Imply Enterprise on Google Kubernetes Engine

This document describes how to deploy and manage Imply Enterprise (formerly Imply Private) on Google Kubernetes Engine (GKE) using the enhanced installation mode. This solution improves setup and management processes for running Imply on Google Cloud Platform (GCP).

Google Cloud components and sample architecture

The following diagram shows the resources resulting from this deployment mode:

GCP sample setup

The architecture includes these components:

VPC - The network that resources will communicate over. An existing VPC can be used, or Imply will create one for you.
Public GKE cluster - The Kubernetes cluster in which all compute resources run. Several node pools are created for the cluster to use when deploying Druid deployments. An existing cluster cannot be used; the cluster will be created by Imply.
Cloud SQL instance - The data store used as the default metadata store for Druid deployments, as well as for the Imply control plane. An existing Cloud SQL instance can be used, or Imply will create one for you.
Cloud Storage bucket - The storage bucket used as the default Deep Storage for all Druid deployments. An existing cloud storage bucket can be used, or Imply will create one for you.
Load balancer - The load balancer that routes requests for the Imply control plane.

Google Cloud machine types

This section lists the Google Cloud machine types available for Imply cluster server instances in GCP, by node type. For more information about machine sizing, see planning.

Master nodes

n2-standard-2: 2 vCPU / 8 GB RAM
n2-standard-8: 8 vCPU / 32 GB RAM
n2-standard-16: 16 vCPU / 64 GB RAM

Query nodes

c2-standard-4: 4 vCPU / 16 GB RAM
c2-standard-8: 8 vCPU / 32 GB RAM
c2-standard-16: 16 vCPU / 64 GB RAM
c2-standard-30: 30 vCPU / 120 GB RAM

Data nodes

n2-highmem-4: 4 vCPU / 32 GB RAM / 750 GB disk
n2-highmem-8: 8 vCPU / 64 GB RAM / 1500 GB disk
n2-highmem-16: 16 vCPU / 128 GB RAM / 3000 GB disk
n1-highmem-32: 32 vCPU / 208 GB RAM / 6000 GB disk
n2-highmem-48: 48 vCPU / 384 GB RAM / 9000 GB disk
n2-highmem-80: 80 vCPU / 640 GB RAM / 3000 GB disk

Note that for each machine type you choose, the installer creates two node pools: the destination node pool and a second node pool with -pool-2 appended to the pool name. The second node pool is intended for use when upgrading the GKE nodes. For more information, see Updating node pools.

Requirements

Before proceeding please ensure the following requirements are fulfilled.

GCP resource requirements

You will need the following Google Cloud resources:

A Google Cloud account
A Google Cloud project
A GCP user with sufficient permissions, with these options:
- Option 1. User with Owner Basic Role
- Option 2. A minimally privileged user with all permissions of the following GCP predefined roles:
  - Cloud SQL Admin
  - Compute Network Admin
  - Kubernetes Engine Admin
  - Storage Admin
  - DNS Administrator (if configuring ingress)
A Monitoring workspace containing the respective project. Allow Google to create the workspace with the same name as your project.

Enable the following APIs before starting:

Compute Engine API
Kubernetes Engine API
Service Networking API
Cloud DNS API (if configuring ingress)

To enable these APIs, navigate to the APIs & Services section of the Google Cloud Web Console, search for them, and enable them.

Tool dependencies

The setup script depends on the following tools:

These tools are included by default in the Google Cloud shell, and for that reason we recommend running the setup script from there.

Internet access

The Imply Manager installer requires Internet connectivity to complete the installation. After installation, the Imply deployment will continue to require Internet access, for example, to download assets from https://hub.docker.com and https://static.imply.io.

If your policies prevent such access, we recommend using Imply Manager on Kubernetes instead.

Data encryption

By default, data for the following GCP resources are encrypted using the default keys for the resources:

Google Kubernetes Engine (GKE) Node pools, for boot disk
Google Cloud Storage (GCS) Buckets
Cloud SQL Instances

Alternatively, you can configure encryption based on your own Customer Managed Encryption Keys.

To use this feature, you need to enable key management service (KMS) in the GCP project, create a key ring, and create the key. See the KMS quickstart for more information.

You also need to configure Service Accounts for Cloud SQL, GCS, and GKE. For more information, see the following documentation:

With KMS enabled for the project, you will be prompted to supply the key information by the setup script, as described in the following section.

