Skip to content

Latest commit

 

History

History
253 lines (178 loc) · 12.2 KB

README.md

File metadata and controls

253 lines (178 loc) · 12.2 KB
Raw

PulSolr

Circle CI

Dependencies

  • Ruby: 3.1.0

SolrCloud machines are running Solr 8.4.1

To install run bundle install

Solr Home Directory

solr_configs/ is configured to be the Solr home directory. Each application's Solr core configuration is a subdirectory of solr_configs/. Relevant Solr analysis library packages can be found within each core's lib/ directory.

Note that solr.xml is used for development and testing on this repository, but not in production. The production file is at https://github.com/pulibrary/princeton_ansible/blob/master/roles/pulibrary.solrcloud/templates/solr.xml.j2

SolrCloud Inventory

Current SolrCloud machines and their collections are enumerated in the solr inventory.

Adding a new core

This repository updates, but does not create, collections. To add a new collection, create its config here and deploy to get the config up to the server. Then use the UI to create the collection. Finally, you can add the collection to the deploy scripts so that it will be updated in future deployments (https://github.com/pulibrary/pul_solr/blob/main/config/collections.yml).

Note: Each collection should be created with a replication factor of 2 at minimum.

Connecting to Solr UI

There are capistrano tasks to connect to the Solr UI for managing solr that can be run from the project directory on your machine. You will need to be connected to VPN for the tasks to run.

  • Production Solr 8

    bundle exec cap production solr:console

  • Staging Solr 8

    bundle exec cap staging solr:console

Managing Orangelight Catalog Configsets

When we need to make a change to the orangelight config set that would break search we need to deploy a new config set, index with it, and then swap it in and retire the previous config set. See ADR#0005 for more details. In this case the following procedure should be followed:

  1. Copy the current production config set into a new directory, incrementing the version number. Change the name of the core in core.properties in the old config set so the new one is used for testing.
  2. Deploy the new config set
  3. Delete the solr collection in use for indexing. Recreate it using the new config set.
  4. Populate the reindex collection
  5. Swap the alias so the reindex collection becomes the production / in-use collection
  6. Delete the old collection and re-create it using the new config set
  7. Delete the old config set from this repository

Managing Configsets

After deploying one may list, upload, update, and delete Configsets using the following Capistrano tasks:

SOLR_URL=http://localhost:8983/solr bundle exec cap development configsets:list
SOLR_URL=http://localhost:8983/solr bundle exec cap development "configsets:update[dpul_new/conf,dpul-config]"
SOLR_URL=http://localhost:8983/solr bundle exec cap development "configsets:delete[dpul-config]"

Please note that, when uploading a directory for a new Configset from this repository, that the /conf subdirectory should be used (e. g. dpul/conf)

Managing Collections

Using Capistrano, one may create, reload, delete, and list Collections using the following tasks:

SOLR_URL=http://localhost:8983/solr bundle exec cap development collections:list
SOLR_URL=http://localhost:8983/solr bundle exec cap development "collections:create[dpul,dpul-config]"
SOLR_URL=http://localhost:8983/solr bundle exec cap development "collections:reload[dpul]"
SOLR_URL=http://localhost:8983/solr bundle exec cap development "collections:delete[dpul]"

SolrCloud Backups

Backups are implemented as a ruby service class wrapped in a rake task that's invoked by cron (scheduled via whenever / capistrano)

If a specific backup did not complete and you want more information, consult the log for the requeststatus and check it with the requeststatus api call.

Restoring a backup

Before restoring:

  1. Check the size of your backup with du -sh. For example, if your backup is /mnt/solr_backup/solr8/production/20240606/catalog-alma-production3-20240606.bk:

    /mnt/solr_backup/solr8/production/20240606/catalog-alma-production3-20240606.bk

  2. Check the size of the root file system on the machine on which you are performing the restore: df -h.

  3. If there is not enough storage on the root file system for the backup, resolve that issue before proceeding.

Restoring a backup (solr docs) is a matter of issuing the proper API call on the solr box, e.g.:

$ curl "http://localhost:8983/solr/admin/collections?action=RESTORE&name=pulfalight-staging-20210111.bk&collection=pulfalight-staging-restore&location=/mnt/solr_backup/staging/20210111"

To find the right values for this curl:

  1. SSH with a tunnel to a solr box (for example, ssh -L 8983:localhost:8983 deploy@lib-solr-staging5d)
  2. Run ls -t /mnt/solr_backup/solr8/production if you want to restore from a production backup, or ls -t /mnt/solr_backup/solr8/staging if you want to restore from a staging backup. The first directory will be the most recent backup.
    • The location parameter will be /mnt/solr_backup/solr8/{Environment}/{Backup Date}. In the example above, it is /mnt/solr_backup/staging/20210111
  3. Do an ls of the directory that you will use as the location parameter.
    For example, ls /mnt/solr_backup/staging/20210111.
    • The name parameter will be the directory name of the collection you want to restore. In the example above, it is pulfalight-staging-20210111.bk, referring to the /mnt/solr_backup/staging/20210111/pulfalight-staging-20210111.bk directory.
  4. Choose a name for a new collection you'd like to restore the data into. In your browser, go to http://localhost:8983/solr/#/~collections and confirm that there is not already a collection by that name.
    • The collection parameter will be the name of the collection.
  5. If the collection is very large (e.g. the catalog), the restore will probably time out. You can still move forward by adding the async parameter. It should be something unique to this particular restore.
  6. Run the curl with the location, name, collection, and -- if the collection is large -- async parameters you determined above.

