Kubernetes backup

StackState for Kubernetes troubleshooting Self-hosted

Overview

The Kubernetes setup for StackState has a built-in backup and restore mechanism that can be configured to store backups to the local clusters, to AWS S3 or to Azure Blob Storage.

Backup scope

The following data can be automatically backed up:

Configuration and topology data stored in StackGraph is backed up when the Helm value backup.stackGraph.enabled is set to true.
Metrics stored in StackState's Victoria Metrics instance(s) is backed up when the Helm value victoria-metrics-0.backup.enabled and victoria-metrics-1.backup.enabled are set to true.
Telemetry data stored in StackState's Elasticsearch instance is backed up when the Helm value backup.elasticsearch.enabled is set to true.

The following data will not be backed up:

In transit topology and telemetry updates stored in Kafka - these only have temporary value and would be of no use when a backup is restored
Master node negotiations state stored in ZooKeeper - this runtime state would be incorrect when restored and will be automatically determined at runtime
Kubernetes configuration state and raw persistent volume state - this state can be rebuilt by re-installing StackState and restoring the backups.
Kubernetes logs - these are ephemeral.

Storage options

Backups are sent to an instance of MinIO (min.io), which is automatically started by the stackstate Helm chart when automatic backups are enabled. MinIO is an object storage system with the same API as AWS S3. It can store its data locally or act as a gateway to AWS S3 (min.io), Azure BLob Storage (min.io) and other systems.

The built-in MinIO instance can be configured to store the backups in three locations:

Enable backups

AWS S3

Encryption

Amazon S3-managed keys (SSE-S3) should be used when encrypting S3 buckets that store the backups.

⚠️ Encryption with AWS KMS keys stored in AWS Key Management Service (SSE-KMS) isn't supported. This will result in errors such as this one in the Elasticsearch logs:

Caused by: org.elasticsearch.common.io.stream.NotSerializableExceptionWrapper: sdk_client_exception: Unable to verify integrity of data upload. Client calculated content hash (contentMD5: ZX4D/ZDUzZWRhNDUyZTI1MTc= in base 64) didn't match hash (etag: c75faa31280154027542f6530c9e543e in hex) calculated by Amazon S3. You may need to delete the data stored in Amazon S3. (metadata.contentMD5: null, md5DigestStream: com.amazonaws.services.s3.internal.MD5DigestCalculatingInputStream@5481a656, bucketName: stackstate-elasticsearch-backup, key: tests-UG34QIV9s32tTzQWdPsZL/master.dat)",

To enable scheduled backups to AWS S3 buckets, add the following YAML fragment to the Helm values.yaml file used to install StackState:

backup:
  enabled: true
  stackGraph:
    bucketName: AWS_STACKGRAPH_BUCKET
  elasticsearch:
    bucketName: AWS_ELASTICSEARCH_BUCKET
victoria-metrics-0:
  backup:
    bucketName: AWS_VICTORIA_METRICS_BUCKET
victoria-metrics-1:
  backup:
    bucketName: AWS_VICTORIA_METRICS_BUCKET
minio:
  accessKey: YOUR_ACCESS_KEY
  secretKey: YOUR_SECRET_KEY
  s3gateway:
    enabled: true
    accessKey: AWS_ACCESS_KEY
    secretKey: AWS_SECRET_KEY

Replace the following values:

YOUR_ACCESS_KEY and YOUR_SECRET_KEY are the credentials that will be used to secure the MinIO system. These credentials are set on the MinIO system and used by the automatic backup jobs and the restore jobs. They're also required if you want to manually access the MinIO system.
- YOUR_ACCESS_KEY should contain 5 to 20 alphanumerical characters.
- YOUR_SECRET_KEY should contain 8 to 40 alphanumerical characters.
AWS_ACCESS_KEY and AWS_SECRET_KEY are the AWS credentials for the IAM user that has access to the S3 buckets where the backups will be stored. See below for the permission policy that needs to be attached to that user.
AWS_STACKGRAPH_BUCKET, AWS_ELASTICSEARCH_BUCKET and AWS_VICTORIA_METRICS_BUCKET are the names of the S3 buckets where the backups should be stored. Note: The names of AWS S3 buckets are global across the whole of AWS, therefore the S3 buckets with the default name (sts-elasticsearch-backup, sts-stackgraph-backup and sts-victoria-metrics-backup ) will probably not be available.

