Ovais TariqThis blog outlines the deployment of Tigris on an Google's Kubernetes Engine (GKE) Autopilot instance.
The installation will use recommended settings for redundancy, allocating more resources than a simple laptop based installation would. For more information on the laptop based installation please consult our previous blog!
If you would rather watch a video, check out the deployment in action on YouTube:
Requirements
Below are the requirements for the installation box and the target Kubernetes environment.
The list of items required:
- Helm
- Google Cloud SDK
- git and tigris-deploy repository
- GKE cluster with sufficient quotas
Installation Host
We will require Helm to perform the installation. It is assumed that the installation host already has access to the deployment target GKE cluster.
The version of helm used in this blog was:
❯ helm version
version.BuildInfo{Version:"v3.10.1", GitCommit:"9f88ccb6aee40b9a0535fcc7efea6055e1ef72c9", GitTreeState:"clean", GoVersion:"go1.19.2"}
To interface with the GKE cluster using kubectl conveniently, you may want to install the GKE plugin. You can install it with this command:
❯ gcloud components install gke-gcloud-auth-plugin
GKE
Fortunately, GKE Autopilot clusters automatically comes with a set of controllers installed. The list includes GKE Ingress that enables the creation of external load balancers for Ingress resources and controllers that manage other aspects of GCP, such as persistent disks.
One of the challenges of ensuring a successful deployment in GCP is to manage quotas efficiently. You will want to ensure quotas allow for sufficient CPU and SSD storage allocation.
Using the defaults of the Helm Chart, the following quotas proved to be sufficient:
Deployment
The installation deploys the following components:
- Kubernetes Operator for FoundationDB
- FoundationDB
- Tigris Search (TypeSense)
- Tigris Server
You can install the components individually or together, using the encompassing tigris-stack Helm Chart. Below I'm going to use this Chart to install Tigris.
Prepare For Deployment
Next, check out the deploy script repository:
❯ git clone git@github.com:tigrisdata/tigris-deploy.git
Cloning into 'tigris-deploy'...
remote: Enumerating objects: 177, done.
remote: Counting objects: 100% (97/97), done.
remote: Compressing objects: 100% (60/60), done.
remote: Total 177 (delta 43), reused 68 (delta 34), pack-reused 80
Receiving objects: 100% (177/177), 87.68 KiB | 568.00 KiB/s, done.
Resolving deltas: 100% (63/63), done.
Navigate to the folder which contains the helm chart of tigris-stack:
❯ cd tigris-deploy/helm/tigris-stack
Deploy Tigris Stack
To ensure there is initial quorum for Tigris Search, we should deploy it initially with a single replica.
❯ helm install tigris-stack . --set tigris-search.replicas=1
W1103 11:56:22.823655 12264 gcp.go:119] WARNING: the gcp auth plugin is deprecated in v1.22+, unavailable in v1.26+; use gcloud instead.
To learn more, consult https://cloud.google.com/blog/products/containers-kubernetes/kubectl-auth-changes-in-gke
W1103 11:56:30.072806 12264 warnings.go:70] Autopilot increased resource requests for Deployment default/tigris-server to meet requirements. See http://g.co/gke/autopilot-resources.
W1103 11:56:30.089432 12264 warnings.go:70] Autopilot increased resource requests for Deployment default/tigris-stack-fdb-operator to meet requirements. See http://g.co/gke/autopilot-resources.
W1103 11:56:30.232424 12264 warnings.go:70] Autopilot set default resource requests on StatefulSet default/tigris-search for container tigris-ts-node-mgr, as resource requests were not specified, and adjusted resource requests to meet requirements. See http://g.co/gke/autopilot-defaults and http://g.co/gke/autopilot-resources.
NAME: tigris-stack
LAST DEPLOYED: Thu Nov 3 11:56:25 2022
NAMESPACE: default
STATUS: deployed
REVISION: 1
TEST SUITE: None
At this point your cluster will likely only have a few nodes:
❯ kubectl get nodes
W1103 11:57:04.068108 12352 gcp.go:119] WARNING: the gcp auth plugin is deprecated in v1.22+, unavailable in v1.26+; use gcloud instead.
To learn more, consult https://cloud.google.com/blog/products/containers-kubernetes/kubectl-auth-changes-in-gke
NAME STATUS ROLES AGE VERSION
gk3-doc-default-pool-ddd321b8-4v8x Ready <none> 42h v1.23.8-gke.1900
gk3-doc-default-pool-e88cea62-9b77 Ready <none> 42h v1.23.8-gke.1900
The pods will be in the Pending state and trigger pod scale-ups:
❯ kubectl get pods
W1103 11:56:43.749022 12327 gcp.go:119] WARNING: the gcp auth plugin is deprecated in v1.22+, unavailable in v1.26+; use gcloud instead.
