Issues running poste.io in Kubernetes

I have tried running poste.io mailserver in Kubernetes (v1.19.6), and I can get it to mostly work. However, there are major issues that keep me from running it in production I have observed so far:

Excessive logging from k8s TCP port checks - I define a k8s service for each TCP port I expose, which is ports 25 (SMTP) and 143 (IMAP). K8s constantly checks the ports, and I can see in ‘Connections realtime’ in the dashboard logging for port 25, and then adding logs of these thousands of connections into /data/log/delivery/conn. My underlying storage ran out of inodes due to this.
The port checks coming from the internal k8s IPs are flagged as dangerous and get automatically blacklisted. This prevents connections even to the web UI. I found a workaround is to immediately login to the web UI after a restart, then quickly disable connection blocking; otherwise web logins get blocked.
K8s deployments do not provide a consistent hostname, and poste.io barely worked with a hostname that didn’t match its domain name. I changed it to a k8s StatefulSet, which does provide a consistent--albeit odd--hostname, and it is slightly better. However, this only gives me a consistent short hostname, not an FQDN. To lookup the FQDN, it needs to go out of the cluster and back into itself, and the health checks fail. When I manually update /etc/hosts so that the FQDN is an alias to localhost, the health checks still show up as failing.

‌

Comments (3)