The IAM user identified by AWS_ACCESS_KEY and AWS_SECRET_KEY must be configured with the following permission policy to access the S3 buckets:

{
    "Version": "2012-10-17",
    "Statement": [
        {
            "Sid": "AllowListMinioBackupBuckets",
            "Effect": "Allow",
            "Action": [
                "s3:ListBucket",
                "s3:GetBucketLocation"
            ],
            "Resource": [
                "arn:aws:s3:::AWS_STACKGRAPH_BUCKET",
                "arn:aws:s3:::AWS_ELASTICSEARCH_BUCKET",
                "arn:aws:s3:::AWS_VICTORIA_METRICS_BUCKET"
            ]
        },
        {
            "Sid": "AllowWriteMinioBackupBuckets",
            "Effect": "Allow",
            "Action": [
                "s3:PutObject",
                "s3:GetObject",
                "s3:DeleteObject"
            ],
            "Resource": [
                "arn:aws:s3:::AWS_STACKGRAPH_BUCKET/*",
                "arn:aws:s3:::AWS_ELASTICSEARCH_BUCKET/*",
                "arn:aws:s3:::AWS_VICTORIA_METRICS_BUCKET/*"
            ]
        }
    ]
}

Azure Blob Storage

To enable backups to an Azure Blob Storage account, add the following YAML fragment to the Helm values.yaml file used to install StackState:

backup:
  enabled: true
minio:
  accessKey: AZURE_STORAGE_ACCOUNT_NAME
  secretKey: AZURE_STORAGE_ACCOUNT_KEY
  azuregateway:
    enabled: true

Replace the following values:

AZURE_STORAGE_ACCOUNT_NAME - the Azure storage account name (learn.microsoft.com)
AZURE_STORAGE_ACCOUNT_KEY - the Azure storage account key (learn.microsoft.com) where the backups should be stored.

The StackGraph, Elasticsearch and Victoria Metrics backups are stored in BLOB containers called sts-stackgraph-backup, sts-elasticsearch-backup and sts-victoria-metrics-backup respectively. These names can be changed by setting the Helm values backup.stackGraph.bucketName, backup.elasticsearch.bucketName, victoria-metrics-0.backup.bucketName and victoria-metrics-1.backup.bucketName respectively.

Kubernetes storage

If MinIO is configured to store its data in Kubernetes storage, a PersistentVolumeClaim (PVC) is used to request storage from the Kubernetes cluster. The kind of storage allocated depends on the configuration of the cluster.

It's advised to use AWS S3 for clusters running on Amazon AWS and Azure Blob Storage for clusters running on Azure for the following reasons:

Kubernetes clusters running in a cloud provider usually map PVCs to block storage, such as Elastic Block Storage for AWS or Azure Block Storage. Block storage is expensive, especially for large data volumes.
Persistent Volumes are destroyed when the cluster that created them is destroyed. That means an (accidental) deletion of your cluster will also destroy all backups stored in Persistent Volumes.
Persistent Volumes can't be accessed from another cluster. That means that it isn't possible to restore StackState from a backup taken on another cluster.