To learn more, consult https://cloud.google.com/blog/products/containers-kubernetes/kubectl-auth-changes-in-gke
NAME READY STATUS RESTARTS AGE
tigris-search-0 0/2 Pending 0 14s
tigris-server-8646cb4b7b-fz6h4 0/1 Pending 0 14s
tigris-server-8646cb4b7b-hmxj9 0/1 Pending 0 14s
tigris-server-8646cb4b7b-qsjw7 0/1 Pending 0 14s
tigris-stack-fdb-operator-8fd845b9-wb4r5 0/1 Pending 0 14s
❯ kubectl describe pod tigris-search-0 | tail
W1103 11:58:18.395905 12695 gcp.go:119] WARNING: the gcp auth plugin is deprecated in v1.22+, unavailable in v1.26+; use gcloud instead.
To learn more, consult https://cloud.google.com/blog/products/containers-kubernetes/kubectl-auth-changes-in-gke
Node-Selectors: <none>
Tolerations: kubernetes.io/arch=amd64:NoSchedule
node.kubernetes.io/not-ready:NoExecute op=Exists for 300s
node.kubernetes.io/unreachable:NoExecute op=Exists for 300s
Events:
Type Reason Age From Message
---- ------ ---- ---- -------
Warning FailedScheduling 108s gke.io/optimize-utilization-scheduler 0/2 nodes are available: 2 Insufficient cpu, 2 Insufficient memory.
Warning FailedScheduling 38s gke.io/optimize-utilization-scheduler 0/3 nodes are available: 1 node(s) had taint {node.kubernetes.io/not-ready: }, that the pod didn't tolerate, 2 Insufficient cpu, 2 Insufficient memory.
Normal TriggeredScaleUp 26s cluster-autoscaler
Tigris will restart a few times before it changes state to Running. This is due to the unavailability of FoundationDB, the key-value store Tigris uses for persistence.
As you can see below, fdb is still in a Pending state when the tigris-server Pods are already up:
❯ kubectl get pods
W1103 12:05:30.762386 14893 gcp.go:119] WARNING: the gcp auth plugin is deprecated in v1.22+, unavailable in v1.26+; use gcloud instead.
To learn more, consult https://cloud.google.com/blog/products/containers-kubernetes/kubectl-auth-changes-in-gke
NAME READY STATUS RESTARTS AGE
fdb-cluster-log-1 0/2 Pending 0 43s
fdb-cluster-log-2 0/2 Pending 0 43s
fdb-cluster-log-3 0/2 Pending 0 42s
fdb-cluster-log-4 0/2 Pending 0 42s
fdb-cluster-log-5 0/2 Pending 0 42s
fdb-cluster-stateless-1 0/2 Pending 0 43s
fdb-cluster-stateless-10 0/2 Pending 0 43s
fdb-cluster-stateless-2 0/2 Pending 0 43s
fdb-cluster-stateless-3 0/2 Pending 0 43s
fdb-cluster-stateless-4 0/2 Pending 0 43s
fdb-cluster-stateless-5 0/2 Pending 0 43s
fdb-cluster-stateless-6 0/2 Pending 0 43s
fdb-cluster-stateless-7 0/2 Pending 0 43s
fdb-cluster-stateless-8 0/2 Pending 0 43s
fdb-cluster-stateless-9 0/2 Pending 0 43s
fdb-cluster-storage-1 0/2 Pending 0 43s
fdb-cluster-storage-2 0/2 Pending 0 43s
fdb-cluster-storage-3 0/2 Pending 0 43s
fdb-cluster-storage-4 0/2 Pending 0 43s
fdb-cluster-storage-5 0/2 Pending 0 43s
tigris-search-0 2/2 Running 1 (5m49s ago) 9m1s
tigris-server-8646cb4b7b-fz6h4 0/1 ContainerCreating 0 9m1s
tigris-server-8646cb4b7b-hmxj9 0/1 CrashLoopBackOff 1 (6s ago) 9m1s
tigris-server-8646cb4b7b-qsjw7 0/1 CrashLoopBackOff 2 (7s ago) 9m1s
tigris-stack-fdb-operator-8fd845b9-zgr4t 1/1 Running 0 5m55s
:info: You can improve the deployment sequence by using more sophisticated deployment methods, such as Synchronization Waves in ArgoCD!