Creating an Imply deployment

The Imply GKE setup script guides you through the deployment process.

Download the setup script

Running the setup script in the Google Cloud shell is recommended, since the shell environment contains all dependencies used by the script, by default.

To download the setup script from the Google Cloud shell, run the following command:

curl -O https://static.imply.io/onprem/setup

Authorize with Google Cloud

Authorize gcloud to access the Cloud Platform with Google user credentials. To do so, run the command gcloud auth login and follow the prompts shown. Ensure that you are authorized as a user with sufficient permissions.

If you are running the setup script from the Google Cloud shell, you are prompted to authorize cloud shell when you start the script.

Start the script

To start the script, run sh setup. You will see output similar to the following:

Welcome to the Imply GCP Installer. (build 50aa6b9)

This installer will guide you through setting up the required Google
Cloud Platform resources for running a managed Imply install in GCP
on Kubernetes.

Select project

Next, select the Google project to work in.

1) golden-bonbon-283506
2) pure-episode-234323
Select the Google Project ID [pure-episode-234323]:

Enter the number corresponding to the project that you want to use and press enter, or press enter with no value to use the default value.

Enter a custom encryption key

If KMS is enabled for the project, you are prompted to choose keys for the project, as follows:

KMS is enabled for your project. You can choose
to use the default keys setup for [the project]
or use Customer Managed Encryption Keys.

Note: When using CMEK, the project's service accounts
      have to be authorized for using KMS keys before
      running setup. Refer to GCP's documentation for
      more information.

The following prompts may appear:

If asked whether you want to use the default key for the resources, choose yes or no to specify another key.
If you choose no, you are prompted to select a key ring in the project and key.
If you are updating an existing project and it uses CMEK, you can choose whether you want to use a ring that already exists in the project. Choose yes or no to specify another key.

Enter license

Next you are prompted to enter a license key.

Enter your Imply License (blank for trial):

Enter your Imply license here and press enter, or press enter with no value to use a trial license. The trial license allows you to use the product for 30 days. If using a trial license, you can update the deployment later with a replacement license if needed.

Select region

Next you are prompted for the deployment region. This is the region that Imply will be deployed to.

1) asia-east1
2) asia-east2
3) asia-northeast1
4) asia-southeast1
5) australia-southeast1
6) europe-north1
7) europe-west1
8) europe-west2
9) europe-west3
10) europe-west4
11) northamerica-northeast1
12) southamerica-east1
13) us-central1
14) us-east1
15) us-east4
16) us-west1
Select a region [us-central1]:

Enter the number corresponding to the region that you want to use for the deployment, and press enter.

Select HA or non-HA deployment

Next choose whether to deploy resources across multiple availability zones in the region. When deployed across multiple availability zones, Imply is more resilient to failure in any one zone.

Running in multiple zones will create all resources with
redundancies in multiple GCP zones. Resources such as optionally
created MySQL and the Kubernetes cluster will have redundancy
configured across these zones. This is recommended in production.
Run in multiple zones? [yes]:

Press enter to accept the default of using multiple zones, or enter no for a single zone deployment and press enter. We recommend using multiple zones.

Select network

Next you are asked whether to create a new VPC or to use an existing VPC. This is the network that the Imply deployment will use when deploying resources.

Creating a separate VPC is recommended but if you require
the Imply cluster to run in an existing VPC to be able to
access resources that cannot otherwise be access you can
select an existing VPC.
Note: When using an existing VPC please ensure that private
service networking is enabled to allow the Imply cluster
to communicate with GCP services.
Create a new VPC [yes]:
CIDR Range for the Imply Cluster to use [10.128.0.0/16]:

Press enter to take the default of creating a new VPC with the default CIDR range specified, or specify your own. If you want to use an existing VPC, enter no and follow the prompts. Ensure that the CIDR range that you specify in this case is available for use.

Select deep storage bucket

Choose whether to create a Deep Storage bucket or to use an existing bucket. This bucket is used as the default deep storage for all Imply deployments.

Choose a GCS bucket to store your data (deep storage). If you
type 'New' (or leave the input blank) a new bucket will be
created for you. If you wish to use an existing bucket, use the
form: gs://my-bucket-1.

Note: If a bucket is created for you, it will be removed on uninstall.
      ALL DATA WILL BE REMOVED AS WELL.
Existing GCS bucket or New [New]:

Press enter to take the default of creating a new storage bucket, or enter no to use an existing bucket, and follow the prompts to give details to identify the bucket.