To enable backups to cluster-local storage, enable MinIO by adding the following YAML fragment to the Helm values.yaml file used to install StackState:

backup:
  enabled: true
minio:
  accessKey: YOUR_ACCESS_KEY
  secretKey: YOUR_SECRET_KEY
  persistence:
    enabled: true

Replace the following values:

YOUR_ACCESS_KEY and YOUR_SECRET_KEY - the credentials that will be used to secure the MinIO system. The automatic backup jobs and the restore jobs will use them. They're also required to manually access the MinIO storage. YOUR_ACCESS_KEY should contain 5 to 20 alphanumerical characters and YOUR_SECRET_KEY should contain 8 to 40 alphanumerical characters.

Configuration and topology data (StackGraph)

Configuration and topology data (StackGraph) backups are full backups, stored in a single file with the extension .graph. Each file contains a full backup and can be moved, copied or deleted as required.

Disable scheduled backups

When backup.enabled is set to true, scheduled StackGraph backups are enabled by default. To disable scheduled StackGraph backups only, set the Helm value backup.stackGraph.scheduled.enabled to false.

Disable restores

When backup.enabled is set to true, StackGraph restores are enabled by default. To disable StackGraph restore functionality only, set the Helm value backup.stackGraph.restore.enabled to false.

Backup schedule

By default, the StackGraph backups are created daily at 03:00 AM server time.

The backup schedule can be configured using the Helm value backup.stackGraph.scheduled.schedule, specified in Kubernetes cron schedule syntax (kubernetes.io).

Backup retention

By default, the StackGraph backups are kept for 30 days. As StackGraph backups are full backups, this can require a lot of storage.

The backup retention delta can be configured using the Helm value backup.stackGraph.scheduled.backupRetentionTimeDelta, specified in Python timedelta format (python.org).

Metrics (Victoria Metrics)

Victoria Metrics use incremental backups without versioning of a bucket, it means that the new backup replaces completely the previous one.

In case you run into one of the following situations:

mount an empty volume to /storage directory of Victoria Metrics instances
delete the /storage directory or files inside from Victoria Metrics instances

The next (empty) backup created will be labeled with a new version and the previous one, before the volume was emptied, will be preserved. Both backups will be from that moment on listed as available for restore.

Metrics (Victoria Metrics) use instant snapshots to store data in incremental backups. Many instances of Victoria Metrics can store backups to the same bucket, each of them will be stored in separated directory. All files located in the directory should be treated as a single whole and can only be moved, copied or deleted as a whole.

High Available deployments should be deployed with two instances of Victoria Metrics. Backups are enabled/configured independently for each of them.

The following code snippets/commands are provided for the first instance of Victoria Metric victoria-metrics-0. To backup/configure the second instance you should use victoria-metrics-1

Enable scheduled backups

Backups of Victoria Metrics are disabled by default. To enabled scheduled Victoria Metrics backups, set the Helm value victoria-metrics-0.backup.enabled to true.

Victoria Metrics backups requires to enable backups.

Enable restores

Restore functionality of Victoria Metrics are disabled by default. To enabled restore functionality of Victoria Metrics, set the Helm value victoria-metrics-0.restore.enabled to true.

Victoria Metrics restore functionality requires to enable backups.

Backup schedule

By default, the Victoria Metrics backups are created every 1h:

victoria-metrics-0 - 25 minutes past the hour
victoria-metrics-1 - 35 minutes past the hour

The backup schedule can be configured using the Helm value victoria-metrics-0.backup.bucketName according cronexpr format

Telemetry data (Elasticsearch)

The telemetry data (Elasticsearch) snapshots are incremental and stored in files with the extension .dat. The files in the Elasticsearch backup storage location should be treated as a single whole and can only be moved, copied or deleted as a whole.

The configuration snippets provided in the section enable backups will enable daily Elasticsearch snapshots.

Disable scheduled snapshots

When backup.enabled is set to true, scheduled Elasticsearch snapshots are enabled by default. To disable scheduled Elasticsearch snapshots only, set the Helm value backup.elasticsearch.scheduled.enabled to false.

Disable restores

When backup.enabled is set to true, Elasticsearch restores are enabled by default. To disable Elasticsearch restore functionality only, set the Helm value backup.elasticsearch.restore.enabled to false.

