cgspace-notes/content/posts/2020-08.md

title	date	author	categories
August, 2020	2020-08-02T15:35:54+03:00	Alan Orth	Notes

2020-08-02

• I spent a few days working on a Java-based curation task to tag items with ISO 3166-1 Alpha2 country codes based on their cg.coverage.country text values
  • It looks up the names in ISO 3166-1 first, and then in our CGSpace countries mapping (which has five or so of Peter's preferred "display" country names)
  • It implements a "force" mode too that will clear existing country codes and re-tag everything
  • It is class based so I can easily add support for other vocabularies, and the technique could even be used for organizations with mappings to ROR and Clarisa...

• The code is currently on my personal GitHub: https://github.com/alanorth/dspace-curation-tasks
  • I still need to figure out how to integrate this with the DSpace build because currently you have to package it and copy the JAR to the dspace/lib directory (not to mention the config)
• I forked the dspace-curation-tasks to ILRI's GitHub and submitted the project to Maven Central so I can integrate it more easily with our DSpace build via dependencies

2020-08-03

• Atmire responded to the ticket about the ongoing upgrade issues
  • They pushed an RC2 version of the CUA module that fixes the FontAwesome issue so that they now use classes instead of Unicode hex characters so our JS + SVG works!
  • They also said they have never experienced the type: 5 site statistics issue, so I need to try to purge those and continue with the stats processing
• I purged all unmigrated stats in a few cores and then restarted processing:

$ curl -s "http://localhost:8081/solr/statistics/update?softCommit=true" -H "Content-Type: text/xml" --data-binary '<delete><query>id:/.*unmigrated.*/</query></delete>'
$ export JAVA_OPTS='-Dfile.encoding=UTF-8 -Xmx2048m'
$ chrt -b 0 dspace dsrun com.atmire.statistics.util.update.atomic.AtomicStatisticsUpdateCLI -t 12 -c statistics

• Andrea from Macaroni Bros emailed me a few days ago to say he's having issues with the CGSpace REST API
  • He said he noticed the issues when they were developing the WordPress plugin to harvest CGSpace for the RTB website: https://www.rtb.cgiar.org/publications/

2020-08-04

• Look into the REST API issues that Macaroni Bros raised last week:
  • The first one was about the collections endpoint returning empty items:
    • https://cgspace.cgiar.org/rest/collections/1445/items?limit=1&offset=2 (offset=2 is correct)
    • https://cgspace.cgiar.org/rest/collections/1445/items?limit=1&offset=3 (offset=3 is empty)
    • https://cgspace.cgiar.org/rest/collections/1445/items?limit=1&offset=4 (offset=4 is correct again)
  • I confirm that the second link returns zero items on CGSpace...
    • I tested on my local development instance and it returns one item correctly...
    • I tested on DSpace Test (currently DSpace 6 with UUIDs) and it returns one item correctly...
    • Perhaps an indexing issue?
  • The second issue is the collections endpoint returning the wrong number of items:
    • https://cgspace.cgiar.org/rest/collections/1445 (numberItems: 63)
    • https://cgspace.cgiar.org/rest/collections/1445/items (real number of items: 61)
  • I confirm that it is indeed happening on CGSpace...
    • And actually I can replicate the same issue on my local CGSpace 5.8 instance:

$ http 'http://localhost:8080/rest/collections/1445' | json_pp | grep numberItems
   "numberItems" : 63,
$ http 'http://localhost:8080/rest/collections/1445/items' jq '. | length'
61

• Also on DSpace Test (which is running DSpace 6!), though the issue is slightly different there:

$ http 'https://dspacetest.cgiar.org/rest/collections/5471c3aa-202e-42f0-96c2-497a18e3b708' | json_pp | grep numberItems
   "numberItems" : 61,
$ http 'https://dspacetest.cgiar.org/rest/collections/5471c3aa-202e-42f0-96c2-497a18e3b708/items' | jq '. | length'
59

• Ah! I exported that collection's metadata and checked it in OpenRefine, where I noticed that two items are mapped twice
  • I dealt with this problem in 2017-01 and the solution is to check the collection2item table:

dspace=# SELECT * FROM collection2item WHERE item_id = '107687';
   id   | collection_id | item_id
--------+---------------+---------
 133698 |           966 |  107687
 134685 |          1445 |  107687
 134686 |          1445 |  107687
(3 rows)

• So for each id you can delete one duplicate mapping:

dspace=# DELETE FROM collection2item WHERE id='134686';
dspace=# DELETE FROM collection2item WHERE id='128819';

• Update countries on CGSpace to be closer to ISO 3166-1 with some minor differences based on Peter's preferred display names

$ cat 2020-08-04-PB-new-countries.csv
cg.coverage.country,correct
CAPE VERDE,CABO VERDE
COCOS ISLANDS,COCOS (KEELING) ISLANDS
"CONGO, DR","CONGO, DEMOCRATIC REPUBLIC OF"
COTE D'IVOIRE,CÔTE D'IVOIRE
"KOREA, REPUBLIC","KOREA, REPUBLIC OF"
PALESTINE,"PALESTINE, STATE OF"
$ ./fix-metadata-values.py -i 2020-08-04-PB-new-countries.csv -db dspace -u dspace -p 'fuuu' -f cg.coverage.country -t 'correct' -m 228

• I had to restart Tomcat 7 three times before all the Solr statistics cores came up properly
  • I started a full Discovery re-indexing