Select metadata store

Choose whether to create a Cloud SQL store for storing metadata about your Imply deployment or to use an existing database.

Choose a MySQL database to serve as the metadata store. It is
recommended that you create a new database for use by Imply.
If you require Imply to use an existing MySQL database, you
may enter those details here instead.

Note: Your SQL connection information may be stored in plaintext
      in a GCS bucket in your account by Terraform.

Note: If a database is created for you, it will be removed on uninstall.
      ALL DATA WILL BE REMOVED AS WELL.

Press enter to take the default of creating a new Cloud SQL instance, or enter no to use an existing database, and follow the prompts to give details to identify the database.

About Cloud SQL sizing

Imply enables disk autoscaling for this metadata store instance. Disk autoscaling is a GCP feature described in Automatic storage increase of the Google Cloud documentation.

The Imply configuration does not itself impose a limit on the disk size. However, in practice, the disk size is limited by the size of the default instance type used by Imply for the metadata store, which is db-n1-standard-2.

The disk size automatically starts scaling up in 6 GB increments initially, and scales up linearly to 25 GB increments for disk sizes above 500 GB. The increment sizes determine when scaling is needed, that is, when the amount of free space reaches 6 GB or less, the autoscaling system scales up disk by 6 GB.

Note that storage does not scale down, other than by manually changing the underlying instances.

Select ingress

Choose whether to set up GKE Ingress for the Imply cluster, allowing secure connections to the Imply Manager. This requires that you have at least one Cloud DNS managed zone configured.

Ingress can be automatically configured using Cloud DNS
and Google Managed Certificates to allow secure connections
to the Imply Manager. This feature requires that a Cloud DNS
managed zone already exist. For more information on setting
up a Cloud DNS zone see: https://cloud.google.com/dns/docs/zones
Note: Security Policy changes may take a few minutes to be reflected and
access to the Imply Manager may be interrupted. It will not affect
any running clusters.
Automatically setup ingress? [no]: yes
Determining available Cloud DNS Managed Zones
1) gcp-imply-io
Select a zone: 1

If you do not have Cloud DNS managed zones configured, press enter to take the default selection, no, as this feature is not supported without Cloud DNS managed zones. Otherwise, to use this feature, enter yes and enter the number corresponding to the Cloud DNS zone that you want to use for the deployment, and press enter. This results in the control plane being hosted in a subdomain of the Cloud DNS zone that you selected.

Attach CloudArmor policy

Next choose whether to have a CloudArmor Policy applied to the services in the cluster. If you select yes, you are shown a list of existing CloudArmor policies. Select one of them to apply.

A Cloud Armor Security Policy can be attached to the Ingress
that will be created. This feature requires that a Security Policy
already exist. For more information on Security Policies see:
https://cloud.google.com/armor/docs/configure-security-policies
Configure a Security Policy? [no]: yes
Determining available Security Policies
1) imply-sp
Select a policy: 1

Confirm details

Next review and confirm the details of the deployment. If correct, press enter and proceed with the deployment. Otherwise, enter no to restart the script.

Wait for deployment to complete

The deployment will take around 30 minutes. Wait for it to finish deploying resources. When deployment finishes, you will receive a message similar to the following.

helm_release.manager: Creation complete after 3m3s [id=imply]

Apply complete! Resources: 32 added, 0 changed, 0 destroyed

Outputs:

bucket =
cidr = 10.128.0.0/16
dns_host = imply.gcp.imply.io
dns_zone = gcp-imply-io
gke_id = imply-gke
license = 
project_id = pure-episode-234323
region = us-central1
sql_endpoint =
sql_password =
sql_username =
vpc =
zone =

You can access your Imply Manager by running:
  gcloud container clusters get-credentials "imply-doc-test-1-gke" --region us-central1 --project engineering-docs-321321
to get the credentials for the Kubernetes cluster and then:
  kubectl port-forward svc/imply-doc-test-1-manager-int 9097
Then visit http://localhost:9097

The details include instructions for how to access your deployment.

Start using Imply

The setup script presents you with the URL for accessing the Imply Manager, from where you can add users, access the data loader, and access Pivot.

Create an Imply Manager admin user

When you run the Imply Manager for the first time, it will prompt you to enter the initial admin user credentials. Provide account settings as prompted. When you finish, the Imply Manager home page appears, where you can add additional users. See Imply Manager users for how to add users.