Give Autopilot enough time to scale up nodes for the deployment. FoundationDB will likely trigger a separate scale-up event on its own.
❯ kubectl get nodes
W1103 12:09:59.375610 16639 gcp.go:119] WARNING: the gcp auth plugin is deprecated in v1.22+, unavailable in v1.26+; use gcloud instead.
To learn more, consult https://cloud.google.com/blog/products/containers-kubernetes/kubectl-auth-changes-in-gke
NAME STATUS ROLES AGE VERSION
gk3-doc-default-pool-ddd321b8-4v8x Ready <none> 42h v1.23.8-gke.1900
gk3-doc-default-pool-e88cea62-9b77 Ready <none> 42h v1.23.8-gke.1900
gk3-doc-nap-10cyk06a-854c84a8-4qss Ready <none> 4m23s v1.23.8-gke.1900
gk3-doc-nap-10cyk06a-854c84a8-6fd2 Ready <none> 4m21s v1.23.8-gke.1900
gk3-doc-nap-10cyk06a-854c84a8-m6hp Ready <none> 4m23s v1.23.8-gke.1900
gk3-doc-nap-10cyk06a-854c84a8-p8zq Ready <none> 4m21s v1.23.8-gke.1900
gk3-doc-nap-10cyk06a-854c84a8-r744 Ready <none> 4m22s v1.23.8-gke.1900
gk3-doc-nap-10cyk06a-854c84a8-xj5b Ready <none> 4m20s v1.23.8-gke.1900
gk3-doc-nap-10cyk06a-9f9e9a3f-4m2r Ready <none> 4m18s v1.23.8-gke.1900
gk3-doc-nap-10cyk06a-9f9e9a3f-d6nm Ready <none> 4m18s v1.23.8-gke.1900
gk3-doc-nap-10cyk06a-9f9e9a3f-ggxv Ready <none> 4m17s v1.23.8-gke.1900
gk3-doc-nap-10cyk06a-9f9e9a3f-lfwl Ready <none> 4m18s v1.23.8-gke.1900
gk3-doc-nap-10cyk06a-9f9e9a3f-s456 Ready <none> 4m18s v1.23.8-gke.1900
gk3-doc-nap-10cyk06a-9f9e9a3f-slg8 Ready <none> 4m19s v1.23.8-gke.1900
gk3-doc-nap-10cyk06a-9f9e9a3f-vg27 Ready <none> 11m v1.23.8-gke.1900
gk3-doc-nap-10cyk06a-9f9e9a3f-xf4k Ready <none> 4m18s v1.23.8-gke.1900
gk3-doc-nap-10cyk06a-9f9e9a3f-xptm Ready <none> 4m18s v1.23.8-gke.1900
gk3-doc-nap-10cyk06a-c0284c87-5hpx Ready <none> 4m13s v1.23.8-gke.1900
gk3-doc-nap-10cyk06a-c0284c87-96c2 Ready <none> 4m12s v1.23.8-gke.1900
gk3-doc-nap-10cyk06a-c0284c87-c7h8 Ready <none> 4m13s v1.23.8-gke.1900
gk3-doc-nap-10cyk06a-c0284c87-klm4 Ready <none> 4m12s v1.23.8-gke.1900
gk3-doc-nap-10cyk06a-c0284c87-mrqp Ready <none> 4m12s v1.23.8-gke.1900
gk3-doc-nap-10cyk06a-c0284c87-wwj2 Ready <none> 4m12s v1.23.8-gke.1900
gk3-doc-nap-qm2jb0jm-1393ada1-bgwt Ready <none> 11m v1.23.8-gke.1900
gk3-doc-nap-qm2jb0jm-6d70fd3a-pxdr Ready <none> 12m v1.23.8-gke.1900
Following the scale up of the nodes, the services to slowly also come up. As it is waiting for foundational services to start,
However, after about 15 minutes the Pods should become available:
❯ kubectl get pods
W1103 12:10:45.077224 16929 gcp.go:119] WARNING: the gcp auth plugin is deprecated in v1.22+, unavailable in v1.26+; use gcloud instead.
