Skip to content

Latest commit

 

History

History
279 lines (214 loc) · 11.4 KB

README.md

File metadata and controls

279 lines (214 loc) · 11.4 KB
Raw

DataONE Indexer

Also see RELEASE-NOTES.md

The DataONE Indexer is a system that processes index tasks created by other components. The DataONE Indexer comprises three main subsystems, each defined by its own helm subsystem chart:

  • index-worker: a subsystem implementing a Worker class to process index jobs in parallel
  • rabbitmq: a deployment of the RabbitMQ queue management system
  • solr: a deployment of the SOLR full text search system
flowchart TB
  subgraph "`**DataONE Indexer
  Helm Chart**`"
    A("`Index Worker
    *(1..n)*`")
    A -- sub chart --> C("`RabbitMQ
    *(1..n)*`")
    A -- sub chart --> D("`solr
    *(1..n)*`")
  end
Loading

Clients are expected to register index task messages to be processed in the RabbitMQ queue. Upon startup, the RabbitMQ workers register themselves as handlers of the index task messages. As messages enter the queue, RabbitMQ dispatches these to registered workers in parallel, and workers in turn process each associated object and insert a new index entry into SOLR.

See LICENSE.md for the details of distributing this software.

Building Docker image

The image can be built with either docker or nerdctl depending on which container environment you have installed. For example using Rancher Desktop configured to use nerdctl:

mvn clean package -DskipTests
nerdctl build -t dataone-index-worker:2.4.0 -f docker/Dockerfile --build-arg TAG=2.4.0 .

If you are building locally for Kubernetes on rancher-desktop, you'll need to set the namespace to k8s.io using a build command such as:

mvn clean package -DskipTests
nerdctl build -t dataone-index-worker:2.4.0 -f docker/Dockerfile --build-arg TAG=2.4.0 \
         --namespace k8s.io  .

Publish the image to GHCR

For the built image to be deployable in a remote kubernetes cluster, it must first be published to an image registry that is visible to Kubernetes. For example, we can make the published image available via the GitHub Container Registry (ghcr.io) so that it can be pulled by Kubernetes. For this to work, the image must be tagged with the ghcr.io URL, so it can be published. Then the image can be pushed to the registry after logging in with a suitable GitHub PAT.

Note that, for the image to be associated with a particular GitHub repository, a metadata LABEL can be added to the image that associates it when it is built - see this entry in the Dockerfile:

LABEL org.opencontainers.image.source="https://github.com/dataoneorg/dataone-indexer"

Commands for pushing the built image (example assuming tag is 2.4.0):

GITHUB_PAT="your-own-secret-GitHub-Personal-Access-Token-goes-here"
TAG=2.4.0

nerdctl tag dataone-index-worker:$TAG ghcr.io/dataoneorg/dataone-index-worker:$TAG
echo $GITHUB_PAT | nerdctl login ghcr.io -u DataONEorg --password-stdin
nerdctl push ghcr.io/dataoneorg/dataone-index-worker:$TAG

Once the image has been pushed, it may be private and will need to be made public and assigned to the dataone-indexer repository if the LABEL wasn't set as described above.

Publishing the Helm Chart

The helm chart may also be published to a helm repository for use by others as a top-level application deployment or as a dependency sub-chart. For example, we can publish the chart via the GitHub Container Registry (ghcr.io). For this to work, the chart must contain an annotation to associate it with the correct repository - see this entry in Chart.yaml:

# OCI Annotations - see https://github.com/helm/helm/pull/11204
sources:
  - https://github.com/dataoneorg/dataone-indexer

The chart must then be packaged:

helm package ./helm

...which creates a zipped tar file named dataone-indexer-{version}.tgz, where {version} reflects the chart version in Chart.yaml. The chart can then be published to the correct ghcr.io URL, after logging in with a suitable GitHub PAT:

(example assumes the chart version is 0.5.0)

GITHUB_PAT="your-own-secret-GitHub-Personal-Access-Token-goes-here"
TAG=2.4.0
helm push  dataone-indexer-0.5.0.tgz  oci://ghcr.io/dataoneorg/charts

NOTE the use of charts in the oci url, in order to distinguish helm charts from docker images.

Deploying the application via Helm

Helm provides a simple mechanism to install all application dependencies and configure the application in a single command. To deploy using helm to a release named d1index and also in a namespace named d1index, and then view the deployed pods and services, use a sequence like:

kubectl create namespace d1index
helm install -n d1index d1index ./helm
kubectl -n d1index get all

and to uninstall the helm, use:

helm -n d1index uninstall d1index

Note that this helm chart also installs rabbitmq and solr, which can be partially configured through the values.yaml file in the parent chart through exported child properties.

Authentication Notes

DataONE Authentication Token

In order to access and index private datasets on a Metacat instance, the dataone-indexer needs an authentication token, which may be obtained from DataONE administrators (see the Metacat Helm README). Upon startup, the indexer expects to find a Kubernetes Secret named: {{ .Release.Name }}-indexer-token, which contains the auth token associated with the key DataONEauthToken. The indexer can operate without this Secret, but will only be able to index public-readable datasets.

RabbitMQ

The rabbitmq service runs under the username and password that are set via values.yaml

rabbitmq:
  auth:
    username: rmq
    existingPasswordSecret: ""      ## (must contain key: `rabbitmq-password`)

...where existingPasswordSecret is the name of a Kubernetes secret that contains the password, identified by a key named rabbitmq-password.

