CGSpace Notes

November, 2016

by Alan Orth in  

2016-11-01

  • Add dc.type to the output options for Atmire’s Listings and Reports module (#286)

Listings and Reports with output type

2016-11-02

  • Migrate DSpace Test to DSpace 5.5 (notes)
  • Run all updates on DSpace Test and reboot the server
  • Looks like the OAI bug from DSpace 5.1 that caused validation at Base Search to fail is now fixed and DSpace Test passes validation! (#63)
  • Indexing Discovery on DSpace Test took 332 minutes, which is like five times as long as it usually takes
  • At the end it appeared to finish correctly but there were lots of errors right after it finished:
2016-11-02 15:09:48,578 INFO  com.atmire.dspace.discovery.AtmireSolrService @ Wrote Collection: 10568/76454 to Index
2016-11-02 15:09:48,584 INFO  com.atmire.dspace.discovery.AtmireSolrService @ Wrote Community: 10568/3202 to Index
2016-11-02 15:09:48,589 INFO  com.atmire.dspace.discovery.AtmireSolrService @ Wrote Collection: 10568/76455 to Index
2016-11-02 15:09:48,590 INFO  com.atmire.dspace.discovery.AtmireSolrService @ Wrote Community: 10568/51693 to Index
2016-11-02 15:09:48,590 INFO  org.dspace.discovery.IndexClient @ Done with indexing
2016-11-02 15:09:48,600 INFO  com.atmire.dspace.discovery.AtmireSolrService @ Wrote Collection: 10568/76456 to Index
2016-11-02 15:09:48,613 INFO  org.dspace.discovery.SolrServiceImpl @ Wrote Item: 10568/55536 to Index
2016-11-02 15:09:48,616 INFO  com.atmire.dspace.discovery.AtmireSolrService @ Wrote Collection: 10568/76457 to Index
2016-11-02 15:09:48,634 ERROR com.atmire.dspace.discovery.AtmireSolrService @
java.lang.NullPointerException
        at org.dspace.discovery.SearchUtils.getDiscoveryConfiguration(SourceFile:57)
        at org.dspace.discovery.SolrServiceImpl.buildDocument(SolrServiceImpl.java:824)
        at com.atmire.dspace.discovery.AtmireSolrService.indexContent(AtmireSolrService.java:821)
        at com.atmire.dspace.discovery.AtmireSolrService.updateIndex(AtmireSolrService.java:898)
        at org.dspace.discovery.SolrServiceImpl.createIndex(SolrServiceImpl.java:370)
        at org.dspace.storage.rdbms.DatabaseUtils$ReindexerThread.run(DatabaseUtils.java:945)
  • DSpace is still up, and a few minutes later I see the default DSpace indexer is still running
  • Sure enough, looking back before the first one finished, I see output from both indexers interleaved in the log:
2016-11-02 15:09:28,545 INFO  org.dspace.discovery.SolrServiceImpl @ Wrote Item: 10568/47242 to Index
2016-11-02 15:09:28,633 INFO  org.dspace.discovery.SolrServiceImpl @ Wrote Item: 10568/60785 to Index
2016-11-02 15:09:28,678 INFO  com.atmire.dspace.discovery.AtmireSolrService @ Processing (55695 of 55722): 43557
2016-11-02 15:09:28,688 INFO  com.atmire.dspace.discovery.AtmireSolrService @ Processing (55703 of 55722): 34476
  • I will raise a ticket with Atmire to ask them

2016-11-06

  • After re-deploying and re-indexing I didn’t see the same issue, and the indexing completed in 85 minutes, which is about how long it is supposed to take

2016-11-07

  • Horrible one liner to get Linode ID from certain Ansible host vars:
$ grep -A 3 contact_info * | grep -E "(Orth|Sisay|Peter|Daniel|Tsega)" | awk -F'-' '{print $1}' | grep linode | uniq | xargs grep linode_id
  • I noticed some weird CRPs in the database, and they don’t show up in Discovery for some reason, perhaps the :
  • I’ll export these and fix them in batch:
dspace=# \copy (select distinct text_value, count(*) from metadatavalue where metadata_field_id=230 group by text_value order by count desc) to /tmp/crp.csv with csv;
COPY 22
  • Test running the replacements:
$ ./fix-metadata-values.py -i /tmp/CRPs.csv -f cg.contributor.crp -t correct -m 230 -d dspace -u dspace -p 'fuuu'
  • Add AMR to ILRI subjects and remove one duplicate instance of IITA in author affiliations controlled vocabulary (#288)

2016-11-08

  • Atmire’s Listings and Reports module seems to be broken on DSpace 5.5

Listings and Reports broken in DSpace 5.5

  • I’ve filed a ticket with Atmire
  • Thinking about batch updates for ORCIDs and authors
  • Playing with SolrClient in Python to query Solr
  • All records in the authority core are either authority_type:orcid or authority_type:person
  • There is a deleted field and all items seem to be false, but might be important sanity check to remember
  • The way to go is probably to have a CSV of author names and authority IDs, then to batch update them in PostgreSQL
  • Dump of the top ~200 authors in CGSpace:
dspace=# \copy (select distinct text_value, count(*) from metadatavalue where metadata_field_id=3 group by text_value order by count desc limit 210) to /tmp/210-authors.csv with csv;