To learn more, consult https://cloud.google.com/blog/products/containers-kubernetes/kubectl-auth-changes-in-gke
NAME READY STATUS RESTARTS AGE
fdb-cluster-log-1 2/2 Running 0 5m57s
fdb-cluster-log-2 2/2 Running 0 5m57s
fdb-cluster-log-3 2/2 Running 0 5m56s
fdb-cluster-log-4 2/2 Running 0 5m56s
fdb-cluster-log-5 2/2 Running 0 5m56s
fdb-cluster-stateless-1 2/2 Running 0 5m57s
fdb-cluster-stateless-10 2/2 Running 0 5m57s
fdb-cluster-stateless-2 2/2 Running 0 5m57s
fdb-cluster-stateless-3 2/2 Running 0 5m57s
fdb-cluster-stateless-4 2/2 Running 0 5m57s
fdb-cluster-stateless-5 2/2 Running 0 5m57s
fdb-cluster-stateless-6 2/2 Running 0 5m57s
fdb-cluster-stateless-7 2/2 Running 0 5m57s
fdb-cluster-stateless-8 2/2 Running 0 5m57s
fdb-cluster-stateless-9 2/2 Running 0 5m57s
fdb-cluster-storage-1 2/2 Running 0 5m57s
fdb-cluster-storage-2 2/2 Running 0 5m57s
fdb-cluster-storage-3 2/2 Running 0 5m57s
fdb-cluster-storage-4 2/2 Running 0 5m57s
fdb-cluster-storage-5 2/2 Running 0 5m57s
tigris-search-0 2/2 Running 1 (11m ago) 14m
tigris-server-8646cb4b7b-95lcf 1/1 Running 0 2m37s
tigris-server-8646cb4b7b-gff64 1/1 Running 2 (3m12s ago) 3m23s
tigris-server-8646cb4b7b-hmxj9 1/1 Running 5 (3m59s ago) 14m
tigris-stack-fdb-operator-8fd845b9-zgr4t 1/1 Running 0 11m
That's it, your Tigris deployment should be now on its way coming up!
Validate Deployment
This time we are going to validate Tigris Server using the Tigris CLI, using a small linux Pod that was deployed in the same namespace as the Tigris Stack.
First we need to install the CLI:
$ curl -sSL https://tigris.dev/cli-linux | sudo tar -xz -C /usr/local/bin
...
$ ls -la /usr/local/bin/tigris
-rwxr-xr-x 1 1001 121 17264640 Nov 3 07:21 /usr/local/bin/tigris
Set TIGRIS_URL to point at the Service endpoint of tigris-server:
$ export TIGRIS_URL=http://tigris-http:80
After that see if you can interact with the Tigris database using the tigris utility:
$ tigris quota limits
{
"ReadUnits": 100,
"WriteUnits": 25,
"StorageSize": 104857600
}
$ tigris server info
{
"server_version": "v1.0.0-beta.17"
}
$ tigris server version
tigris server version at http://tigris-http:80 is v1.0.0-beta.17
$ tigris create database robert
$ tigris list databases
robert
Preparing For Production
Scaling Search Out
To ensure Search is also redundant, once the deployment has progressed past transient state, Tigris Search should be scaled up to multiple replicas. In order to maintain quorum, the number of replicas should be set to an odd number, at a minimum of 3.
Below command will increase the number of Tigris Search replicas to 5 which should be a sufficiently large number of replicas for an initial Production deployment:
❯ helm upgrade tigris-stack . --set tigris-search.replicas=5
W1103 18:12:06.790278 82440 gcp.go:119] WARNING: the gcp auth plugin is deprecated in v1.22+, unavailable in v1.26+; use gcloud instead.
To learn more, consult https://cloud.google.com/blog/products/containers-kubernetes/kubectl-auth-changes-in-gke
W1103 18:12:14.011524 82440 warnings.go:70] Autopilot increased resource requests for Deployment default/tigris-stack-fdb-operator to meet requirements. See http://g.co/gke/autopilot-resources.
W1103 18:12:14.362641 82440 warnings.go:70] Autopilot increased resource requests for Deployment default/tigris-server to meet requirements. See http://g.co/gke/autopilot-resources.
W1103 18:12:14.711610 82440 warnings.go:70] Autopilot increased resource requests for StatefulSet default/tigris-search to meet requirements. See http://g.co/gke/autopilot-resources.
Release "tigris-stack" has been upgraded. Happy Helming!
NAME: tigris-stack
LAST DEPLOYED: Thu Nov 3 18:12:08 2022
NAMESPACE: default
STATUS: deployed
REVISION: 2
TEST SUITE: None
You can verify that additional replicas were started, using kubectl:
❯ kubectl get pods | grep tigris
W1103 18:12:33.301669 82537 gcp.go:119] WARNING: the gcp auth plugin is deprecated in v1.22+, unavailable in v1.26+; use gcloud instead.
