cgspace-notes/content/posts/2021-03.md

title	date	author	categories
March, 2021	2021-03-01T10:13:54+02:00	Alan Orth	Notes

2021-03-01

• Discuss some OpenRXV issues with Abdullah from CodeObia
  • He's trying to work on the DSpace 6+ metadata schema autoimport using the DSpace 6+ REST API
  • Also, we found some issues building and running OpenRXV currently due to ecosystem shift in the Node.js dependencies

2021-03-02

• I fixed three build and runtime issues in OpenRXV:
  • fix highcharts-angular and ngx-tour-core build
  • frontend/package.json: Pin @types/ramda at 0.27.34
• Then I merged a few fixes that Abdullah had worked on last week

2021-03-03

• I fixed another frontend build warning on OpenRXV
• Then I updated the frontend container to use Node.js 12 and Ubuntu 20.04
• Also, I added a GitHub Actions workflow to build the frontend
• I did some testing of Abdullah's patch for the values mapping search on OpenRXV
  • It still doesn't work with multi-word values, so I recorded a video with wf-recorder and uploaded it to the issue for him to investigate

2021-03-04

• Peter is having issues with the workflow since yesterday
  • I looked at the Munin stats and see a high number of database locks since yesterday

• I looked at the number of connections in PostgreSQL and it's definitely high again:

$ psql -c 'SELECT * FROM pg_locks pl LEFT JOIN pg_stat_activity psa ON pl.pid = psa.pid;' | wc -l
1020

• I reported it to Atmire to take a look, on the same issue we had been tracking this before
• Abenet asked me to add a new ORCID for ILRI staff member Zoe Campbell
• I added it to the controlled vocabulary and then tagged her existing items on CGSpace using my add-orcid-identifier.py script:

$ cat 2021-03-04-add-zoe-campbell-orcid.csv 
dc.contributor.author,cg.creator.identifier
"Campbell, Zoë","Zoe Campbell: 0000-0002-4759-9976"
"Campbell, Zoe A.","Zoe Campbell: 0000-0002-4759-9976"
$ ./ilri/add-orcid-identifiers-csv.py -i 2021-03-04-add-zoe-campbell-orcid.csv -db dspace -u dspace -p 'fuuu'

• I still need to do cleanup on the journal articles metadata
  • Peter sent me some cleanups but I can't use them in the search/replace format he gave
  • I think it's better to export the metadata values with IDs and import cleaned up ones as CSV

localhost/dspace63= > \COPY (SELECT dspace_object_id AS id, text_value as "cg.journal" FROM metadatavalue WHERE dspace_object_id IN (SELECT uuid FROM item) AND metadata_field_id=251) to /tmp/2021-02-24-journals.csv WITH CSV HEADER;
COPY 32087

• I used OpenRefine to remove all journal values that didn't have one of these values: ; ( )
  • Then I cloned the cg.journal field to cg.volume and cg.issue
  • I used some GREL expressions like these to extract the journal name, volume, and issue:

value.partition(';')[0].trim() # to get journal names
value.partition(/[0-9]+\([0-9]+\)/)[1].replace(/^(\d+)\(\d+\)/,"$1") # to get journal volumes
value.partition(/[0-9]+\([0-9]+\)/)[1].replace(/^\d+\((\d+)\)/,"$1") # to get journal issues

• Then I uploaded the changes to CGSpace using dspace metadata-import
• Margarita from CCAFS was asking about an error deleting some items that were showing up in Google and should have been private
  • The error was "Authorization denied for action OBSOLETE (DELETE) on BITSTREAM:bd157345-448e ..."
  • I searched the DSpace issue tracker and found several issues reporting this:
    • DS-3985 Delete item fails
    • DS-4004 Authorization denied Exception when trying to delete permanently an item, collection or community as a non-Admin user
    • DS-4297 Authorization error when trying to delete item by submitter/administrator
  • The issue is apparently with non-admin users who are in the admin and submit groups of the owning collection...
  • In this case the item was uploaded to the CCAFS Reports collection, and Margarita is a non-admin user who is a member of the collection's admin and submit groups, exactly as the issue described
  • I added a comment about our issue to DS-4297
• Yesterday Abenet added me to a WLE collection approver/editer steps so we can try to figure out why Niroshini is having issues adding metadata to Udana's submissions
  • I edited Udana's submission to CGSpace:
    • corrected the title
    • added language English
    • changed the link to the external item page instead of PDF
    • added SDGs from the external item page
    • added AGROVOC subjects from the external item page
    • added pagination (extent)
    • changed the license to "other" because CC-BY-NC-ND is not printed anywhere in the PDF or external item page

