How to manually clean up logs

Cleaning up Ceph logs

Moving Ceph out of read-only mode

If you installed AI Center and use Ceph storage, take the following steps to move Ceph out of read-only mode:

1. Check if Ceph is at full capacity:

   kubectl -n rook-ceph exec -it deploy/rook-ceph-tools -- ceph status

   If Ceph is at full capacity, you must adjust the read-only threshold to start up the rgw gateways.
2. Scale down the ML skills:

   kubectl -n uipath scale deployment <skill> --replicas=0

3. Put the cluster in write mode:

   ceph osd set-full-ratio 0.95 <95 is the default value so you could increase to 96 and go up 
   incrementall>

4. Run Garbage Collection:

   kubectl -n rook-ceph exec -it deploy/rook-ceph-tools -- radosgw-admin gc process --include-all

5. When storage goes down, run the following commands:

   kubectl -n rook-ceph exec -it deploy/rook-ceph-tools -- ceph status
   kubectl -n rook-ceph exec -it deploy/rook-ceph-tools -- ceph df

   At this point, the storage should be lower, and the cluster should be healthy.

Disabling streaming logs

To ensure everything is in a good state, disable streaming logs by taking the following steps.

1. Disable aut-sync on UiPath and AI Center.
2. Disable streaming logs for AI Center.
3. If you have ML skills that have already been deployed, run the following commands:

   kubectl set env deployment [REPLICASET_NAME] LOGS_STREAMING_ENABLED=false

4. Find out which buckets use the most space:

   kubectl -n rook-ceph exec deploy/rook-ceph-tools -- radosgw-admin bucket stats | jq -r '["BucketName","NoOfObjects","SizeInKB"], ["--------------------","------","------"], (.[] | [.bucket, .usage."rgw.main"."num_objects", .usage."rgw.main".size_kb_actual]) | @tsv' | column -ts $'\t'

5. Install s3cmd to prepare for cleaning up the sf-logs:

   pip3 install awscli s3cmd
   export PATH=/usr/local/bin:$PATH

6. Clean up the sf-logs logs. For details, see How to clean up old logs stored in the sf-logs bundle.
7. Complete the cleanup operation:

   kubectl -n rook-ceph exec -it deploy/rook-ceph-tools -- radosgw-admin gc process --include-all

8. If the previous steps do not solve the issue, clean up the AI Center data.
9. Check if the storage was reduced:

   kubectl -n rook-ceph exec -it deploy/rook-ceph-tools -- ceph df

10. Once storage is no longer full, reduce the backfill setting:

    kubectl -n rook-ceph exec -it deploy/rook-ceph-tools -- ceph osd set-full-ratio 0.95

11. Check if the ML skills are affected by the multipart upload issue:

    echo $(kubectl -n rook-ceph exec deploy/rook-ceph-tools -- radosgw-admin bucket list --max-entries 10000000 --bucket train-data | jq '[.[] | select (.name | contains("_multipart")) | .meta.size] | add') | numfmt --to=iec-i

    If they are affected by this issue, and the returned value is high, you may need to do a backup and restore.

Cleaning up s3 logs

If you use an s3-compatible storage provider, take the following steps to clean up your logs:

1. Get the key for storage access.
2. Find the large items:

   export PATH=/usr/local/bin:$PATH
   
   kubectl -n rook-ceph get secrets ceph-object-store-secret -o yaml
    
    base64 decode the OBJECT_STORAGE_ACCESSKEY and OBJECT_STORAGE_SECRETKEY and OBJECT_STORAGE_EXTERNAL_HOST
   
   run: aws configure

   aws s3 ls --endpoint-url <AWS-ENDPOINT> --no-verify-ssl --recursive s3://<bucket name>
   
   So for sf-logs
   
   aws s3 ls --endpoint-url <AWS-ENDPOINT> --no-verify-ssl --recursive s3://sf-logs
   
   aws s3 ls --endpoint-url <AWS-ENDPOINT> --no-verify-ssl --recursive s3://sf-logs --recursive --human-readable --sumarize

3. Delete the sf-logs. For more details, see the AWS documentation.

   aws s3 rm --endpoint-url <AWS-ENDPOINT> --no-verify-ssl --recursive s3://sf-logs --include="2022* --exclude="2022_12_8"
   
   So craft an include and exclude command to help with this. use --dryrun first

4. Delete the train-data.