CGSpace Notes

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

September, 2024

in  Notes

2024-09-01

  • Upgrade CGSpace to DSpace 7.6.2

2024-09-05

  • Finalize work on migrating DSpace Angular from Yarn to NPM

2024-09-06

  • This morning Tomcat crashed due to an OOM kill:
Sep 06 00:00:24 server systemd[1]: tomcat9.service: A process of this unit has been killed by the OOM killer.
Sep 06 00:00:25 server systemd[1]: tomcat9.service: Main process exited, code=killed, status=9/KILL
Sep 06 00:00:25 server systemd[1]: tomcat9.service: Failed with result 'oom-kill'.
  • According to the system journal, it was a Node.js dspace-angular process that tried to allocate memory and failed, thus invoking the OOM killer
  • Currently I see high memory usage in those processes:
$ pm2 status
┌────┬──────────────┬─────────────┬─────────┬─────────┬──────────┬────────┬──────┬───────────┬──────────┬──────────┬──────────┬──────────┐
│ id │ name         │ namespace   │ version │ mode    │ pid      │ uptime │ ↺    │ status    │ cpu      │ mem      │ user     │ watching │
├────┼──────────────┼─────────────┼─────────┼─────────┼──────────┼────────┼──────┼───────────┼──────────┼──────────┼──────────┼──────────┤
│ 0  │ dspace-ui    │ default     │ 7.6.3-… │ cluster │ 994      │ 4D     │ 0    │ online    │ 0%       │ 3.4gb    │ dspace   │ disabled │
│ 1  │ dspace-ui    │ default     │ 7.6.3-… │ cluster │ 1015     │ 4D     │ 0    │ online    │ 0%       │ 3.4gb    │ dspace   │ disabled │
│ 2  │ dspace-ui    │ default     │ 7.6.3-… │ cluster │ 1029     │ 4D     │ 0    │ online    │ 0%       │ 3.4gb    │ dspace   │ disabled │
│ 3  │ dspace-ui    │ default     │ 7.6.3-… │ cluster │ 1042     │ 4D     │ 0    │ online    │ 0%       │ 3.4gb    │ dspace   │ disabled │
└────┴──────────────┴─────────────┴─────────┴─────────┴──────────┴────────┴──────┴───────────┴──────────┴──────────┴──────────┴──────────┘
  • I bet if I look in the logs I’d find some kind of heavy traffic on the frontend, causing high caching for Angular SSR

2024-09-08

  • Analyzing memory use in our DSpace hosts, which have 32GB of memory
    • Effective cache of PostgreSQL is estimated at 11GB, which seems way high since the database is only 2GB
    • Realistically this should be how we adjust, with PostgreSQL using ~8GB (or less) and each dspace-angular process pinned at 2GB…

Total - Solr - Tomcat Postgres - Nginx - Angular 31366 − (1024×4.4) − 7168 − (8×1024) − 512 - (4x2048) = 2796.4 left…

  • I put some of these changes in on DSpace Test and will monitor this week

2024-09-10

  • Some bot in South Africa made a ton of requests on the API and made the load hit the roof:
# grep -E '10/Sep/2024:[10-11]' /var/log/nginx/api-access.log | awk '{print $1}' | sort | uniq -c | sort -h
...
 149720 102.182.38.90
  • They are using several user agents so are obviously a bot:
Mozilla/5.0 (Windows NT 10.0; Win64; x64; rv:130.0) Gecko/20100101 Firefox/130.0
Mozilla/5.0 (Windows NT 10.0; Win64; x64; rv:109.0) Gecko/20100101 Firefox/111.0
Mozilla/5.0 (Windows; U; Windows NT 6.1; rv:11.0) Gecko/20100101 Firefox/11.0
  • I added them to the list of bot networks in nginx and the load went down

2024-09-11

  • Upgrade DSpace 7 Test to Ubuntu 24.04
  • I did some minor maintenance to test dspace-statistics-api with Python 3.12
    • I tagged version 1.4.4 and released it on GitHub

2024-09-14

  • Noticed a persistent higher than usual load on CGSpace and checked the server logs
    • Found some new data center subnets to block because they were making thousands of requests with normal user agents
  • I enabled HTTP/3 in nginx
  • I enabled the SSR patch in Angular: https://github.com/DSpace/dspace-angular/issues/3110

2024-09-16

  • Experiment with the dspace-statistics-api-js on DSpace 7 Test
    • In the past it always caused Solr to run out of memory, but I increased Solr’s heap from 2g to 3g and it runs without crashing
    • I attached VisualVM to Solr with a 3g and 4g heap and iterated over 1260 pages of results in the dspace-statistics-api-js:

Solr with 3g heap

Solr with 4g heap

2024-09-23

  • Upgrade PostgreSQL from version 14 to 15 on DSpace Test the same way I did last year:
# apt update
# apt install postgresql-15
# Update configs with Ansible
# systemctl stop tomcat9
# pg_ctlcluster 14 main stop
# tar -cvzpf var-lib-postgresql-14.tar.gz /var/lib/postgresql/14
# tar -cvzpf etc-postgresql-14.tar.gz /etc/postgresql/14
# pg_ctlcluster 15 main stop
# pg_dropcluster 15 main
# pg_upgradecluster 14 main
# pg_ctlcluster 15 main start
...

ERROR:  could not find function "xml_is_well_formed" in file "/usr/lib/postgresql/15/lib/pgxml.so"
ERROR:  function public.xml_is_well_formed(text) does not exist
ERROR:  could not find function "xml_is_well_formed" in file "/usr/lib/postgresql/15/lib/pgxml.so"
ERROR:  function public.xml_valid(text) does not exist

$ su - postgres
$ cat /tmp/generate-reindex.sql
SELECT 'REINDEX TABLE CONCURRENTLY ' || quote_ident(relname) || ' /*' || pg_size_pretty(pg_total_relation_size(C.oid)) || '*/;'
FROM pg_class C
LEFT JOIN pg_namespace N ON (N.oid = C.relnamespace)
WHERE nspname = 'public'
  AND C.relkind = 'r'
  AND nspname !~ '^pg_toast'
ORDER BY pg_total_relation_size(C.oid) ASC;
$ psql dspace < /tmp/generate-reindex.sql > /tmp/reindex.sql
$ <trim the extra stuff from /tmp/reindex.sql>
$ psql dspace < /tmp/reindex.sql
  • The database shrunk by 186MB!

2024-09-29

  • I upgraded the database on CGSpace to PostgreSQL 15