2016-11-09

  • CGSpace crashed so I quickly ran system updates, applied one or two of the waiting changes from the 5_x-prod branch, and rebooted the server
  • The error was Timeout waiting for idle object but I haven’t looked into the Tomcat logs to see what happened
  • Also, I ran the corrections for CRPs from earlier this week

2016-11-10

  • Helping Megan Zandstra and CIAT with some questions about the REST API
  • Playing with find-by-metadata-field, this works:
$ curl -s -H "accept: application/json" -H "Content-Type: application/json" -X POST "http://localhost:8080/rest/items/find-by-metadata-field" -d '{"key": "cg.subject.ilri","value": "SEEDS"}'
  • But the results are deceiving because metadata fields can have text languages and your query must match exactly!
dspace=# select distinct text_value, text_lang from metadatavalue where resource_type_id=2 and metadata_field_id=203 and text_value='SEEDS';
 text_value | text_lang
------------+-----------
 SEEDS      |
 SEEDS      |
 SEEDS      | en_US
(3 rows)
  • So basically, the text language here could be null, blank, or en_US
  • To query metadata with these properties, you can do:
$ curl -s -H "accept: application/json" -H "Content-Type: application/json" -X POST "http://localhost:8080/rest/items/find-by-metadata-field" -d '{"key": "cg.subject.ilri","value": "SEEDS"}' | jq length
55
$ curl -s -H "accept: application/json" -H "Content-Type: application/json" -X POST "http://localhost:8080/rest/items/find-by-metadata-field" -d '{"key": "cg.subject.ilri","value": "SEEDS", "language":""}' | jq length
34
$ curl -s -H "accept: application/json" -H "Content-Type: application/json" -X POST "http://localhost:8080/rest/items/find-by-metadata-field" -d '{"key": "cg.subject.ilri","value": "SEEDS", "language":"en_US"}' | jq length
  • The results (55+34=89) don’t seem to match those from the database:
dspace=# select count(text_value) from metadatavalue where resource_type_id=2 and metadata_field_id=203 and text_value='SEEDS' and text_lang is null;
 count
-------
    15
dspace=# select count(text_value) from metadatavalue where resource_type_id=2 and metadata_field_id=203 and text_value='SEEDS' and text_lang='';
 count
-------
     4
dspace=# select count(text_value) from metadatavalue where resource_type_id=2 and metadata_field_id=203 and text_value='SEEDS' and text_lang='en_US';
 count
-------
    66
  • So, querying from the API I get 55 + 34 = 89 results, but the database actually only has 85…
  • And the find-by-metadata-field endpoint doesn’t seem to have a way to get all items with the field, or a wildcard value
  • I’ll ask a question on the dspace-tech mailing list
  • And speaking of text_lang, this is interesting:
dspacetest=# select distinct text_lang from metadatavalue where resource_type_id=2;
 text_lang
-----------

 ethnob
 en
 spa
 EN
 es
 frn
 en_
 en_US

 EN_US
 eng
 en_U
 fr
(14 rows)
  • Generate a list of all these so I can maybe fix them in batch:
dspace=# \copy (select distinct text_lang, count(*) from metadatavalue where resource_type_id=2 group by text_lang order by count desc) to /tmp/text-langs.csv with csv;
COPY 14
  • Perhaps we need to fix them all in batch, or experiment with fixing only certain metadatavalues:
dspace=# update metadatavalue set text_lang='en_US' where resource_type_id=2 and metadata_field_id=203 and text_value='SEEDS';
UPDATE 85
  • The fix-metadata.py script I have is meant for specific metadata values, so if I want to update some text_lang values I should just do it directly in the database
  • For example, on a limited set:
dspace=# update metadatavalue set text_lang=NULL where resource_type_id=2 and metadata_field_id=203 and text_value='LIVESTOCK' and text_lang='';
UPDATE 420
  • And assuming I want to do it for all fields:
dspacetest=# update metadatavalue set text_lang=NULL where resource_type_id=2 and text_lang='';
UPDATE 183726
  • After that restarted Tomcat and PostgreSQL (because I’m superstitious about caches) and now I see the following in REST API query:
$ curl -s -H "accept: application/json" -H "Content-Type: application/json" -X POST "http://localhost:8080/rest/items/find-by-metadata-field" -d '{"key": "cg.subject.ilri","value": "SEEDS"}' | jq length
71
$ curl -s -H "accept: application/json" -H "Content-Type: application/json" -X POST "http://localhost:8080/rest/items/find-by-metadata-field" -d '{"key": "cg.subject.ilri","value": "SEEDS", "language":""}' | jq length
0
$ curl -s -H "accept: application/json" -H "Content-Type: application/json" -X POST "http://localhost:8080/rest/items/find-by-metadata-field" -d '{"key": "cg.subject.ilri","value": "SEEDS", "language":"en_US"}' | jq length
  • Not sure what’s going on, but Discovery shows 83 values, and database shows 85, so I’m going to reindex Discovery just in case

