19:16:54Z
finished loading latest-truthy.nt to /Volumes/Torterra/wikidata/data at 2020-07-20T09:09:04Z

Trivia

Torterra is a Pokemon 250px-389Torterra.png

Scripts

wikidata2jena

#!/bin/bash
# WF 2020-05-10

# global settings
jena=apache-jena-3.16.0
tgz=$jena.tar.gz
jenaurl=http://mirror.easyname.ch/apache/jena/binaries/$tgz
base=/Volumes/Torterra/wikidata
data=$base/data
tdbloader=$jena/bin/tdb2.tdbloader

getjena() {
# download
if [ ! -f $tgz ]
then
  echo "downloading $tgz from $jenaurl"
	wget $jenaurl
else
  echo "$tgz already downloaded"
fi
# unpack
if [ ! -d $jena ]
then
  echo "unpacking $jena from $tgz"
	tar xvzf $tgz
else
  echo "$jena already unpacked"
fi
# create data directory
if [ ! -d $data ]
then
  echo "creating $data directory"
  mkdir -p $data
else
  echo "$data directory already created"
fi
}

#
# show the given timestamp
#
timestamp() {
 local msg="$1"
 local ts=$(date -u +"%Y-%m-%dT%H:%M:%SZ")
 echo "$msg at $ts"
}

#
# load data for the given data dir and input
#
loaddata() {
	local data="$1"
	local input="$2"
  timestamp "start loading $input to $data"
  $tdbloader --loader=parallel --loc "$data" "$input" > tdb2-out.log 2> tdb2-err.log
	timestamp "finished loading $input to $data"
}

getjena
export TMPDIR=$base/tmp
if [ ! -d $TMPDIR ]
then
  echo "creating temporary directory $TMPDIR"
  mkdir $TMPDIR
else
  echo "using temporary directory $TMPDIR"
fi
loaddata $data latest-truthy.nt

fuseki

This scripts starts a Fuseki Server and makes it available for Querying the wikidata set imported

#!/bin/bash
# WF 2020-06-25
# Jena Fuseki server installation
# see https://jena.apache.org/documentation/fuseki2/fuseki-run.html
version=3.16.0
fuseki=apache-jena-fuseki-$version
if [ ! -d $fuseki ]
then
  if [ ! -f $fuseki.tar.gz ]
  then
    wget http://archive.apache.org/dist/jena/binaries/$fuseki.tar.gz
  else
    echo $fuseki.tar.gz already downloaded
  fi
  echo "unpacking $fuseki.tar.gz"
  tar xvfz $fuseki.tar.gz
else
  echo $fuseki already downloaded and unpacked
fi
cd $fuseki
java -jar fuseki-server.jar --tdb2 --loc=../data /wikidata

log2csv

This script extracts the import progress from the import log to be able to analyze it e.g. with a spreadsheet software as used for the diagram above.

#!/bin/bash
# create a CSV file from the loader log to analyze import speed
# degradation
# 2020-07-16 WF
target=capri:/bitplan/Projekte/2020/ConfIDent/wikidata/import

for log in tdb2-err.log 
do
  csv=$(echo $log | sed 's/.log/.csv/')
  echo "extracting $csv from $log"
  # Example
  # 07:33:45 INFO  loader          :: Add: 1.420.000.000 latest-truthy.nt (Batch: 54.036 / Avg: 38.369)
  # 07:33:45 INFO  loader          ::   Elapsed: 37.008,24 seconds [2020/07/16 07:33:45 MESZ]
  grep Elapsed -B1 $log | awk '
BEGIN {
  print ("elapsed;add;batch;avg")
}

/Add:/ {
   add=$6
   batch=$9
   avg=$12
   gsub ("\\)","",avg)
   #print add
   #print batch
   #print avg
   next
}
/Elapsed:/ {
  elapsed=$6
  printf("%s;%s;%s;%s\n",elapsed,add,batch,avg)
}
' > $csv
done

scp tdb2-err.csv $target/importStats2020-07.csv