# Transport Layer Security (TLS) The Percona Operator for MongoDB uses Transport Layer Security (TLS) cryptographic protocol for the following types of communication: * Internal - communication between Percona Server for MongoDB instances in the cluster * External - communication between the client application and the cluster The internal certificate is also used as an authorization method. Certificates for TLS security can be generated in several ways. By default, the Operator generates long-term certificates automatically if there are no certificate secrets available. Other options are the following ones: * the Operator can use a specifically installed *cert-manager*, which will automatically generate and renew short-term TLS certificates, * certificates can be generated manually. You can also use pre-generated certificates available in the `deploy/ssl-secrets.yaml` file for test purposes, but we strongly recommend **avoiding their usage on any production system**! The following subsections explain how to configure TLS security with the Operator yourself, as well as how to temporarily disable it if needed. ## Install and use the *cert-manager* ### About the *cert-manager* The [cert-manager :octicons-link-external-16:](https://cert-manager.io/docs/) is a Kubernetes certificate management controller which widely used to automate the management and issuance of TLS certificates. It is community-driven, and open source. When you have already installed *cert-manager* and deploy the operator, the operator requests a certificate from the *cert-manager*. The *cert-manager* acts as a self-signed issuer and generates certificates. The Percona Operator self-signed issuer is local to the operator namespace. This self-signed issuer is created because Percona Server for MongoDB requires all certificates issued by the same CA (Certificate authority). Self-signed issuer allows you to deploy and use the Percona Operator without creating a cluster issuer separately. ### Installation of the *cert-manager* The steps to install the *cert-manager* are the following: * create a namespace, * disable resource validations on the cert-manager namespace, * install the cert-manager. The following commands perform all the needed actions: ``` {.bash data-prompt="$" } $ kubectl apply -f https://github.com/jetstack/cert-manager/releases/download/v{{ certmanagerrecommended }}/cert-manager.yaml --validate=false ``` After the installation, you can verify the *cert-manager* by running the following command: ``` {.bash data-prompt="$" } $ kubectl get pods -n cert-manager ``` The result should display the *cert-manager* and webhook active and running: ``` {.text .no-copy} NAME READY STATUS RESTARTS AGE cert-manager-7d59dd4888-tmjqq 1/1 Running 0 3m8s cert-manager-cainjector-85899d45d9-8ncw9 1/1 Running 0 3m8s cert-manager-webhook-84fcdcd5d-697k4 1/1 Running 0 3m8s ``` Once you create the database with the Operator, it will automatically trigger cert-manager to create certificates. Whenever you check certificates for expiration, you will find that they are valid and short-term. ## Generate certificates manually To generate certificates manually, follow these steps: 1. Provision a Certificate Authority (CA) to generate TLS certificates, 2. Generate a CA key and certificate file with the server details, 3. Create the server TLS certificates using the CA keys, certs, and server details. The set of commands generate certificates with the following attributes: * `Server-pem` - Certificate * `Server-key.pem` - the private key * `ca.pem` - Certificate Authority You should generate certificates twice: one set is for external communications, and another set is for internal ones. A secret created for the external use must be added to the `spec.secrets.ssl` key of the `deploy/cr.yaml` file. A certificate generated for internal communications must be added to the `spec.secrets.sslInternal` key of the `deploy/cr.yaml` file. !!! note If you only create the external certificate, then the Operator will not generate the internal one, but instead use certificate you have provided for both external and internal communications. Supposing that your cluster name is `my-cluster-name`, the instructions to generate certificates manually are as follows: ``` {.bash data-prompt="$" } $ CLUSTER_NAME=my-cluster-name $ NAMESPACE=default $ cat < ca-config.json { "signing": { "default": { "expiry": "87600h", "usages": ["signing", "key