Sentry is a cross-platform crash reporting and aggregation platform.
This repository aims to support Sentry >=10 and move out from the deprecated Helm charts official repo.
Big thanks to the maintainers of the deprecated chart. This work has been partly inspired by it.
helm repo add sentry https://sentry-kubernetes.github.io/charts
For now the full list of values is not documented but you can get inspired by the values.yaml specific to each directory.
Make sure to revert the changes on Clickhouse replica counts if the change doesn't suit you.
This version introduces changes to definitions of ingest-consumers and workers. These changes allow to balance ingestion pipeline with more granularity.
- Ingest consumers: Templates for Deployment and HPA manifests are now separate for ingest-consumer-events, ingest-consumer-attachments and ingest-consumer-transactions.
- Workers: Templates for two additional worker Deployments added, each of them with its own HPA. By default they're configured for error- and transaction-related tasks processing, but queues to consume can be redefined for both.
Since labels are immutable in k8s Deployments, helm upgrade --force should be used to recreate ingest-consumer Deployments.
As an alternative, existing ingest-consumer Deployments can be removed manually with kubectl delete before upgrading helm release.
This version introduces a significant change by dropping support for Kafka Zookeeper and transitioning to Kafka Kraft mode. This change requires action on your part to ensure a smooth upgrade.
- Kafka Upgrade: We have upgraded from Kafka
23.0.7to27.1.2. This involves moving from Zookeeper to Kraft, requiring a fresh setup of Kafka.
-
Backup Your Data: Ensure all your data is backed up before starting the migration process.
-
Retrieve the Cluster ID from Zookeeper by executing:
kubectl exec -it <your-zookeeper-pod> -- zkCli.sh get /cluster/id
-
Deploy at least one Kraft controller-only in your deployment with zookeeperMigrationMode=true. The Kraft controllers will migrate the data from your Kafka ZkBroker to Kraft mode.
To do this, add the following values to your Zookeeper deployment when upgrading:
controller: replicaCount: 1 controllerOnly: true zookeeperMigrationMode: true broker: zookeeperMigrationMode: true kraft: enabled: true clusterId: "<your_cluster_id>"
-
Wait until all brokers are ready. You should see the following log in the broker logs:
INFO [KafkaServer id=100] Finished catching up on KRaft metadata log, requesting that the KRaft controller unfence this broker (kafka.server.KafkaServer) INFO [BrokerLifecycleManager id=100 isZkBroker=true] The broker has been unfenced. Transitioning from RECOVERY to RUNNING. (kafka.server.BrokerLifecycleManager)
In the controllers, the following message should show up:
Transitioning ZK migration state from PRE_MIGRATION to MIGRATION (org.apache.kafka.controller.FeatureControlManager)
-
Once all brokers have been successfully migrated, set 'broker.zookeeperMigrationMode=false' to fully migrate them.
broker: zookeeperMigrationMode: false
-
To conclude the migration, switch off migration mode on controllers and stop Zookeeper:
controller: zookeeperMigrationMode: false zookeeper: enabled: false
After the migration is complete, you should see the following message in your controllers:
[2023-07-13 13:07:45,226] INFO [QuorumController id=1] Transitioning ZK migration state from MIGRATION to POST_MIGRATION (org.apache.kafka.controller.FeatureControlManager) -
(Optional) If you would like to switch to a non-dedicated cluster, set 'controller.controllerOnly=false'. This will cause controller-only nodes to switch to controller+broker nodes.
At this point, you could manually decommission broker-only nodes by reassigning its partitions to controller-eligible nodes.
For more information about decommissioning a Kafka broker, check the official documentation.
Bumped dependencies:
- memcached > 6.5.9
- kafka > 23.0.7 - This is a major update, but only kafka version is updated. See bitnami charts' update note
- clickhouse > 3.7.0 - Supports priorityClassName and max_suspicious_broken_parts config.