2016-11-14

  • I applied Atmire’s suggestions to fix Listings and Reports for DSpace 5.5 and now it works
  • There were some issues with the dspace/modules/jspui/pom.xml, which is annoying because all I did was rebase our working 5.1 code on top of 5.5, meaning Atmire’s installation procedure must have changed
  • So there is apparently this Tomcat native way to limit web crawlers to one session: Crawler Session Manager
  • After adding that to server.xml bots matching the pattern in the configuration will all use ONE session, just like normal users:
$ http --print h https://dspacetest.cgiar.org 'User-Agent:Mozilla/5.0 (compatible; Googlebot/2.1; +http://www.google.com/bot.html)'
HTTP/1.1 200 OK
Connection: keep-alive
Content-Encoding: gzip
Content-Language: en-US
Content-Type: text/html;charset=utf-8
Date: Mon, 14 Nov 2016 19:47:29 GMT
Server: nginx
Set-Cookie: JSESSIONID=323694E079A53D5D024F839290EDD7E8; Path=/; Secure; HttpOnly
Transfer-Encoding: chunked
Vary: Accept-Encoding
X-Cocoon-Version: 2.2.0
X-Robots-Tag: none

$ http --print h https://dspacetest.cgiar.org 'User-Agent:Mozilla/5.0 (compatible; Googlebot/2.1; +http://www.google.com/bot.html)'
HTTP/1.1 200 OK
Connection: keep-alive
Content-Encoding: gzip
Content-Language: en-US
Content-Type: text/html;charset=utf-8
Date: Mon, 14 Nov 2016 19:47:35 GMT
Server: nginx
Transfer-Encoding: chunked
Vary: Accept-Encoding
X-Cocoon-Version: 2.2.0
  • The first one gets a session, and any after that — within 60 seconds — will be internally mapped to the same session by Tomcat
  • This means that when Google or Baidu slam you with tens of concurrent connections they will all map to ONE internal session, which saves RAM!

2016-11-15

  • The Tomcat JVM heap looks really good after applying the Crawler Session Manager fix on DSpace Test last night:

Tomcat JVM heap (day) after setting up the Crawler Session Manager Tomcat JVM heap (week) after setting up the Crawler Session Manager

  • Seems the default regex doesn’t catch Baidu, though:
$ http --print h https://dspacetest.cgiar.org 'User-Agent:Mozilla/5.0 (compatible; Baiduspider/2.0; +http://www.baidu.com/search/spider.html)'
HTTP/1.1 200 OK
Connection: keep-alive
Content-Encoding: gzip
Content-Language: en-US
Content-Type: text/html;charset=utf-8
Date: Tue, 15 Nov 2016 08:49:54 GMT
Server: nginx
Set-Cookie: JSESSIONID=131409D143E8C01DE145C50FC748256E; Path=/; Secure; HttpOnly
Transfer-Encoding: chunked
Vary: Accept-Encoding
X-Cocoon-Version: 2.2.0

$ http --print h https://dspacetest.cgiar.org 'User-Agent:Mozilla/5.0 (compatible; Baiduspider/2.0; +http://www.baidu.com/search/spider.html)'
HTTP/1.1 200 OK
Connection: keep-alive
Content-Encoding: gzip
Content-Language: en-US
Content-Type: text/html;charset=utf-8
Date: Tue, 15 Nov 2016 08:49:59 GMT
Server: nginx
Set-Cookie: JSESSIONID=F6403C084480F765ED787E41D2521903; Path=/; Secure; HttpOnly
Transfer-Encoding: chunked
Vary: Accept-Encoding
X-Cocoon-Version: 2.2.0
  • Adding Baiduspider to the list of user agents seems to work, and the final configuration should be:
<!-- Crawler Session Manager Valve helps mitigate damage done by web crawlers -->
<Valve className="org.apache.catalina.valves.CrawlerSessionManagerValve"
       crawlerUserAgents=".*[bB]ot.*|.*Yahoo! Slurp.*|.*Feedfetcher-Google.*|.*Baiduspider.*" />
  • Looking at the bots that were active yesterday it seems the above regex should be sufficient:
$ grep -o -E 'Mozilla/5\.0 \(compatible;.*\"' /var/log/nginx/access.log | sort | uniq
Mozilla/5.0 (compatible; Baiduspider/2.0; +http://www.baidu.com/search/spider.html)" "-"
Mozilla/5.0 (compatible; bingbot/2.0; +http://www.bing.com/bingbot.htm)" "-"
Mozilla/5.0 (compatible; Googlebot/2.1; +http://www.google.com/bot.html)" "-"
Mozilla/5.0 (compatible; YandexBot/3.0; +http://yandex.com/bots)" "-"
Mozilla/5.0 (compatible; YandexImages/3.0; +http://yandex.com/bots)" "-"
  • Neat maven trick to exclude some modules from being built:
$ mvn -U -Dmirage2.on=true -Dmirage2.deps.included=false -Denv=localhost -P \!dspace-lni,\!dspace-rdf,\!dspace-sword,\!dspace-swordv2 clean package
  • We absolutely don’t use those modules, so we shouldn’t build them in the first place