encipherment", "server auth", "client auth"] } } } EOF $ cat < ca.pem.old $ kubectl get secret/my-cluster-name-ssl-internal -o jsonpath='{.data.tls\.crt}' | base64 --decode > tls.pem.old $ kubectl get secret/my-cluster-name-ssl-internal -o jsonpath='{.data.tls\.key}' | base64 --decode > tls.key.old ``` 3. Combine new and current `ca.pem` into a `ca.pem.combined` file: ``` {.bash data-prompt="$" } $ cat ca.pem ca.pem.old >> ca.pem.combined ``` 4. Create a new Secrets object with *old* TLS certificate (`tls.pem.old`) and key (`tls.key.old`), but a *new combined* `ca.pem` (`ca.pem.combined`): ``` {.bash data-prompt="$" } $ kubectl delete secret/my-cluster-name-ssl-internal $ kubectl create secret generic my-cluster-name-ssl-internal --from-file=tls.crt=tls.pem.old --from-file=tls.key=tls.key.old --from-file=ca.crt=ca.pem.combined --type=kubernetes.io/tls ``` 5. The cluster will go through a rolling reconciliation, but it will do it without problems, as every node has old TLS certificate/key, and both new and old CA certificates. 6. If new TLS certificate and key weren’t generated on step 1, do that now. 7. Create a new Secrets object for the second time: use new TLS certificate (`server.pem` in the example) and its key (`server-key.pem`), and again the combined CA certificate (`ca.pem.combined`): ``` {.bash data-prompt="$" } $ kubectl delete secret/my-cluster-name-ssl-internal $ kubectl create secret generic my-cluster-name-ssl-internal --from-file=tls.crt=server.pem --from-file=tls.key=server-key.pem --from-file=ca.crt=ca.pem.combined --type=kubernetes.io/tls ``` 8. The cluster will go through a rolling reconciliation, but it will do it without problems, as every node already has a new CA certificate (as a part of the combined CA certificate), and can successfully allow joiners with new TLS certificate to join. Joiner node also has a combined CA certificate, so it can authenticate against older TLS certificate. 9. Create a final Secrets object: use new TLS certificate (`server.pmm`) and its key (`server-key.pem`), and just the new CA certificate (`ca.pem`): ``` {.bash data-prompt="$" } $ kubectl delete secret/my-cluster-name-ssl-internal $ kubectl create secret generic my-cluster-name-ssl-internal --from-file=tls.crt=server.pem --from-file=tls.key=server-key.pem --from-file=ca.crt=ca.pem --type=kubernetes.io/tls ``` 10. The cluster will go through a rolling reconciliation, but it will do it without problems: the old CA certificate is removed, and every node is already using new TLS certificate and no nodes rely on the old CA certificate any more. ### Update certificates with downtime If your certificates have been already expired (or if you continue to use the Operator version prior to 1.9.0), you should move through the *pause - update Secrets - unpause* route as follows. 1. Pause the cluster [in a standard way](pause.md), and make sure it has reached its paused state. 2. If cert-manager is used, delete issuer and TLS certificates: ``` {.bash data-prompt="$" } $ { kubectl delete issuer/my-cluster-name-psmdb-ca-issuer issuer/my-cluster-name-psmdb-issuer kubectl delete certificate/my-cluster-name-ssl certificate/my-cluster-name-ssl-internal } ``` 3. Delete Secrets to force the SSL reconciliation: ``` {.bash data-prompt="$" } $ kubectl delete secret/my-cluster-name-ssl secret/my-cluster-name-ssl-internal ``` 4. Check certificates to make sure reconciliation have succeeded. 5. Unpause the cluster [in a standard way](pause.md), and make sure it has reached its running state. ## Run Percona Server for MongoDB without TLS Omitting TLS is also possible, but we recommend that you run your cluster with the TLS protocol enabled. To disable TLS protocol (e.g. for demonstration purposes) set the `spec.allowUnsafeConfigurations` key to `true` in the `deploy/cr.yaml` file and and make sure that there are no certificate secrets available. This is the only condition under which the cluster will work without TLS. !!! warning Normally, the Operator prevents users from configuring a cluster with unsafe parameters (starting it with less than 3 replica set instances or without TLS, etc.), automatically changing such unsafe parameters to safe defaults. If you switch the cluster to the *unsafe configurations permissive mode*, you will not be able to switch it back by setting `spec.allowUnsafeConfigurations` key to `false`, the flag will be ignored.