Snapshot schedule

By default, Elasticsearch snapshots are created daily at 03:00 AM server time.

The backup schedule can be configured using the Helm value backup.elasticsearch.scheduled.schedule, specified in Elasticsearch cron schedule syntax (elastic.co).

Snapshot retention

By default, Elasticsearch snapshots are kept for 30 days, with a minimum of 5 snapshots and a maximum of 30 snapshots.

The retention time and number of snapshots kept can be configured using the following Helm values:

backup.elasticsearch.scheduled.snapshotRetentionExpireAfter, specified in Elasticsearch time units (elastic.co).
backup.elasticsearch.scheduled.snapshotRetentionMinCount
backup.elasticsearch.scheduled.snapshotRetentionMaxCount

By default, the retention task itself runs daily at 1:30 AM UTC (elastic.co). If you set snapshots to expire faster than within a day, for example for testing purposes, you will need to change the schedule for the retention task.

Snapshot indices

By default, a snapshot is created for Elasticsearch indices with names that start with sts.

The indices for which a snapshot is created can be configured using the Helm value backup.elasticsearch.scheduled.indices, specified in JSON array format (w3schools.com).

Restore backups and snapshots

Scripts to list and restore backups and snapshots can be found in the restore directory of the StackState Helm chart repository (github.com). To use the scripts, download them from GitHub or checkout the repository.

Before you use the scripts, ensure that:

The kubectl binary is installed and configured to connect to:
1. The Kubernetes cluster where StackState has been installed.
2. The namespace within that cluster where StackState has been installed.
The following Helm values have been correctly set:
1. backup.enabled is set to true.
2. backup.stackGraph.restore.enabled isn't set to false (to access StackGraph backups).
3. backup.elasticsearch.restore.enabled isn't set to false (to access Elasticsearch snapshots).
4. victoria-metrics-0.restore.enabled or victoria-metrics-1.restore.enabled isn't set to false (to access Victoria Metrics snapshots).

List StackGraph backups

To list the StackGraph backups, execute the following command:

./restore/list-stackgraph-backups.sh

The output should look like this:

job.batch/stackgraph-list-backups-20210222t111942 created
Waiting for job to start...
=== Listing StackGraph backups in bucket "sts-stackgraph-backup"...
sts-backup-20210215-0300.graph
sts-backup-20210216-0300.graph
sts-backup-20210217-0300.graph
sts-backup-20210218-0300.graph
sts-backup-20210219-0300.graph
sts-backup-20210220-0300.graph
sts-backup-20210221-0300.graph
sts-backup-20210222-0300.graph
===
job.batch "stackgraph-list-backups-20210222t111942" deleted

The timestamp when the backup was taken is part of the backup name.

Lines in the output that start with Error from server (BadRequest): are expected. They appear when the script is waiting for the pod to start.

Restore a StackGraph backup

To avoid the unexpected loss of existing data, a backup can only be restored on a clean environment by default. If you are completely sure that any existing data can be overwritten, you can override this safety feature by using the command -force. Only execute the restore command when you are sure that you want to restore the backup.

To restore a StackGraph backup on a clean environment, select a backup name and pass it as the first parameter in the following command:

./restore/restore-stackgraph-backup.sh sts-backup-20210216-0300.graph

To restore a StackGraph backup on an environment with existing data, select a backup name and pass it as the first parameter in the following command next to a second parameter -force:

Note that existing data will be overwritten when the backup is restored.

Only do this if you are completely sure that any existing data can be overwritten.