If you ran this with an async ID in the async param, you can check the progress in your browser at: http://localhost:8983/solr/admin/collections?action=REQUESTSTATUS&requestid={your async id}

Specs

Start Solr via lando, then run the specs

lando start
rspec

To stop Solr again do lando stop

Fixtures

To get fixtures for specs, you can't just pull json documents out of existing solr cores because then you won't get the index-only fields.

Orangelight fixtures

You can get most records at the bibdata path, e.g. https://bibdata.princeton.edu/bibliographic/10166399/solr

If you need a scsb record, you have to go on the bibdata worker box, look in /data/scsb_temp, and grep for the id you want, e.g.

grep -r SCSB-3330744 .

Copy the line you found into a .xml file in your local copy of marc_liberation. Then grab the first two lines from the file in which you found that record, e.g.

head -n2 scsbupdate20171226_030300_19.xml

Copy those as the first two lines in your .xml file.

Then get the indexed json with the following command in the marc_liberation file: bundle exec traject -c marc_to_solr/lib/traject_config.rb -t xml scsb.xml -w Traject::JsonWriter

Finally, copy the output and paste it into a fixture file in your pul_solr project. It's nice to fix the formatting of the json; in vim you can do this with a little python utility ala :%!python -m json.tool

Pulfalight fixtures

In pulfalight we are pulling fixture documents straight from solr, which means we can't test against fields that are indexed but not stored. Pull a fixture like, e.g.:

http://localhost:8983/solr/pulfalight-dev/select?id=MC001-02-03&qt=document

Heap Dump

If you have no idea what is happening on solr you may want to try dumping the heap. This is difficult when GC is freezing everything. Keep trying and eventually you may be lucky!

  1. Run as pulsys user:

     sudo nsenter -t [pid_of_solr_process] -m # Enter the namespace of the Solr PID, so that systemd security settings don't block the dump
 sudo su - deploy # jmap must be run as the same user as solr
 jmap -dump:format=b,file=/home/deploy/solr.hprof [pid_of_solr_process]

  2. Then you'll want to look at it. Download the file to your machine

     scp deploy@lib-solrN:/home/deploy/solr.hprof .

  3. Download and install the eclipse memory analyzer application.

  4. You need to assign enough heap to the app to hold the entire heap that you dumped on the server. When you unzip it you get a mat directory:

    1. After installing the Memory Analyzer, open the Applications directory in Finder.
    2. Right click on Memory Analyzer and select "Show Package Contents"
    3. expand contents > eclipse > right click on "MemoryAnalyzer.ini" to edit.
    4. Change "-Xmx" to be a number bigger than the file you have.

  5. You're supposed to be able to double-click the 'mat' file but that doesn't work. you have to "Show Package Contents" > Contents > MacOS > run 'MemoryAnalyzer'

  6. Congratulations! You opened the application. Now open the heap file and wait a long time while it parses the file and the progress bar jumps around. It took about an hour for a 20g file for us.

  7. You want to look at the dominator tree to see how much heap is used by each object. Right-click the biggest one's thread (higher in the tree) > Java Basics > Thread Overview and Stacks. Expand the thread click the "total" button at the bottom so all of them will open up. Expand the first column (Object stack frame). Expand 'org.eclipse.jetty.servlet.ServletHandler.doHandle'. Click the first (local) frame. Look on the left, double-click the + to expand more properties, the thing that broke it was _originalURI. right-click > copy value.

Solr Docker

The docker directory contains a Dockerfile that serves as the base docker image for running Solr in CI and Lando.

This image:

  • adds a security.json which allows us to make changes to solr via basic auth. It runs with an embedded zookeeper on a separate port.
  • adds Solr plugins downloaded from the solrcloud role in princeton_ansible
  • contains scripts for solr setup in circleci and lando.

Update and Rebuild

quay.io

This is Red Hat's container registry (they also call it a repository), and where Ops is shifting to keeping images (as of May 2023). Contact the Ops team to become a member of the pulibrary organization. You must be a member in order to push an image to the remote repository.

Building and pushing an image:

cd docker/
docker login quay.io # login to quay.io
docker buildx create --use # only necessary the first time you want to build an image
docker buildx build --platform linux/arm64/v8,linux/amd64 -t quay.io/pulibrary/ci-solr:{solr version}-{Dockerfile version} --push .

For example, if you are building with solr 8.4 and Dockerfile 1.0.0:

cd docker/
docker login # login to quay.io
docker buildx create --use
docker buildx build --platform linux/arm64/v8,linux/amd64 -t qay.io/pulibrary/ci-solr:8.4-v1.0.0 --push .

Github Container Registry

You can also push to your own Github Container Registry, if you are just testing something out:

  1. Login to the container registry
  2. cd docker
  3. docker buildx create --use # only necessary the first time
  4. docker buildx build --platform linux/arm64/v8,linux/amd64 -t ghcr.io/[username]/ci-solr:{solr version}-{Dockerfile version} --push .

Docker hub - deprecated

Old images were pushed to the pulibrary docker hub organization - for older tags see https://hub.docker.com/r/pulibrary/ci-solr/tags