CGSpace Notes

Documenting day-to-day work on the CGSpace repository.

May, 2020

by Alan Orth in  Notes

2020-05-02

  • Peter said that CTA is having problems submitting an item to CGSpace
    • Looking at the PostgreSQL stats it seems to be the same issue that Tezira was having last week, as I see the number of connections in ‘idle in transaction’ and ‘waiting for lock’ state are increasing again
    • I see that CGSpace (linode18) is still using PostgreSQL JDBC driver version 42.2.11, and there were some bugs related to transactions fixed in 42.2.12 (which I had updated in the Ansible playbooks, but not deployed yet)

2020-05-03

  • Purge a few remaining bots from CGSpace Solr statistics that I had identified a few months ago
    • lua-resty-http/0.10 (Lua) ngx_lua/10000
    • omgili/0.5 +http://omgili.com
    • IZaBEE/IZaBEE-1.01 (Buzzing Abound The Web; https://izabee.com; info at izabee dot com)
    • Twurly v1.1 (https://twurly.org)
    • Pattern/2.6 +http://www.clips.ua.ac.be/pattern
    • CyotekWebCopy/1.7 CyotekHTTP/2.0
  • This is only about 2,500 hits total from the last ten years, and half of these bots no longer seem to exist, so I won’t bother submitting them to the COUNTER-Robots project
  • I noticed that our custom themes were incorrectly linking to the OpenSearch XML file
    • The bug was fixed for Mirage2 in 2015
    • Note that this did not prevent OpenSearch itself from working
    • I will patch this on our DSpace 5.x and 6.x branches

2020-05-06

  • Atmire responded asking for more information about the Solr statistics processing bug in CUA so I sent them some full logs
    • Also I asked again about the Maven variable interpolation issue for cua.version.number, and if they would be willing to upgrade CUA to use Font Awesome 5 instead of 4.

2020-05-07

  • Linode sent an alert that there was high CPU usage on CGSpace (linode18) early this morning
    • I looked at the nginx logs using goaccess and I found a few IPs making lots of requests around then:
# cat /var/log/nginx/*.log /var/log/nginx/*.log.1 | grep -E "07/May/2020:(01|03|04)" | goaccess --log-format=COMBINED -
  • The two main IPs making requests around then are 188.134.31.88 and 212.34.8.188
    • The first is in Russia and it is hitting mostly XMLUI Discover links using dozens of different user agents, a total of 20,000 requests this week
    • The second IP is CodeObia testing AReS, a total of 171,000 hits this month
    • I will purge both of those IPs from the Solr stats using my check-spider-ip-hits.sh script:
$ ./check-spider-ip-hits.sh -f /tmp/ips -s statistics -p
Purging 171641 hits from 212.34.8.188 in statistics
Purging 20691 hits from 188.134.31.88 in statistics

Total number of bot hits purged: 192332
  • And then I will add 188.134.31.88 to the nginx bad bot list and tell CodeObia to please use a “bot” user agent
  • I also changed the nginx config to block requests with blank user agents

2020-05-11

  • Bizu said she was having issues submitting to CGSpace last week
    • The issue sounds like the one Tezira and CTA were having in the last few weeks
    • I looked at the PostgreSQL graphs and see there are a lot of connections in “idle in transaction” and “waiting for lock” state:

PostgreSQL connections

  • I think I’ll downgrade the PostgreSQL JDBC driver from 42.2.12 to 42.2.10, which was the version we were using before these issues started happening
  • Atmire sent some feedback about my ongoing issues with their CUA module, but none of it was conclusive yet
    • Regarding Font Awesome 5 they will check how much work it will take and give me a quote
  • Abenet said some users are questioning why the statistics dropped so much lately, so I made a post to Yammer to explain about the robots
  • Last week Peter had asked me to add a new ILRI author’s ORCID iD
    • I added it to the controlled vocabulary and tagged the user’s existing ~11 items in CGSpace using this CSV file with my add-orcid-identifiers-csv.py script:
$ cat 2020-05-11-add-orcids.csv
dc.contributor.author,cg.creator.id
"Lutakome, P.","Pius Lutakome: 0000-0002-0804-2649"
"Lutakome, Pius","Pius Lutakome: 0000-0002-0804-2649"
$ ./add-orcid-identifiers-csv.py -i 2020-05-11-add-orcids.csv -db dspace -u dspace -p 'fuuu' -d
  • Run system updates on CGSpace (linode18) and reboot it
    • I had to restart Tomcat five times before all Solr statistics cores came up OK, ugh.