NOTE: it appears that this information is cached on a PersistentVolumeClaim that is created automatically by rabbitmq. If the credentials are changed in values.yaml and/or the secret, therefore, authentication will fail because they will conflict with the cached values in the PVC. If you are just testing, the problem can be resolved by deleting the PVC. In production, the PVC would also be used for maintaining durable queues, and so it may not be reasonable to delete the PVC. You can get the name and identifiers of the PVCs with kubectl -n d1index get pvc.

Running the IndexWorker in the docker container

The docker image assumes that the deployment configuration file exists to configure endpoint addresses and credentials. To run the indexer, ensure that theDATAONE_INDEXER_CONFIG is set in the environment and contains the absolute path to the configuration file for the indexer. This path must be accessible in the container, so you will likely want to mount a volume to provide the edited properties file. You can then run it using a command like:

nerdctl run -it \
    -e DATAONE_INDEXER_CONFIG=/var/lib/dataone-indexer/dataone-indexer.properties \
    -v `pwd`/helm/config/dataone-indexer.properties:/var/lib/dataone-indexer/dataone-indexer.properties \
    dataone-index-worker:2.4.0

A Note on SOLR Authentication

The helm installation does not currently configure solr with authentication enabled, since the service is not exposed outside the Kubernetes cluster. Mentions of logins in the following sections can therefore be ignored. However, this should be changed to use authentication if connecting to a solr instance outside the cluster.

Checking if SOLR is configured

Logging in using the SOLR_AUTHENTICATION_OPTS and SOLR_AUTH_TYPE env variables (if applicable) allows the solr command to be executed to check the server status:

$ export SOLR_AUTH_TYPE=basic
$ export SOLR_AUTHENTICATION_OPTS="-Dbasicauth=${SOLR_ADMIN_USERNAME}:${SOLR_ADMIN_PASSWORD}"
$ solr status -z ${SOLR_ZK_HOSTS} -c ${SOLR_COLLECTION}

Found 1 Solr nodes:

Solr process 8 running on port 8983
{
  "solr_home":"/opt/bitnami/solr/server/solr",
  "version":"9.0.0 a4eb7aa123dc53f8dac74d80b66a490f2d6b4a26 - janhoy - 2022-05-05 01:00:08",
  "startTime":"2022-10-11T07:08:50.155Z",
  "uptime":"0 days, 0 hours, 21 minutes, 52 seconds",
  "memory":"70.9 MB (%13.8) of 512 MB",
  "cloud":{
    "ZooKeeper":"d1index-zookeeper:2181/solr",
    "liveNodes":"3",
    "collections":"1"}}

SOLR Dashboard

Once the SOLR server is up and running, connect to the SOLR Dashboard by creating a kube proxy, and then browse to the local address:

k8 port-forward -n d1index service/d1index-solr 8983:8983 & echo "Solr URL: 127.0.0.1:8983/solr/"

You'll need to log in with the helm-configured SOLR admin user and password, if applicable.

Once the proxy is set up, you can also run API calls from the ConfigSet API and Collections API.

curl -u ${SOLR_ADMIN_USERNAME}:${SOLR_ADMIN_PASSWORD} http://localhost:8983/solr/admin/configs?action=CREATE\&name=dataone-index --header "Content-Type:text/xml" -X POST -d @dataone-index.zip
{
  "responseHeader":{
    "status":0,
    "QTime":5974}}
curl -u ${SOLR_ADMIN_USERNAME}:${SOLR_ADMIN_PASSWORD} http://localhost:8983/solr/admin/configs?action=list
curl -u ${SOLR_ADMIN_USERNAME}:${SOLR_ADMIN_PASSWORD} http://localhost:8983/solr/admin/collections?action=list

Admin tools for rabbitmq

Once rabbitmq is configured, the web console can be accessed by port-forwarding.

k8 -n jones port-forward pod/d1index-rabbitmq-0 15672:15672 &

then login to the Rabbitmq web console: http://localhost:15672

You can also download a copy of rabbitmqadmin from http://localhost:15672/cli/rabbitmqadmin, and the rabbitmqadmin command can be used to interact with the server. First, you need to set up a config file for rabbitmqadmin that provides some default values:

$ cat rmq.conf
[default]
hostname = d1index-rabbitmq-headless
port = 15672
username = rmq
password = your-client-pw-here
declare_vhost = / # Used as default for declare / delete only
vhost = /         # Used as default for declare / delete / list
  • List exchanges and queues
    • rabbitmqadmin -c rmq.conf -N default -U rmq -p $RMQPW list exchanges --vhost=/
    • rabbitmqadmin -c rmq.conf -N default -U rmq -p $RMQPW list queues --vhost=/
  • Declare exchanges, queues, and bindings
    • rabbitmqadmin -c rmq.conf -N default declare exchange name=testexchange type=direct -U rmq -p $RMQPW --vhost=/
    • rabbitmqadmin -c rmq.conf -N default declare queue name=testqueue type=direct -U rmq -p $RMQPW --vhost=/
    • rabbitmqadmin -c rmq.conf -N default -U rmq -p $RMQPW declare binding source=testexchange destination=testqueue routing_key=testqueue --vhost=/
  • Publish a bunch of messages to a queue
for n in $(seq 1 30); do echo $n; rabbitmqadmin -c rmq.conf -N default -U rmq -p $RMQPW publish exchange=testexchange routing_key=testqueue payload="Message: ${n}" --vhost=/; done

History

This is a refactored version of the original DataONE d1_cn_index_processor that runs completely independently of other DataONE Coordinating Node services. It is intended to be deployed in a Kubernetes cluster environment, but is written such that it can also be deployed in other environments as needed.