- zookeeper > 11.4.11 - 2 Major updates from v9 to v11. See To v10 upgrade notes and To v11 upgrade notes
- rabbitmq > 11.16.2
Bumped dependencies:
- kafka > 22.1.3 - now supports Kraft. Note that the upgrade is breaking and that you have to start a new kafka from scratch to use it.
Example:
kafka:
zookeeper:
enabled: false
kraft:
enabled: true
Chart dependencies has been upgraded because of sentry requirements. Changes:
- The minimum required version of Postgresql is 14.5 (works with 15.x too)
Bumped dependencies:
- postgresql > 12.5.1 - latest wersion of chart with postgres 15
If Kafka is complaining about unknown or missing topic, please connect to kafka-0 and run
/opt/bitnami/kafka/bin/kafka-topics.sh --create --topic ingest-replay-recordings --bootstrap-server localhost:9092
Sentry version from 22.10.0 onwards should be using chart 17.x.x
- post process forwarder events and transactions topics are split in Sentry 22.10.0
You can delete the deployment "sentry-post-process-forward" as it's no longer needed.
sentry-worker may failed to start by #774.
If you encountered this issue, please reset counters-0, triggers-0 queues.
system.secret-key is removed
See https://github.com/sentry-kubernetes/charts/tree/develop/sentry#sentry-secret-key
Chart dependencies has been upgraded because of bitnami charts removal. Changes:
nginx.service.port: 80>nginx.service.ports.http: 80kafka.service.port>kafka.service.ports.client
Bumped dependencies:
- redis > 16.12.1 - latest version of chart
- kafka > 16.3.2 - chart aligned with zookeeper dependency, upgraded kafka to 3.11
- rabbit > 8.32.2 - latest 3.9.* image version of chart
- postgresql > 10.16.2 - latest wersion of chart with postgres 11
- nginx > 12.0.4 - latest version of chart
ClickHouse was reconfigured with sharding and replication in-mind, If you are using external ClickHouse, you don't need to do anything.
WARNING: You will lose current event data
Otherwise, you should delete the old ClickHouse volumes in-order to upgrade to this version.
The service annotions have been moved from the service section to the respective service's service sub-section. So what was:
service:
annotations:
alb.ingress.kubernetes.io/healthcheck-path: /_health/
alb.ingress.kubernetes.io/healthcheck-port: traffic-portwill now be set per service:
sentry:
web:
service:
annotations:
alb.ingress.kubernetes.io/healthcheck-path: /_health/
alb.ingress.kubernetes.io/healthcheck-port: traffic-port
relay:
service:
annotations:
alb.ingress.kubernetes.io/healthcheck-path: /api/relay/healthcheck/ready/
alb.ingress.kubernetes.io/healthcheck-port: traffic-portIf you were using clickhouse tabix externally, we disabled it per default.
If you were using clickhouse ImagePullSecrets, we unified the way it's used.
to simplify 1st time installations, backup value on clickhouse has been changed to false.
clickhouse.clickhouse.configmap.remote_servers.replica.backup
- the default value of features.orgSubdomains is now "false"
- the default mode of relay is now "proxy". You can change it through the values.yaml file
- we removed the
githubSsovariable for the oauth github configuration. It was using the old environment variable, that doesn't work with Sentry anymore. Just use the common github.xxxx configuration for both oauth & the application integration.
- The sentry.configYml value is now in a real yaml format
- If you were previously using
relay.asHook, the value is nowasHook
As Relay is now part of this chart your need to make sure you enable either Nginx or the Ingress. Please read the next paragraph for more information.
If you are using an ingress gateway (like istio), you have to change your inbound path from sentry-web to nginx.
By default, NGINX is enabled to allow sending the incoming requests to Sentry Relay or the Django backend depending on the path. When Sentry is meant to be exposed outside of the Kubernetes cluster, it is recommended to disable NGINX and let the Ingress do the same. It's recommended to go with the go to Ingress Controller, NGINX Ingress but others should work as well.