2020-05-12

  • Peter noticed that CGSpace is no longer on AReS, because I blocked all requests that don’t specify a user agent
    • I’ve temporarily disabled that restriction and asked Moayad to look into how he can specify a user agent in the AReS harvester

2020-05-13

  • Atmire responded about Font Awesome and said they can switch to version 5 for 16 credits
    • I told them to go ahead
  • Also, Atmire gave me a small workaround for the cua.version.number interpolation issue and said they would look into the crash that happens when processing our Solr stats
  • Run system updates and reboot AReS server (linode20) for the first time in almost 100 days
    • I notice that AReS now has some of CGSpace’s data in it (but not all) since I dropped the user-agent restriction on the REST API yesterday

2020-05-17

  • Create an issue in the OpenRXV project for Moayad to change the default harvester user agent (#36)

2020-05-18

  • Atmire responded and said they still can’t figure out the CUA statistics issue, though they seem to only be trying to understand what’s going on using static analysis
    • I told them that they should try to run the code with the Solr statistics that I shared with them a few weeks ago

2020-05-19

  • Add ORCID identifier for Sirak Bahta
    • I added it to the controlled vocabulary and tagged the user’s existing ~40 items in CGSpace using this CSV file with my add-orcid-identifiers-csv.py script:
$ cat 2020-05-19-add-orcids.csv
dc.contributor.author,cg.creator.id
"Bahta, Sirak T.","Sirak Bahta: 0000-0002-5728-2489"
$ ./add-orcid-identifiers-csv.py -i 2020-05-19-add-orcids.csv -db dspace -u dspace -p 'fuuu' -d
  • An IITA user is having issues submitting to CGSpace and I see there are a rising number of PostgreSQL connections waiting in transaction and in lock:

PostgreSQL connections

  • This is the same issue Tezira, Bizu, and CTA were having in the last few weeks and it I already downgraded the PostgreSQL JDBC driver version to the last version I was using before this started (42.2.10)
    • I will downgrade it to version 42.2.9 for now…
    • The only other thing I can think of is that I upgraded Tomcat to 7.0.103 in March
  • Run system updates on DSpace Test (linode26) and reboot it
  • Run system updates on CGSpace (linode18) and reboot it
    • After the system came back up I had to restart Tomcat 7 three times before all the Solr statistics cores came up OK
  • Send Atmire a snapshot of the CGSpace database for them to possibly troubleshoot the CUA issue with DSpace 6

2020-05-20

  • Send CodeObia some logos and footer text for the next phase of OpenRXV development (#18)

2020-05-25

  • Add ORCID identifier for CIAT author Manuel Francisco
    • I added it to the controlled vocabulary and tagged the user’s existing ~27 items in CGSpace using this CSV file with my add-orcid-identifiers-csv.py script:
$ cat 2020-05-25-add-orcids.csv
dc.contributor.author,cg.creator.id
"Díaz, Manuel F.","Manuel Francisco Diaz Baca: 0000-0001-8996-5092"
"Díaz, Manuel Francisco","Manuel Francisco Diaz Baca: 0000-0001-8996-5092"
$ ./add-orcid-identifiers-csv.py -i 2020-05-25-add-orcids.csv -db dspace -u dspace -p 'fuuu' -d
  • Last week Maria asked again about searching for items by accession or issue date
    • A few months ago I had told her to search for the ISO8601 date in Discovery search, which appears to work because it filters the results down quite a bit
    • She pointed out that the results include hits that don’t exactly match, for example if part of the search string appears elsewhere like in the timestamp
    • I checked in Solr and the results are the same, so perhaps it’s a limitation in Solr…?
    • So this effectively means that we don’t have a way to create reports for items in an arbitrary date range shorter than a year:
      • DSpace advanced search is buggy or simply not designed to work like that
      • AReS Explorer currently only allows filtering by year, but will allow months soon
      • Atmire Listings and Reports only allows a “Timespan” of a year