Manage a cluster

To start using Imply, create a cluster by clicking New cluster. In the New cluster page, choose the instance types for the cluster.

GCP service account details

Click Create cluster when finished. The Servers page appears. It takes a few minutes for the servers in the cluster to appear on the page. Once they do, their status reports "waiting" until deployment completes, at which point their status should be OK. You can then connect to the cluster, as follows.

See cluster management documentation for information on creating and administering Imply clusters. Administration tasks you perform from the Imply Manager include stopping the cluster, starting the cluster, scaling it up, and adding data tiers.

Connect to a cluster

Once the servers finish starting and join the cluster, you can access the data loader and Pivot UI as follows:

Click Overview from the left menu for the cluster.
Click Open under cluster actions. Pivot appears.
Click Load data to access the data loader for the cluster.

If you are new to Imply, follow the Quickstart to learn how to get started quickly with sample data. If you want to load your own data from Google Cloud Storage, see the next section.

Load data

The following steps take you through an example of loading data into your cluster from a Google Cloud Storage (GCS) bucket. Your ingestion source types are not limited to GCS, but GCS is a likely data source type for an Imply Enterprise on Google Kubernetes Engine (GKE) deployment. For other ingestion types, see Druid ingestion documentation.

By default, the service accounts on the Druid service pods only have data read and write access to the deep storage bucket. To ingest data from another storage bucket, you need to give the agent service account permissions needed to access that bucket. See Google's IAM documentation for details. The agent service account is named imply-default-asa.

Give service account permission to data bucket

Give the service account access to the resource from which you would like to ingest data. In this example, we will be ingesting data from Google Storage Bucket imply-doc-demo-ingest.

Use the Google Cloud Web Console to navigate to your service account, we'll need to look up the email address of the agent service account. Find the agent service account, which is named imply-default-asa, and copy the email address for the account.

GCP service account details

In the Google Cloud web console, navigate to the bucket details for the bucket that you would like to ingest data from, and click the permissions tab.

GCP bucket permissions

Add the service account with role storage admin using the service account email.

GCP bucket add service account permission

Now the service account has the permission needed to read and write to the bucket.

Ingest data using Druid console

Navigate to the Druid Web Console, click new ingestion spec and choose the Google Cloud Storage ingestion type.

GCP choose Google Cloud Storage ingestion type

Follow the on-screen instructions for providing the path to the data that you want to ingest, and any filtering or aggregation options that you'd like to apply.

You can also download Wikipedia sample data to try. Upload the file to your Google Cloud bucket and provide the path to the file inside the Druid console.

After the data is ingested successfully, you will see a corresponding ingestion task listed in the Ingestion tab with status, SUCCESS.

GCP successful ingestion task

Once Druid loads all of the segments data, the corresponding datasource will show as Fully available under the Datasources tab:

GCP segments fully loaded

Query the data

The datasource is now ready to be queried. See the query tutorial documentation for details.

Updating an Imply deployment

To update an existing Imply deployment, run the setup script as described in Creating an Imply Deployment. Give the selection for the deployment that you'd like to update. When asked if you'd like to update, enter yes.

Select the Google Project ID [zachsherman]:
1) New
2) default
Select existing deployment to manage or New to create one [New]: 2
An existing deployment was found, update? [yes]:

You will be given the details of the deployment, and prompted whether the details are correct. If you are using an updated version of the script, and don't want to make any explicit changes to your deployment, enter yes. If you would like to make explicit changes to your deployment enter no.

License:
Project ID:  zachsherman
Region:      us-central1
Zone:        [Multizone]
VPC:         [Created]
CIDR:        10.128.0.0/16
Bucket:      [Created]
MySQL:       [Created]
Ingress:     no
Is the above information correct [yes]:

If you entered no, from there, you will be permitted to modify the deployment. Follow the steps as described earlier in this document.

Updating node pools

You may occasionally need to update node pool software underlying the Imply deployment, say to update the Google Kubernetes Engine version. With a single node pool, this would mean down time for the cluster.

To allow for non-interrupting upgrades, the installer creates a second node pool for every node machine type created. The name of these machine pools are appended with -pool-2. These pool-2 nodes are unused until needed.

When you need to update the node pools, follow these steps:

Update the pool-2 node pools following the instructions on Upgrading nodes in the Google Kubernetes Engine (GKE) documentation.
Deploy the cluster to the second set in the Imply Manager by clicking the Deploy to pool 2 checkbox.