Note: if you are using NGINX Ingress, please set this annotation on your ingress : nginx.ingress.kubernetes.io/use-regex: "true".
If you are using additionalHostNames the nginx.ingress.kubernetes.io/upstream-vhost annotation might also come in handy.
It sets the Host header to the value you provide to avoid CSRF issues.
nginx:
ingress:
annotations:
cert-manager.io/cluster-issuer: "letsencrypt-prod"
enabled: true
hostname: fqdn
ingressClassName: "nginx"
tls: true
Snuba only supports a UTC timezone for Clickhouse. Please keep the initial value!
Following Helm Chart best practices the new version introduces some breaking changes, all configuration for external
resources moved to separate config branches: externalClickhouse, externalKafka, externalRedis, externalPostgresql.
Here is a mapping table of old values and new values:
| Before | After |
|---|---|
postgresql.postgresqlHost |
externalPostgresql.host |
postgresql.postgresqlPort |
externalPostgresql.port |
postgresql.postgresqlUsername |
externalPostgresql.username |
postgresql.postgresqlPassword |
externalPostgresql.password |
postgresql.postgresqlDatabase |
externalPostgresql.database |
postgresql.postgresSslMode |
externalPostgresql.sslMode |
redis.host |
externalRedis.host |
redis.port |
externalRedis.port |
redis.password |
externalRedis.password |
As this chart runs in helm 3 and also tries its best to follow on from the original Sentry chart. There are some steps that needs to be taken in order to correctly upgrade.
From the previous upgrade, make sure to get the following from your previous installation:
- Redis Password (If Redis auth was enabled)
- PostgreSQL Password
Both should be in the
secretsof your original 9.0 release. Make a note of both of these values.
Due to an issue where transferring from Helm 2 to 3. Statefulsets that use the following: heritage: {{ .Release.Service }} in the metadata field will error out with a Forbidden error during the upgrade. The only workaround is to delete the existing statefulsets (Don't worry, PVC will be retained):
kubectl delete --all sts -n <Sentry Namespace>
Once the statefulsets are deleted. Next steps is to convert the helm release from version 2 to 3 using the helm 3 plugin:
helm3 2to3 convert <Sentry Release Name>
Finally, it's just a case of upgrading and ensuring the correct params are used:
If Redis auth enabled:
helm upgrade -n <Sentry namespace> <Sentry Release> . --set redis.usePassword=true --set redis.password=<Redis Password> --set postgresql.postgresqlPassword=<Postgresql Password>
If Redis auth is disabled:
helm upgrade -n <Sentry namespace> <Sentry Release> . --set postgresql.postgresqlPassword=<Postgresql Password>
Please also follow the steps for Major version 3 to 4 migration
By default, PostgreSQL is installed as part of the chart. To use an external PostgreSQL server set postgresql.enabled to false and then set postgresql.postgresHost and postgresql.postgresqlPassword. The other options (postgresql.postgresqlDatabase, postgresql.postgresqlUsername and postgresql.postgresqlPort) may also want changing from their default values.
To avoid issues when upgrade this chart, provide postgresql.postgresqlPassword for subsequent upgrades. This is due to an issue in the PostgreSQL chart where password will be overwritten with randomly generated passwords otherwise. See https://github.com/helm/charts/tree/master/stable/postgresql#upgrade for more detail.
This chart is capable of mounting the sentry-data PV in the Sentry worker and cron pods. This feature is disabled by default, but is needed for some advanced features such as private sourcemaps.
You may enable mounting of the sentry-data PV across worker and cron pods by changing filestore.filesystem.persistence.persistentWorkers to true. If you plan on deploying Sentry containers across multiple nodes, you may need to change your PVC's access mode to ReadWriteMany and check that your PV supports mounting across multiple nodes.
- Lint in Pull requests
- Public availability through Github Pages
- Automatic deployment through Github Actions
- Symbolicator deployment
- Testing the chart in a production environment
- Improving the README