Skip to content
This documentation applies to Codacy Self-hosted v7.0.0

For the latest updates and improvements, see the latest Cloud documentation instead.

System requirements#

Before installing Codacy Self-hosted you must ensure that you have the following infrastructure correctly provisioned and configured:

The next sections describe in detail how to set up these prerequisites.

Git provider#

To use Codacy Self-hosted, you must use one or more of our supported Git providers. In particular, if you're using a self-hosted Git provider, make sure that your version is supported by Codacy.

Kubernetes or MicroK8s cluster setup#

The cluster running Codacy must satisfy the following requirements:

  • The infrastructure hosting the cluster must be provisioned with the hardware and networking requirements described below
  • The orchestration platform managing the cluster must be one of:
    • Kubernetes version 1.15.* to 1.20.* (1.20 recommended)
    • MicroK8s version 1.15.* to 1.19.* (1.19 recommended)
  • The NGINX Ingress controller must be installed and correctly set up in the cluster

Cluster networking requirements#

The cluster must be configured to accept and establish connections on the following ports:

Service Protocol/Port Notes
Inbound SSH TCP/22 MicroK8s only, to access the infrastructure remotely.
Inbound HTTP TCP/80 Allow access to the Codacy website and API endpoints
Inbound HTTPS TCP/443 Allow access to the Codacy website and API endpoints
Outbound PostgreSQL TCP/5432 Connection to the PostgreSQL DBMS
Outbound SMTP TCP/25 Connection to your SMTP server
Outbound SMTPS TCP/465 Connection to your SMTP server over TLS/SSL
Outbound Docker Hub * Connection to Docker Hub to download the required container images
Outbound Git provider * Connection to the ports required by your remote Git provider

Cluster hardware requirements#

The high-level architecture described in the next section is important in understanding how Codacy uses and allocates hardware resources. Below we also provide guidance on resource provisioning for typical scenarios.

For a custom hardware resource recommendation, please contact us at support@codacy.com.

Codacy architecture#

You can look at Codacy separately as two groups of components:

  • The "Platform" contains the UI and other components important to treat and show results
  • The "Analysis" is the swarm of workers that run between one and four linters simultaneously, depending on factors such as the number of files or the programming languages used in your projects

High-level Codacy architecture

Since all components are running on a cluster, you can increase the number of pod replicas in every deployment to give you more resilience and throughput, at a cost of increased resource usage.

The following is a simplified overview of how to calculate resource allocation for the group of components "Platform" and "Analysis":

Group of components vCPU Memory
Platform
(1 pod replica per component)
4 8 GB
Analysis
(1 Analysis Worker pod with up to 4 linters)
5
(per Analysis Worker)
10 GB
(per Analysis Worker)

Standard cluster provisioning#

As described in the section above, Codacy's architecture allows scaling the "Analysis" group of components, meaning that the resources needed for Codacy depend mainly on the rate of commits done by your team that Codacy will be analyzing.

The resources recommended on the following table are based on our experience and are also the defaults in the values-production.yaml file. You might need to adapt these defaults taking into account your use case. In particular, you should set the value of global.workerManager.workers.config.dedicatedMax to the maximum number of concurrent analysis depending on the available resources and number of replicas per component.

Note

For MicroK8s clusters we added an extra 1.5 vCPU and 1.5 GB memory to the "Platform" to account for the MicroK8s platform itself running on the same machine.

Installation type Pod replicas per component Max. concurrent analysis Platform resources Analysis resources ~ Total resources
Kubernetes
Small Installation
1 2 4 vCPUs
8 GB RAM
10 vCPUs
20 GB RAM
16 vCPUs
32 GB RAM
Kubernetes
Medium Installation (default)
2 4 8 vCPUs
16 GB RAM
20 vCPUs
40 GB RAM
32 vCPUs
64 GB RAM
Kubernetes
Big Installation
2+ 10+ 8+ vCPUs
16+ GB RAM
50+ vCPUs
100+ GB RAM
60+ vCPUs
110+ GB RAM
MicroK8s
Minimum
1 2 5.5 vCPUs
9.5 GB RAM
10 vCPUs
20 GB RAM
16 vCPUs
32 GB RAM
MicroK8s
Recommended (default)
1+ 2 9.5+ vCPUs
17.5+ GB RAM
10 vCPUs
20 GB RAM
21+ vCPUs
40+ GB RAM

The storage requirements recommended on the following table depend mainly on the number of repositories that Codacy will be analyzing and should be used as a guideline to determine your installation requirements.

Component Bundled in the chart? Minimum recommended
NFS Yes 200 GB
RabbitMQ Yes 8 GB
Minio Yes 20 GB
PostgreSQL No (external DB recommended) 500 GB+

PostgreSQL server setup#

Codacy requires a database server to persist data that must satisfy the following requirements:

  • The infrastructure hosting the database server must be provisioned with the hardware requirements described below
  • The DBMS server must be PostgreSQL version 10.15
  • The PostgreSQL server must be configured to accept connections from the cluster
  • The Codacy databases and a dedicated user must be created using the instructions below

Important

Google, the developer of Kubernetes, doesn't recommend running database servers on your cluster. As such, consider using a managed solution like Amazon RDS or Google Cloud SQL, or running the PostgreSQL server on a dedicated virtual machine.

We recommend that you use a managed solution to reduce maintenance and configuration costs of the PostgreSQL server. The main cloud providers all have this service that you can use, for example:

PostgreSQL hardware requirements#

The following are the minimum specifications recommended for provisioning the PostgreSQL server:

vCPUs Memory Storage Max. concurrent connections
4 8 GB 500 GB+ 300

Preparing PostgreSQL for Codacy#

Before installing Codacy you must create a set of databases that will be used by Codacy to persist data. We also recommend that you create a dedicated user for Codacy, with access permissions only to the databases that are specific to Codacy:

  1. Connect to the PostgreSQL server as a database admin user. For example, using the psql command line client:

    psql -U postgres -h <PostgreSQL server hostname>
    
  2. Create the dedicated user that Codacy will use to connect to PostgreSQL. Make sure that you change the username and password to suit your security needs:

    CREATE USER codacy WITH PASSWORD 'codacy';
    ALTER ROLE codacy WITH CREATEDB;
    

    Take note of the username and password you define, as you will require them later to configure the connection from Codacy to the PostgreSQL server.

  3. Make sure that you can connect to the PostgreSQL database using the newly created user. For example, using the psql command line client:

    psql -U codacy -d postgres -h <PostgreSQL server hostname>
    
  4. Create the databases required by Codacy:

    CREATE DATABASE accounts WITH OWNER=codacy;
    CREATE DATABASE analysis WITH OWNER=codacy;
    CREATE DATABASE results WITH OWNER=codacy;
    CREATE DATABASE metrics WITH OWNER=codacy;
    CREATE DATABASE filestore WITH OWNER=codacy;
    CREATE DATABASE jobs WITH OWNER=codacy;
    CREATE DATABASE listener WITH OWNER=codacy;
    CREATE DATABASE crow WITH OWNER=codacy;
    

Share your feedback 📢

Did this page help you?

Thanks for the feedback! Is there anything else you'd like to tell us about this page?

We're sorry to hear that. Please let us know what we can improve:

Alternatively, you can create a more detailed issue on our GitHub repository.

Thanks for helping improve the Codacy documentation.

If you have a question or need help please contact support@codacy.com.

Last modified February 25, 2022