2021-03-05

• I migrated the Docker bind mount for the AReS Elasticsearch container to a Docker volume:

$ docker-compose -f docker/docker-compose.yml down
$ docker volume create docker_esData_7
$ docker container create --name es_dummy -v docker_esData_7:/usr/share/elasticsearch/data:rw elasticsearch:7.6.2
$ docker cp docker/esData_7/nodes es_dummy:/usr/share/elasticsearch/data
$ docker rm es_dummy
# edit docker/docker-compose.yml to switch from bind mount to volume
$ docker-compose -f docker/docker-compose.yml up -d

• The trick is that when you create a volume like "myvolume" from a docker-compose.yml file, Docker will create it with the name "docker_myvolume"
  • If you create it manually on the command line with docker volume create myvolume then the name is literally "myvolume"
• I still need to make the changes to git master and add these notes to the pull request so Moayad and others can benefit
• Delete the openrxv-items-temp index to test a fresh harvesting:

$ curl -XDELETE 'http://localhost:9200/openrxv-items-temp'

2021-03-05

• Check the results of the AReS harvesting from last night:

$ curl -s 'http://localhost:9200/openrxv-items-temp/_count?q=*&pretty'
{
  "count" : 101761,
  "_shards" : {
    "total" : 1,
    "successful" : 1,
    "skipped" : 0,
    "failed" : 0
  }
}

• Set the current items index to read only and make a backup:

$ curl -X PUT "localhost:9200/openrxv-items/_settings" -H 'Content-Type: application/json' -d' {"settings": {"index.blocks.write":true}}'
$ curl -s -X POST http://localhost:9200/openrxv-items/_clone/openrxv-items-2021-03-05

• Delete the current items index and clone the temp one to it:

$ curl -XDELETE 'http://localhost:9200/openrxv-items'
$ curl -X PUT "localhost:9200/openrxv-items-temp/_settings" -H 'Content-Type: application/json' -d'{"settings": {"index.blocks.write": true}}'
$ curl -s -X POST http://localhost:9200/openrxv-items-temp/_clone/openrxv-items

• Then delete the temp and backup:

$ curl -XDELETE 'http://localhost:9200/openrxv-items-temp'
{"acknowledged":true}%
$ curl -XDELETE 'http://localhost:9200/openrxv-items-2021-03-05'

• I made some pull requests to OpenRXV:
  • docker/docker-compose.yml: Use docker volumes
  • docker/docker-compose.yml: Pin Redis to version 5
• I deployed the latest changes from the last few days on AReS production

2021-03-07

• I realized there is something wrong with the Elasticsearch indexes on AReS
  • On a new test environment I see openrxv-items is correctly an alias of openrxv-items-final:

$ curl -s 'http://localhost:9200/_alias/' | python -m json.tool | less
...
    "openrxv-items-final": {
        "aliases": {
            "openrxv-items": {}
        }
    },

• But on AReS production openrxv-items has somehow become an index:

$ curl -s 'http://localhost:9200/_alias/' | python -m json.tool | less
...
    "openrxv-items": {
        "aliases": {}
    },
    "openrxv-items-final": {
        "aliases": {}
    },
    "openrxv-items-temp": {
        "aliases": {}
    },

• I fixed the issue on production by cloning the openrxv-items index to openrxv-items-final, deleting openrxv-items, and then re-creating it as an alias:

$ curl -X PUT "localhost:9200/openrxv-items/_settings" -H 'Content-Type: application/json' -d'{"settings": {"index.blocks.write": true}}'
$ curl -s -X POST http://localhost:9200/openrxv-items/_clone/openrxv-items-2021-03-07
$ curl -XDELETE 'http://localhost:9200/openrxv-items-final'
$ curl -s -X POST http://localhost:9200/openrxv-items/_clone/openrxv-items-final
$ curl -XDELETE 'http://localhost:9200/openrxv-items'
$ curl -s -X POST 'http://localhost:9200/_aliases' -H 'Content-Type: application/json' -d'{"actions" : [{"add" : { "index" : "openrxv-items-final", "alias" : "openrxv-items"}}]}'

• Delete backups and remove read-only mode on openrxv-items:

$ curl -XDELETE 'http://localhost:9200/openrxv-items-2021-03-07'
$ curl -X PUT "localhost:9200/openrxv-items/_settings" -H 'Content-Type: application/json' -d'{"settings": {"index.blocks.write": false}}'