2020-05-29

  • Linode alerted to say that the CPU load on CGSpace (linode18) was high for a few hours this morning
    • Looking at the nginx logs for this morning with goaccess:
# cat /var/log/nginx/*.log.1 | grep -E "29/May/2020:(02|03|04|05)" | goaccess --log-format=COMBINED -
  • The top is 172.104.229.92, which is the AReS harvester (still not using a user agent, but it’s tagged as a bot in the nginx mapping)
  • Second is 188.134.31.88, which is a Russian host that we also saw in the last few weeks, using a browser user agent and hitting the XMLUI (but it is tagged as a bot in nginx as well)
  • Another one is 51.158.106.4, which is some Scaleway IP making requests to XMLUI with different browser user agents that I am pretty sure I have seen before but never blocked
    • According to Solr it has made about 800 requests this year, but still… it’s a bot.
  • One I don’t think I’ve seen before is 95.217.58.146, which is making requests to XMLUI with a Drupal user agent
  • Atmire got back to me about the Solr CUA issue in the DSpace 6 upgrade and they cannot reproduce the error
    • The next step is for me to migrate DSpace Test (linode26) to DSpace 6 and try to reproduce the error there

2020-05-31

  • Start preparing to migrate DSpace Test (linode26) to the 6_x-dev-atmire-modules branch
    • Run all system updates and reboot
    • For now I will disable all yearly Solr statistics cores except the current statistics one
    • Prepare PostgreSQL with a clean snapshot of CGSpace’s DSpace 5.8 database:
$ sudo su - postgres
$ dropdb dspacetest
$ createdb -O dspacetest --encoding=UNICODE dspacetest
$ psql dspacetest -c 'alter user dspacetest superuser;'
$ pg_restore -d dspacetest -O --role=dspacetest /tmp/cgspace_2020-05-31.backup
$ psql dspacetest -c 'alter user dspacetest nosuperuser;'
# run DSpace 5 version of update-sequences.sql!!!
$ psql -f /home/dspace/src/git/DSpace/dspace/etc/postgres/update-sequences.sql dspacetest
$ psql dspacetest -c "DELETE FROM schema_version WHERE version IN ('5.8.2015.12.03.3');"
$ psql dspacetest -c 'CREATE EXTENSION pgcrypto;'
$ exit
  • Now switch to the DSpace 6.x branch and start a build:
$ chrt -i 0 ionice -c2 -n7 nice -n19 mvn -U -Dmirage2.on=true -Dmirage2.deps.included=false package
...
[ERROR] Failed to execute goal on project additions: Could not resolve dependencies for project org.dspace.modules:additions:jar:6.3: Failed to collect dependencies at com.atmire:atmire-listings-and-reports-api:jar:6.x-2.10.8-0-SNAPSHOT: Failed to read artifact descriptor for com.atmire:atmire-listings-and-reports-api:jar:6.x-2.10.8-0-SNAPSHOT: Could not transfer artifact com.atmire:atmire-listings-and-reports-api:pom:6.x-2.10.8-0-SNAPSHOT from/to atmire.com-snapshots (https://atmire.com/artifactory/atmire.com-snapshots): Not authorized , ReasonPhrase:Unauthorized. -> [Help 1]
  • Great! I will have to send Atmire a note about this… but for now I can sync over my local ~/.m2 directory and the build completes
  • After the Maven build completed successfully I installed the updated code with Ant (make sure to delete the old spring directory):
$ cd dspace/target/dspace-installer
$ rm -rf /blah/dspacetest/config/spring
$ ant update
  • Database migrations take 10:18.287s during the first startup…
    • perhaps when we do the production CGSpace migration I can do this in advance and tell users not to make any submissions?