./restore/restore-stackgraph-backup.sh sts-backup-20210216-0300.graph -force

The output should look like this:

job.batch/stackgraph-restore-20210222t112142 created
Waiting for job to start...
=== Downloading StackGraph backup "sts-backup-20210216-0300.graph" from bucket "sts-stackgraph-backup"...
download: s3://sts-stackgraph-backup/sts-backup-20210216-1252.graph to ../../tmp/sts-backup-20210216-0300.graph
=== Importing StackGraph data from "sts-backup-20210216-0300.graph"...
WARNING: An illegal reflective access operation has occurred
WARNING: Illegal reflective access by org.codehaus.groovy.vmplugin.v7.Java7$1 (file:/opt/docker/lib/org.codehaus.groovy.groovy-2.5.4.jar) to constructor java.lang.invoke.MethodHandles$Lookup(java.lang.Class,int)
WARNING: Please consider reporting this to the maintainers of org.codehaus.groovy.vmplugin.v7.Java7$1
WARNING: Use --illegal-access=warn to enable warnings of further illegal reflective access operations
WARNING: All illegal access operations will be denied in a future release
===
job.batch "stackgraph-restore-20210222t112142" deleted

In case you are running a restore command missing the -force flag on a non-empty database the output will contain an error like this:

ERROR com.stackvista.graph.migration.Restore - Restore isn't possible in a non empty.

Lines that starts with WARNING: are expected. They're generated by Groovy running in JDK 11 and can be ignored.

List Victoria Metrics backups

To list the Victoria Metrics backups, execute the following command:

./restore/list-victoria-metrics-backups.sh

The output should look like this:

job.batch/victoria-metrics-list-backups-20231016t125557 created
Waiting for job to start...
Waiting for job to start...
=== Fetching timestamps of last completed incremental backups
victoria-metrics-0 victoria-metrics-0-20231128160000 "Wed, 29 Nov 2023 07:36:00 GMT"
victoria-metrics-0 victoria-metrics-0-20231129092200 "Wed, 29 Nov 2023 10:56:00 GMT"

===
job.batch "victoria-metrics-list-backups-20231016t125557" deleted

where you can see the Victoria metrics instance, the specific backup version and the last time a backup was completed.

Restore a Victoria Metrics backup

Restore functionality always overrides data. You must be careful to avoid the unexpected loss of existing data.

Restore functionality requires to stop an instance of Victoria Metric while the process.

All new metrics will be cached by vmagent while the restore process, please ensure the vmagent has enough memory to cache metrics.

To restore a Victoria Metrics backup, select an instance name and a backup version and pass them as parameters in the following command:

./restore/restore-victoria-metrics-backup.sh victoria-metrics-0 victoria-metrics-0-20231128160000

The output should look like this:

=== Scaling down the Victoria Metrics instance
statefulset.apps/stackstate-victoria-metrics-0 scaled
=== Allowing pods to terminate
=== Starting restore job
job.batch/victoria-metrics-restore-backup-20231017t092607 created
=== Restore job started. Follow the logs with the following command:
kubectl logs --follow job/victoria-metrics-restore-backup-20231017t092607
=== After the job has completed clean up the job and scale up the Victoria Metrics instance pods again with the following commands:
kubectl delete job victoria-metrics-restore-backup-20231017t092607
kubectl scale statefulsets stackstate-victoria-metrics-0 --replicas=1

Then follow logs to check the job status

2023-10-17T07:26:42.564Z	info	VictoriaMetrics/lib/backup/actions/restore.go:194	restored 53072307269 bytes from backup in 0.445 seconds; deleted 639118752 bytes; downloaded 1204539 bytes
2023-10-17T07:26:42.564Z	info	VictoriaMetrics/app/vmrestore/main.go:64	gracefully shutting down http server for metrics at ":8421"
2023-10-17T07:26:42.564Z	info	VictoriaMetrics/app/vmrestore/main.go:68	successfully shut down http server for metrics in 0.000 seconds

After completion (ensure if the backup has been restored successfully), it's needed to follow commands printed by the earlier command:

delete the restore job
scale up the Victoria Metrics instance

List Elasticsearch snapshots

To list the Elasticsearch snapshots, execute the following command:

./restore/list-elasticsearch-snapshots.sh

The output should look like this:

job.batch/elasticsearch-list-snapshots-20210224t133115 created
Waiting for job to start...
Waiting for job to start...
=== Listing Elasticsearch snapshots in snapshot repository "sts-backup" in bucket "sts-elasticsearch-backup"...
sts-backup-20210219-0300-mref7yrvrswxa02aqq213w
sts-backup-20210220-0300-yrn6qexkrdgh3pummsrj7e
sts-backup-20210221-0300-p481sih8s5jhre9zy4yw2o
sts-backup-20210222-0300-611kxendsvh4hhkoosr4b7
sts-backup-20210223-0300-ppss8nx40ykppss8nx40yk
===
job.batch "elasticsearch-list-snapshots-20210224t133115" deleted

The timestamp when the backup was taken is part of the backup name.

Delete Elasticsearch indices

To delete existing Elasticsearch indices so that a snapshot can be restored, follow these steps.

Stop indexing - scale down all *2es deployments to 0:
```
kubectl scale --replicas=0 deployment/e2es
```

Open a port-forward to the Elasticsearch master:

kubectl port-forward service/stackstate-elasticsearch-master 9200:9200

Get a list of all indices:

curl "http://localhost:9200/_cat/indices?v=true"

The output should look like this:

health status index                          uuid                   pri rep docs.count docs.deleted store.size pri.store.size
green  open   sts_internal_events-2022.09.20 fTk7iEYtQI6ruVFuwNnbPw   1   0        125            0     71.7kb         71.7kb
green  open   .geoip_databases               GYA_c3i6QKenfFehnwBcAA   1   0         41            0     38.8mb         38.8mb
green  open   sts_multi_metrics-2022.09.20   MT1DceQPTDWVIfBJdl7qUg   3   0       1252            0    550.1kb        550.1kb

Delete an index with a following command:

curl -X DELETE "http://localhost:9200/INDEX_NAME?pretty"

Replace INDEX_NAME with the name of the index to delete, for example:

curl -X DELETE "http://localhost:9200/sts_internal_events-2021.02.19?pretty"

The output should be:
```
{
"acknowledged" : true
}
```

Restore an Elasticsearch snapshot

When a snapshot is restored, existing indices won't be overwritten.

See delete Elasticsearch indices.

To restore an Elasticsearch snapshot, select a snapshot name and pass it as the first parameter in the following command line. You can optionally specify a second parameter with a comma-separated list of the indices that should be restored. If not specified, all indices that match the Helm value backup.elasticsearch.scheduled.indices will be restored (default "sts*"):

./restore/restore-elasticsearch-snapshot.sh \
  sts-backup-20210223-0300-ppss8nx40ykppss8nx40yk \
  "<INDEX_TO_RESTORE>,<INDEX_TO_RESTORE>"

The output should look like this:

job.batch/elasticsearch-restore-20210229t152530 created
Waiting for job to start...
Waiting for job to start...
=== Restoring Elasticsearch snapshot "sts-backup-20210223-0300-ppss8nx40ykppss8nx40yk" from snapshot repository "sts-backup" in bucket "sts-elasticsearch-backup"...
{
  "snapshot" : {
    "snapshot" : "sts-backup-20210223-0300-ppss8nx40ykppss8nx40yk",
    "indices" : [
      ".slm-history-1-000001",
      "ilm-history-1-000001",
      "sts_internal_events-2021.02.19"
    ],
    "shards" : {
      "total" : 3,
      "failed" : 0,
      "successful" : 3
    }
  }
}
===
job.batch "elasticsearch-restore-20210229t152530" deleted

The indices restored are listed in the output, as well as the number of failed and successful restore actions.

After the indices have been restored, scale up all *2es deployments:

kubectl scale --replicas=1 deployment/e2es

PreviousBackup and Restore NextConfiguration backup

Last updated 4 months ago