@@ -131,3 +131,31 @@ If the upgrade fails for any reason, please run the `helm rollback` command and
131131
132132Once you complete this process, you'll be able to access Coder using the
133133external IP address of the ingress controller in your cluster.
134+
135+ # ## Scaling Coder & PostgreSQL
136+
137+ Each Coder replica maintains a pool of connections to Postgres, capped at a
138+ maximum of 30 connections by default. This value strikes a balance between
139+ allowing multiple concurrent requests and using up connections on the Postgres
140+ database.
141+
142+ To reconfigure this value, set the environment variable
143+ ` CODER_MAX_DB_CONNECTIONS` to your desired value in the `coderd.extraEnvs`
144+ section of the Helm chart.
145+
146+ > **Important**: You must ensure that Coder's maximum connections per replica
147+ > times the number of Coder replicas is less than the maximum number of
148+ > connections your Postgres instance allows, assuming Coder is the only
149+ > application using the Postgres instance (less if other applications will also
150+ > consume connections). For example, with 3 Coder replicas and the default max
151+ > connections of 30, you must ensure Postgres is configured to allow 90
152+ > connections. If Postgres runs out of connections Coder may fail API requests
153+ > or builds.
154+
155+ By default, Postgres allows a maximum of 100 open connections (less 3 reserved
156+ for superusers). Consult
157+ [Postgres documentation](https://www.postgresql.org/docs/12/runtime-config-connection.html),
158+ or your database administrator for information on how to change this value.
159+
160+ Consider monitoring the [`go_sql_open_connections`](../../admin/prometheus.md)
161+ metric to see how many connections Coder uses in different load scenarios.
0 commit comments