  • I had a mistake in my Solr internal URL parameter so DSpace couldn’t find it, but once I fixed that DSpace starts up OK!
  • Once the initial Discovery reindexing was completed (after three hours or so!) I started the Solr statistics UUID migration:
$ export JAVA_OPTS="-Xmx1024m -Dfile.encoding=UTF-8"
$ dspace solr-upgrade-statistics-6x -i statistics -n 250000
$ dspace solr-upgrade-statistics-6x -i statistics -n 1000000
$ dspace solr-upgrade-statistics-6x -i statistics -n 1000000
...
  • It’s taking about 35 minutes for 1,000,000 records…
  • Some issues towards the end of this core:
Exception: Error while creating field 'p_group_id{type=uuid,properties=indexed,stored,multiValued}' from value '10'
org.apache.solr.client.solrj.impl.HttpSolrServer$RemoteSolrException: Error while creating field 'p_group_id{type=uuid,properties=indexed,stored,multiValued}' from value '10'
        at org.apache.solr.client.solrj.impl.HttpSolrServer.executeMethod(HttpSolrServer.java:552)
        at org.apache.solr.client.solrj.impl.HttpSolrServer.request(HttpSolrServer.java:210)
        at org.apache.solr.client.solrj.impl.HttpSolrServer.request(HttpSolrServer.java:206)
        at org.apache.solr.client.solrj.request.AbstractUpdateRequest.process(AbstractUpdateRequest.java:124)
        at org.apache.solr.client.solrj.SolrServer.add(SolrServer.java:68)
        at org.apache.solr.client.solrj.SolrServer.add(SolrServer.java:54)
        at org.dspace.util.SolrUpgradePre6xStatistics.batchUpdateStats(SolrUpgradePre6xStatistics.java:161)
        at org.dspace.util.SolrUpgradePre6xStatistics.run(SolrUpgradePre6xStatistics.java:456)
        at org.dspace.util.SolrUpgradePre6xStatistics.main(SolrUpgradePre6xStatistics.java:365)
        at sun.reflect.NativeMethodAccessorImpl.invoke0(Native Method)
        at sun.reflect.NativeMethodAccessorImpl.invoke(NativeMethodAccessorImpl.java:62)
        at sun.reflect.DelegatingMethodAccessorImpl.invoke(DelegatingMethodAccessorImpl.java:43)
        at java.lang.reflect.Method.invoke(Method.java:498)
        at org.dspace.app.launcher.ScriptLauncher.runOneCommand(ScriptLauncher.java:229)
        at org.dspace.app.launcher.ScriptLauncher.main(ScriptLauncher.java:81)
  • So basically there are some documents that have IDs that have not been converted to UUID, and have not been labeled as “unmigrated” either…
    • Of these 101,257 documents, 90,000 are of type 5 (search), 9,000 are type storage, and 800 are type view, but it’s weird because if I look at their type/statistics_type using a facet the storage ones disappear…
    • For now I will export these documents from the statistics core and then delete them:
$ ./run.sh -s http://localhost:8081/solr/statistics -a export -o statistics-unmigrated.json -k uid -f '(*:* NOT id:/.{36}/) AND (*:* NOT id:/.+-unmigrated/)'
$ curl -s "http://localhost:8081/solr/statistics/update?softCommit=true" -H "Content-Type: text/xml" --data-binary "<delete><query>(*:* NOT id:/.{36}/) AND (*:* NOT id:/.+-unmigrated/)</query></delete>"
  • Now the UUID conversion script says there is nothing left to convert, so I can try to run the Atmire CUA conversion utility:
$ export JAVA_OPTS="-Xmx1024m -Dfile.encoding=UTF-8"
$ dspace dsrun com.atmire.statistics.util.update.atomic.AtomicStatisticsUpdateCLI -t 1
  • Experiment a bit with the Python country-converter library as it can convert between different formats (like ISO 3166 and UN m49)
    • We need to eventually find a format we can use for all CGIAR DSpaces…