Document updated on Nov 11, 2022
Production best practices
Setting up KrakenD is a straightforward process, but here are some not-so-obvious recommendations to get a good start when going live. This section has generalistic advice, even though every KrakenD installation is different. So let’s dip our toe into the deployment waters!
Architecture recommendations
High Availability
Hardware can fail anytime, and a Gateway is critical enough to have redundancy. Having a cluster of machines operating the service assures high availability. It would be best if you always planned to have at least a couple of KrakenD servers/containers running in case one of them gets in trouble, even when you have low traffic.
KrakenD can run in different regions and data centers transparently without any problem, as its nodes do not need to communicate with each other.
Place a balancer in front of KrakenD
Put a load balancer in front of KrakenD to distribute traffic between the different nodes of the cluster (Kubernetes already does this for you). Use always at least two KrakenD instances for High Availability.
Server dimensioning
Dimension KrakenD nodes according to your expected needs and throughput.
Use several gateways
The API gateway doesn’t need to be unique. We recommend using an independent KrakenD installation per consumer type. For instance, your iOS development team might need a gateway with different views of the consumed content compared to the Web Team. Needs and content in each team differ in each endpoint, and every team could optimize the contract for each case.
Use HTTP2
Enable HTTP2 between your balancer and KrakenD API gateway for the best performance. There is nothing additional you need to configure in KrakenD.
SSL Certificates
Even that you can start KrakenD with SSL, you can add your public SSL certificate in the load balancer or PaaS and use internal certificates, or even no certificates at all (termination), between the load balancer and KrakenD.
Prepare for failure
Add a circuit breaker to your backends to prevent KrakenD from pushing to a failing system. In addition, if you know that a particular backend does not support more than a number of requests, add a maximum number of requests using the proxy rate limit.
Configuration recommendations
KrakenD will behave according to the configuration file, here are some recommendations:
Add logging and suppress unnecessary information
If you don’t add any logging, KrakenD will spit on stdout all the activity of the gateway. This behavior is not recommended for production.
Enable the logging with CRITICAL
, ERROR
, or WARNING
levels at most. Avoid INFO
and DEBUG
levels in production at all times.
Below there is a recommended configuration in production for a good performance:
{
"version": 3,
"extra_config": {
"telemetry/logging": {
"level": "ERROR",
"syslog": false,
"stdout": true
}
}
}
If you send the logs out to an ELK or a GELF server, use "stdout": false
to avoid having double output.
When the output of KrakenD stdout is not important to you, set the logging level to CRITICAL
and redirect its output to /dev/null
to have even more performance. To do that, start KrakenD with:
krakend run -c krakend.json >/dev/null 2>&1
Remove access logs
Removing the access log increases the number of requests per second the gateway can serve on high concurrency.Disable the access log to gain more speed. You will still have the problems logged during runtime, but the requests won’t be outputted.
{
"version": 3,
"router": {
"extra_config": {
"disable_access_log": true
}
}
}
Cache JWT keys
If you use token validation and use a jwk_url
(instead of a jwk_local_path
), enable the caching. Otherwise, each endpoint will require to download the JWK over HTTP for every request, e.g.:
{
"endpoint": "/protected/resource",
"extra_config": {
"auth/validator": {
"alg": "RS256",
"jwk_url": "https://some/.well-known/jwks.json",
"cache": true
}
}
}
Monitoring
Enable traces and metrics
Make sure you have visibility of what is going on. Choose any of the systems where you can send the metrics and enable them. There are many choices, but choose wisely and do not enable them all!. If you don’t use a SaaS provider, a good self-hosted start would be:
- A Grafana Dashboard
- Fueled by InfluxDB
- And full traces by Jaeger
Pay attention to the cardinality of the metrics. Logs and metrics might produce a lot of data and CPU activity. Aggregate and consolidate data in InfluxDB (e.g., when looking at the past year’s metrics, you don’t need minute resolution, and days will be enough).
Deployment recommendations
Release through a CI/CD pipeline
Automate the go-live process through a CI/CD pipeline that builds and checks KrakenD configuration before deploying.
At least your pipeline should have:
krakend check -d -t -c krakend.[tmpl|json|yml]
krakend check -c krakend.json --lint
If you don’t use flexible configuration, you can do it all in one line: krakend check -d - t -c krakend.json --lint
Use Docker and immutable containers
Creating an immutable Docker image with your desired configuration takes a few seconds in your CI/CD pipeline on Docker deployments. Create a Dockerfile
with at least the following code and deploy the resulting image in production:
FROM krakend/krakend-ee:2.1
COPY krakend.json /etc/krakend/krakend.json
Read more on Docker artifacts
Use blue/green or similar deployment strategy
As with Apache, Nginx, Varnish, and other stateless services, changing the configuration require a restart. When deploying new changes, use a technique like blue/green deployment or similar to make the deployment transparent for the user.
This scenario can be automated and is available in Kubernetes and in all major cloud providers. The idea is that you spin up new machines with the latest configuration and then shift the traffic from the old instances to the new ones.
This methodology ensures that there is no downtime when applying changes. Of course, on-premises installations can also make a similar approach, but the implementations depend on the underlying infrastructure.
Code organization
Name your configurations
Add a name
key in the configuration file with helpful information to identify your cluster’s specific version. Whatever type of information you write inside the name
is open to your imagination. Any value you write is available in the metrics for inspection.
{
"version": 3,
"name": "Production Cluster rev-db6a182"
}
During the build in the pipeline, it might be a good idea to replace the content of the name
attribute with content showing the deployed version (the short SHA from the commit, maybe).
Add comments and metadata (@
)
During startup, KrakenD ignores from the configuration anything that it doesn’t recognize, which means that your krakend.json
(or whatever format you use) allows you to include additional metadata and fields that make sense to your company. Use it to add your meta language, tags, comments, bot integrations, etc., for better integration with your CI/CD system, deployment process, or a better future comprehension of the file.
$schema
to validate your configuration, unknown attributes will trigger a warning during validation. To add your configurations schema-compatible, and have them ignored by KrakenD, prefix them with one of the following characters: @
, $
, _
or #
.For instance, you could add @comment
fields. KrakenD does not use the field, and it passes the JSON schema validation. Finding it might be fresh air for the developer next to you.
{
"endpoint": "/cookies",
"input_headers": ["Cookie" ],
"@comment": "At this early stage of the implementation, we still need to send cookies to the backend.",
"backend": [{
"url_pattern": "/srv/legacy"
}]
}
Split the configuration in multiple repos or folders
In large organizations with several teams using a shared gateway, you can split the endpoints into groups using folders or even different repositories. With the flexible configuration, you can have teams working in its dedicated space and aggregate all endpoints during build time without conflicts touching the same files.
Most KrakenD configurations tend to be large and with repetitive blocks. Define a basic skeleton of configurations that will be used across all teams.