2020-08-05

• Port my dspace-curation-tasks to DSpace 6 and tag version 6.0-SNAPSHOT
• I downloaded the UN M.49 CSV file to start working on updating the CGSpace regions
  • First issue is they don't version the file so you have no idea when it was released
  • Second issue is that three rows have errors due to not using quotes around "China, Macao Special Administrative Region"
• Bizu said she was having problems approving tasks on CGSpace
  • I looked at the PostgreSQL locks and they have skyrocketed since yesterday:

• Seems that something happened yesterday afternoon at around 5PM...
  • For now I will just run all updates on the server and reboot it, as I have no idea what causes this issue
  • I had to restart Tomcat 7 three times after the server came back up before all Solr statistics cores came up properly
• I checked the nginx logs around 5PM yesterday to see who was accessing the server:

# cat /var/log/nginx/*.log /var/log/nginx/*.log.1 | grep -E '04/Aug/2020:(17|18)' | goaccess --log-format=COMBINED -

• I see the Macaroni Bros are using their new user agent for harvesting: RTB website BOT
  • But that pattern doesn't match in the nginx bot list or Tomcat's crawler session manager valve because we're only checking for [Bb]ot!
  • So they have created thousands of Tomcat sessions:

$ cat dspace.log.2020-08-04 | grep -E "(63.32.242.35|64.62.202.71)" | grep -E 'session_id=[A-Z0-9]{32}' | sort | uniq | wc -l
5693

• DSpace itself uses a case-sensitive regex for user agents so there are no hits from those IPs in Solr, but I need to tweak the other regexes so they don't misuse the resources
  • Perhaps [Bb][Oo][Tt]...
• I see another IP 104.198.96.245, which is also using the "RTB website BOT" but there are 70,000 hits in Solr from earlier this year before they started using the user agent
  • I purged all the hits from Solr, including a few thousand from 64.62.202.71
• A few more IPs causing lots of Tomcat sessions yesterday:

$ cat dspace.log.2020-08-04 | grep "38.128.66.10" | grep -E 'session_id=[A-Z0-9]{32}' | sort | uniq | wc -l
1585
$ cat dspace.log.2020-08-04 | grep "64.62.202.71" | grep -E 'session_id=[A-Z0-9]{32}' | sort | uniq | wc -l
5691

• 38.128.66.10 isn't creating any Solr statistics due to our DSpace agents pattern, but they are creating lots of sessions so perhaps I need to force them to use one session in Tomcat:

Mozilla/5.0 (Windows NT 5.1) brokenlinkcheck.com/1.2

• 64.62.202.71 is using a user agent I've never seen before:

Mozilla/5.0 (compatible; +centuryb.o.t9[at]gmail.com)

• So now our "bot" regex can't even match that...
  • Unless we change it to [Bb]\.?[Oo]\.?[Tt]\.?... which seems to match all variations of "bot" I can think of right now, according to regexr.com:

RTB website BOT
Altmetribot
Mozilla/5.0 (compatible; Googlebot/2.1; +http://www.google.com/bot.html)
Mozilla/5.0 (compatible; +centuryb.o.t9[at]gmail.com)
Mozilla/5.0 (compatible; BLEXBot/1.0; +http://webmeup-crawler.com/)

• And another IP belonging to Turnitin (the alternate user agent of Turnitinbot):

$ cat dspace.log.2020-08-04 | grep "199.47.87.145" | grep -E 'sessi
on_id=[A-Z0-9]{32}' | sort | uniq | wc -l
2777

• I will add Turnitin to the Tomcat Crawler Session Manager Valve regex as well...

2020-08-06

• I have been working on processing the Solr statistics with the Atmire tool on DSpace Test the last few days:
  • statistics:
    • 2,040,385 docs: 2h 28m 49s
  • statistics-2019:
    • 8,960,000 docs: 12h 7s
    • 1,780,575 docs: 2h 7m 29s
  • statistics-2018:
    • 1,970,000 docs: 12h 1m 28s
    • 360,000 docs: 2h 54m 56s (Linode rebooted)
    • 1,110,000 docs: 7h 1m 44s (Restarted Tomcat, oops)
• I decided to start the 2018 core over again, so I re-synced it from CGSpace and started again from the solr-upgrade-statistics-6x tool and now I'm having the same issues with Java heap space that I had last month
  • The process kept crashing due to memory, so I increased the memory to 3072m and finally 4096m...
  • Also, I decided to try to purge all the -unmigrated docs that it had found so far to see if that helps...
  • There were about 466,000 records unmigrated so far, most of which were type: 5 (SITE statistics)
  • Now it is processing again...
• I developed a small Java class called FixJpgJpgThumbnails to remove ".jpg.jpg" thumbnails from the THUMBNAIL bundle and replace them with their originals from the ORIGINAL bundle
  • The code is based on RemovePNGThumbnailsForPDFs.java by Andrea Schweer
  • I incorporated it into my dspace-curation-tasks repository, then renamed it to cgspace-java-helpers
  • In testing I found that I can replace ~4,000 thumbnails on CGSpace!

2020-08-07

• I improved the RemovePNGThumbnailsForPDFs.java a bit more to exclude infographics and original bitstreams larger than 100KiB
  • I ran it on CGSpace and it cleaned up 3,769 thumbnails!
  • Afterwards I ran dspace cleanup -v to remove the deleted thumbnails