Skip to content

Latest commit

 

History

History
174 lines (120 loc) · 7.25 KB

UPGRADE.md

File metadata and controls

174 lines (120 loc) · 7.25 KB
Raw

Rename localrep to api: substra-backend 0.29.0 -> 0.30.0

Execute this command in the backend DB, before running django migrations,

UPDATE django_content_type SET app_label='api' WHERE app_label='localrep';
UPDATE django_migrations SET app='api' WHERE app='localrep';
ALTER TABLE localrep_algo RENAME TO api_algo;
ALTER TABLE localrep_algoinput RENAME TO api_algoinput;
ALTER TABLE localrep_algooutput RENAME TO api_algooutput;
ALTER TABLE localrep_channelorganization RENAME TO api_channelorganization;
ALTER TABLE localrep_computeplan RENAME TO api_computeplan;
ALTER TABLE localrep_computetask RENAME TO api_computetask;
ALTER TABLE localrep_computetaskinput RENAME TO api_computetaskinput;
ALTER TABLE localrep_computetaskoutput RENAME TO api_computetaskoutput;
ALTER TABLE localrep_computetaskoutputasset RENAME TO api_computetaskoutputasset;
ALTER TABLE localrep_datamanager RENAME TO api_datamanager;
ALTER TABLE localrep_datasample RENAME TO api_datasample;
ALTER TABLE localrep_datasample_data_managers RENAME TO api_datasample_data_managers;
ALTER TABLE localrep_lastevent RENAME TO api_lastevent;
ALTER TABLE localrep_model RENAME TO api_model;
ALTER TABLE localrep_performance RENAME TO api_performance;
ALTER TABLE localrep_profilingstep RENAME TO api_profilingstep;
ALTER TABLE localrep_taskdatasamples RENAME TO api_taskdatasamples;
ALTER TABLE localrep_taskprofiling RENAME TO api_taskprofiling;

Migration steps to change Celery Broker from RabbitMQ to Redis from substra-backend helm chart 21.1.0 and docker image 0.24.0 to upper version

Migration without task in queues

Set connect-backend events Deployment replicas to 0.

Wait for RabbitMQ queues to be empty with rabbitmqctl list_queues

Upgrade connect-backend upper than 0.18.0 to deploy Redis instead of RabbitMQ.

Set connect-backend events Deployment replicas to the number you set in your deployment if it hasn't been done by the upgrade.

Migration with tasks in queues

Set connect-backend worker Replicaset to 0.

Deploy a temporary RabbitMQ to keep the scheduled tasks : helm install tmp-rabbit -n NAMESPACE bitnami/rabbitmq

Migrate data from connect-backend RabbitMQ to tmp-rabbit RabbitMQ

celery -b "amqp://<username>:<password>@<url-rabbitmq-connect-backend>:<port>/" inspect scheduled > scheduled_tasks_rabbitmq.txt
celery migrate "amqp://<username>:<password>@<url-rabbitmq-connect-backend>:<port>/" "amqp://<username>:<password>@<url-tmp-rabbitmq>:<port>/"
celery -b "amqp://<username>:<password>@<url-tmp-rabbitmq>:<port>/" inspect scheduled  > scheduled_tasks_tmp_rabbitmq.txt
diff scheduled_tasks_rabbitmq.txt scheduled_tasks_tmp_rabbitmq.txt

Upgrade connect-backend upper than 0.18.0 to deploy Redis instead of RabbitMQ. Be sure to have a backup, it will remove connect-backend RabbitMQ. RabbitMQ backup documentation : https://www.rabbitmq.com/backup.html

Migrate data from tmp-rabbit RabbitMQ to connect-backend Redis

celery migrate "amqp://<username>:<password>@<url-tmp-rabbitmq>:<port>/" "redis://<username>:<password>@<url-redis-connect-backend>:<port>/"
celery -b "redis://<username>:<password>@<url-redis-connect-backend>:<port>/" inspect scheduled  > scheduled_tasks_redis.txt
diff scheduled_tasks_tmp_rabbitmq.txt scheduled_tasks_redis.txt

Set connect-backend worker Replicaset to the number you set in your deployment.