• Linode sent alerts about the CPU usage on CGSpace yesterday and the day before
  • Looking in the logs I see a few IPs making heavy usage on the REST API and XMLUI:

# zcat --force /var/log/nginx/access.log /var/log/nginx/access.log.1 /var/log/nginx/access.log.2.gz /var/log/nginx/access.log.3.gz | grep -E '0[56]/Mar/2021' | goaccess --log-format=COMBINED -

• I see the usual IPs for CCAFS and ILRI importer bots, but also 143.233.242.132 which appears to be for GARDIAN:

# zgrep '143.233.242.132' /var/log/nginx/access.log.1 | grep -c Delphi
6237
# zgrep '143.233.242.132' /var/log/nginx/access.log.1 | grep -c -v Delphi
6418

• They seem to make requests twice, once with the Delphi user agent that we know and already mark as a bot, and once with a "normal" user agent
  • Looking in Solr I see they have been using this IP for awhile, as they have 100,000 hits going back into 2020
  • I will add this IP to the list of bots in nginx and purge it from Solr with my check-spider-ip-hits.sh script
• I made a few changes to OpenRXV:
  • Migrated away from links to use networks
  • Converted the backend container to use a custom image that includes unoconv so we don't have to manually install it anymore

2021-03-08

• I approved the WLE item that I edited last week, and all the metadata is there: https://hdl.handle.net/10568/111810
  • So I'm not sure what Niroshini's issue with metadata is...
• Peter sent a message yesterday saying that his item finally got committed
  • I looked at the Munin graphs and there was a MASSIVE spike in database activity two days ago, and now database locks are back down to normal levels (from 1000+):

$ psql -c 'SELECT * FROM pg_locks pl LEFT JOIN pg_stat_activity psa ON pl.pid = psa.pid;' | wc -l
13

• On 2021-03-03 the PostgreSQL transactions started rising:

• After that the connections and locks started going up, peaking on 2021-03-06:

• I sent another message to Atmire to ask if they have time to look into this
• CIFOR is pressuring me to upload the batch items from last week
  • Vika sent me a final file with some duplicates that Peter identified removed
  • I extracted and re-applied my basic corrections from last week in OpenRefine, then ran the items through csv-metadata-quality checker and uploaded them to CGSpace
  • In total there are 1,088 items
• Udana from IWMI emailed to ask about CGSpace thumbnails
• Udana from IWMI emailed to ask about an item uploaded recently that does not appear in AReS
  • The item was added to the archive on 2021-03-05, and I last harvested on 2021-03-06, so this might be an issue of a missing item
• Abenet got a quote from Atmire to buy 125 credits for 3750€
• Maria at Bioversity sent some feedback about duplicate items on AReS
• I'm wondering if the issue of the openrxv-items-final index not getting cleared after a successful harvest (which results in having 200,000, then 300,000, etc items) has to do with the alias issue I fixed yesterday
  • I will start a fresh harvest on AReS without now to check, but first back up the current index just in case:

$ curl -X PUT "localhost:9200/openrxv-items-final/_settings" -H 'Content-Type: application/json' -d'{"settings": {"index.blocks.write": true}}'
$ curl -s -X POST http://localhost:9200/openrxv-items-final/_clone/openrxv-items-final-2021-03-08
# start harvesting on AReS

• As I saw on my local test instance, even when you cancel a harvesting, it replaces the openrxv-items-final index with whatever is in openrxv-items-temp automatically, so I assume it will do the same now

2021-03-09

• The harvesting on AReS finished last night and everything worked as expected, with no manual intervention
  • This means that the issue we were facing for a few months was due to the openrxv-items index being deleted and re-created as a standalone index instead of an alias of openrxv-items-final
• Talk to Moayad about OpenRXV development
  • We realized that the missing/duplicate items issue is probably due to the long harvesting time on the REST API, as the time between starting the harvesting on page 0 and finishing the harvesting on page 900 (in the CGSpace example), some items will have been added to the repository, which causes the pages to shift
  • I proposed a solution in the GitHub issue, where we consult the site's XML sitemap after harvesting to see if we missed any items, and then we harvest them individually
• Peter sent me a list of 356 DOIs from Altmetric that don't have our Handles, so we need to Tweet them
  • I used my doi-to-handle.py script to generate a list of handles and titles for him:

$ ./ilri/doi-to-handle.py -i /tmp/dois.txt -o /tmp/handles.txt -db dspace -u dspace -p 'fuuu'