The workload is automatically transferred to pool-2. When you subsequently need to update the node pools again, you can update the first node pools, and then unselect the checkbox to return operation to the first set of node pools.

Deleting an Imply deployment

To delete an existing Imply deployment, run the setup script as described in Creating an Imply Deployment. Give the selection for the deployment that you'd like to update, and when asked if you'd like to update, enter no. You will then be given the option to uninstall the deployment.

1) zachsherman
2) zachsherman-test
Select the Google Project ID [zachsherman]: 
1) New
2) default
Select existing deployment to manage or New to create one [New]: 2
An existing deployment was found, update? [yes]: no
Uninstall deployment? [no]: yes

Follow the prompts from here.

Advanced topics

The following section describes advanced use cases of the deployment.

Direct-access Pivot

Direct-access Pivot is supported in GKE but has some differences in how it works in Imply Hybrid (formerly Imply Cloud). All configuration options are the same, however, so reviewing the Direct-access Pivot docs is a good first step.

The main difference is in how the URL for Pivot access field interacts with the Pivot configuration and underlying Kubernetes resources. For example, setting the field to a name such as pivot.company.com at installation time will result in an Ingress that points to Pivot on port 80 of the Nginx Ingress Controller. This means that once DNS has been configured to point to the public or private load balancers, you can use the name provided.

WARNING!
Using the same domain as configured to access your Imply Manager without a URL path could render your Manager inaccessible.

If your URL for Pivot access field contains a path as well (e.g., imply.mycompany.com/10257016-8bfd-433a-9e98-f62dd23bce1e/pivot) then Pivot will be automatically configured to use the serverRoot from the path configured. This means that if you had configured an Ingress during installation with a domain name, you could use a path to enable routing with the same domain and certificate that have already been configured. The easiest way to configure this would be to copy the public load balancer path and append /pivot to the end of it, as in the example above.

If the field is configured with a different domain than the Manager, it will still work, but you will need to configure the domain and TLS certificate separately. To configure this from the GCP Console, select Network Service, then Load balancing. You can choose whether to make it available publicly on the internet or only within your VPC. From there you will want to configure the Backend configuration. If you have many services available, it may be hard to determine the correct one to select. To figure out the correct service in a new GCP Console tab, select Kubernetes Engine, Services & Ingresses, and then select the Imply cluster named imply-default-gke, by default. Then select the ingress-nginx-controller. On the Details page under Annotations, you should see it displayed as the Network Endpoint Group name. It should look something like k8s1-31d25674-default-ingress-nginx-controller-80-c05a40bc, which you can select in the Backend configuration screen.

Once that is completed, select the Frontend configuration. Here you can configure options like reusing an existing TLS certificate or having one generated, as well as SSL policies. Review the Google Cloud Load Balancing documentation for a full list of configuration options.

Note that when accessing Direct access Pivot any configured CloudArmor policy will automatically apply as well.

Multiple deployments

The setup script can be used to create multiple Imply deployments. If you do not currently have any deployments, you will not be prompted for a deployment name, and instead a name of default will be automatically given for your first deployment. If you have at least one deployment, you will be prompted to select which deployment to use or can create a new deployment at this time.

Select the Google Project ID [pure-episode-234323]:
1) New
2) default
Select existing deployment to manage or New to create one [New]:

If using a non-default deployment, the name of the agent service account will be imply-{DEPLOYMENT-NAME}-asa.

Common issues

The following section goes over some common issues that you may face in the field and how to resolve them.

Deployment terraform state locked

The setup script uses terraform to create, update, or delete resources in Google Cloud for the Imply deployment. If an error occurs while terraform is in the process of modifying resources, such as the user's terminal window being closed, the next time the setup script is run, terraform will complain that it was not able to acquire a lock on the state for the deployment. Such a message would look something like the following:

GCP terraform lock issue

To resolve this situation, ensure that no other operations are being performed on the deployment, and unlock the state file using terraform force-unlock. If this fails, delete the lock file from the Google storage bucket noted in the message.

Cluster create or update fails to scale up node pool

You may encounter an issue when creating or updating an existing cluster, where a failure occurs because a particular node pool fails to scale to the required number of nodes, as shown below:

GCP scale-out failure manager

This can be caused by quota limits in the account being hit. To fix this issue, you will need to increase the corresponding quota.