Migration steps for Merge Metric and Algo view from substra-backend 0.11.0 to upper version

Be sure to backup Django model databases and MinIO data before doing those steps

Backup and restore data

Backup and restore MinIO data

There are several buckets that need to be backup

substra-algo
substra-datamanager
substra-datasample
substra-metrics
substra-model

Backup

kubectl exec -it -n NAMESPACE $(kubectl get pod -l app.kubernetes.io/name=minio  -n NAMESPACE -o name) -- mc mirror --remove --preserve $MINIO_ENV/<bucket> $BACKUPS_DIR/$BACKUP_NAME
kubectl cp NAMESPACE/$(kubectl get pod -l app.kubernetes.io/name=minio  -n NAMESPACE -o custom-columns=":metadata.name" --no-headers):/$BACKUPS_DIR/$BACKUP_NAME /PATH/TO/BACKUPS/<bucket>

Restore

kubectl cp /PATH/TO/BACKUPS/<bucket> NAMESPACE/$(kubectl get pod -l app.kubernetes.io/name=minio  -n NAMESPACE -o custom-columns=":metadata.name" --no-headers):$BACKUPS_DIR/$BACKUP_NAME -c server
kubectl exec -it -n NAMESPACE $(kubectl get pod -l app.kubernetes.io/name=minio  -n NAMESPACE -o name) -- mc mirror --remove --preserve $BACKUPS_DIR/$BACKUP_NAME $MINIO_ENV/<bucket>

Backup and restore Django databases

Backup

localrep
kubectl exec -it -n NAMESPACE $(kubectl get pod -l app.kubernetes.io/name=substra-backend-server  -n NAMESPACE -o name) -c server -- python manage.py dumpdata localrep --indent=4 > /PATH/TO/BACKUPS/localrep.json
substrapp
kubectl exec -it -n NAMESPACE $(kubectl get pod -l app.kubernetes.io/name=substra-backend-server  -n NAMESPACE -o name) -c server -- python manage.py dumpdata substrapp --indent=4 > /PATH/TO/BACKUPS/substrapp.json

Restore

localrep
# Copy backup into the pod
kubectl cp /PATH/TO/BACKUPS/localrep.json NAMESPACE/$(kubectl get pod -l app.kubernetes.io/name=substra-backend-server -n NAMESPACE -o custom-columns=":metadata.name" --no-headers):localrep.json -c server
# Load data with Django
kubectl exec -it -n NAMESPACE $(kubectl get pod -l app.kubernetes.io/name=substra-backend-server  -n NAMESPACE -o name) -c server -- python manage.py loaddata localrep.json
substrapp
# Copy backup into the pod
kubectl cp /PATH/TO/BACKUPS/substrapp.json NAMESPACE/$(kubectl get pod -l app.kubernetes.io/name=substra-backend-server -n NAMESPACE -o custom-columns=":metadata.name" --no-headers):substrapp.json -c server
# Load data with Django
kubectl exec -it -n NAMESPACE $(kubectl get pod -l app.kubernetes.io/name=substra-backend-server  -n NAMESPACE -o name) -c server -- python manage.py loaddata substrapp.json

Upgrade procedure

For each org or a substra network you need to do manual minio migration before update substra-backend

Example with a substra-backend deployed in kubernetes namespace NAMESPACE

1 - Copy data from metrics minio bucket to algos one

kubectl exec -it -n NAMESPACE $(kubectl get pod -l app.kubernetes.io/name=minio  -n NAMESPACE -o name) -- mc cp --recursive /data/substra-metrics/metrics/ /data/substra-algo/algos

2 - Update substra-backend to the last version

Update your substra-backend version

3 - Sanity verification

Verify that we can access and download "metrics" with ALGO_METRICS algo from algo view !

substra config --profile PROFILE URL
substra login --profile PROFILE --username USER --password PASSWORD

# Check list works
substra list metric --profile PROFILE

# For each metric KEY show description and  download file
substra list metric --profile PROFILE | awk '(NR>1) {print $1}' | xargs -I % substra describe metric % --profile PROFILE
substra list metric --profile PROFILE | awk '(NR>1) {print $1}' | xargs -I % substra download metric % --profile PROFILE

3 - Clean data from metrics minio bucket

kubectl exec -it -n NAMESPACE $(kubectl get pod -l app.kubernetes.io/name=minio  -n NAMESPACE -o name) -- mc rm --force --recursive /data/substra-metrics