To learn more, consult https://cloud.google.com/blog/products/containers-kubernetes/kubectl-auth-changes-in-gke
tigris-search-0 2/2 Running 8 (25m ago) 6h16m
tigris-search-1 0/2 Pending 0 19s
tigris-search-2 0/2 Pending 0 19s
tigris-search-3 0/2 Pending 0 18s
tigris-search-4 0/2 Pending 0 18s
tigris-server-8646cb4b7b-95lcf 1/1 Running 0 6h4m
tigris-server-8646cb4b7b-gff64 1/1 Running 2 (6h5m ago) 6h5m
tigris-server-8646cb4b7b-hmxj9 1/1 Running 5 (6h5m ago) 6h16m
tigris-stack-fdb-operator-8fd845b9-zgr4t 1/1 Running 0 6h12m
The replicas should catch up quickly as there isn't a lot of search index to be synchronized. However, GKE Autopilot might need to scale up the nodes prior:
❯ kubectl describe pod tigris-search-1 | tail
W1103 18:14:04.069915 83269 gcp.go:119] WARNING: the gcp auth plugin is deprecated in v1.22+, unavailable in v1.26+; use gcloud instead.
To learn more, consult https://cloud.google.com/blog/products/containers-kubernetes/kubectl-auth-changes-in-gke
Node-Selectors: <none>
Tolerations: kubernetes.io/arch=amd64:NoSchedule
node.kubernetes.io/not-ready:NoExecute op=Exists for 300s
node.kubernetes.io/unreachable:NoExecute op=Exists for 300s
Events:
Type Reason Age From Message
---- ------ ---- ---- -------
Warning FailedScheduling 110s gke.io/optimize-utilization-scheduler 0/24 nodes are available: 24 Insufficient cpu, 24 Insufficient memory.
Normal TriggeredScaleUp 74s cluster-autoscaler pod triggered scale-up: [{https://www.googleapis.com/compute/v1/projects/mystic-berm-360500/zones/us-west2-c/instanceGroups/gk3-doc-nap-2qbw2tfi-b7486e29-grp 0->1 (max: 1000)} {https://www.googleapis.com/compute/v1/projects/mystic-berm-360500/zones/us-west2-a/instanceGroups/gk3-doc-nap-2qbw2tfi-efcf60fb-grp 0->1 (max: 1000)}]
Warning FailedScheduling 23s gke.io/optimize-utilization-scheduler 0/26 nodes are available: 2 node(s) had taint {node.kubernetes.io/not-ready: }, that the pod didn't tolerate, 24 Insufficient cpu, 24 Insufficient memory.
It should take only a minute or two to get them up Running:
❯ kubectl get pods | grep tigris-search
W1103 18:15:05.957816 83699 gcp.go:119] WARNING: the gcp auth plugin is deprecated in v1.22+, unavailable in v1.26+; use gcloud instead.
To learn more, consult https://cloud.google.com/blog/products/containers-kubernetes/kubectl-auth-changes-in-gke
tigris-search-0 2/2 Running 8 (27m ago) 6h18m
tigris-search-1 2/2 Running 0 2m52s
tigris-search-2 2/2 Running 0 2m52s
tigris-search-3 2/2 Running 0 2m51s
tigris-search-4 2/2 Running 0 2m51s
Ending TLS
For a Production installation you will want to add a certificate to your load balancer. However, as this step does not have any Tigris specific detail, we are going to skip detailing this step.
Wrapping Up!
I hope above could illustrate how easy it is to deploy Tigris to GKE Autopilot! Feel free to compare it to the article about deploying Tigris to EKS where we discussed the steps necessary to deploy it to AWS!
If you have any suggestions for us on Tigris related subjects that you think people might find interesting, feel free to reach out to us on either our Tigris Community Slack channel or our Tigris Discord server!
Hope you enjoyed reading or watching this blog or vlog! If you did, stay tuned as next we are going to cover a few interesting subjects such as performing logical backups and restores with Tigris!
Tigris is the data platform built for developers! Use it as a scalable, ACID transactional, real-time backend for your serverless applications. Build data-rich features without worrying about slow queries or missing indexes. Seamlessly implement search within your applications with its embedded search engine. Connect serverless functions with its event streams to build highly responsive applications that scale automatically.
Get early access and try out Tigris for your next application. Join our Slack or Discord community to ask any questions you might have.
