PMM-14697 - HAProxy configuration for dynamic DNS discovery #760
Conversation
theTibi
commented
Jan 9, 2026
theTibi
commented
- Introduced maxReplicas parameter in values.yaml to set the maximum number of PMM replicas for HAProxy server-template.
- Updated haproxy-configmap.yaml to utilize server-template for dynamic DNS-based discovery, allowing automatic pod discovery during scaling without requiring HAProxy restarts.
- Enhanced comments for clarity on the configuration and behavior of HAProxy with respect to PMM pods.
- Introduced maxReplicas parameter in values.yaml to set the maximum number of PMM replicas for HAProxy server-template. - Updated haproxy-configmap.yaml to utilize server-template for dynamic DNS-based discovery, allowing automatic pod discovery during scaling without requiring HAProxy restarts. - Enhanced comments for clarity on the configuration and behavior of HAProxy with respect to PMM pods.
theTibi
requested review from
JiriCtvrtka and
idoqo
and removed request for
a team
- Documented the requirement for the Raft leader to be on pmm-0 when scaling down to a single PMM replica. - Included instructions on the scaling process to prevent PMM from becoming unreachable during scaling operations.
| # Use server-template for dynamic DNS-based discovery of PMM pods | ||
| # This automatically discovers pods when scaling up/down without requiring HAProxy restart | ||
| # HAProxy will re-resolve DNS based on 'hold valid' setting in the resolver (10s) | ||
| server-template pmm 1-{{ $.Values.maxReplicas | default 10 }} {{ $.Values.service.name | default "monitoring-service" }}.{{ $.Release.Namespace }}.svc.cluster.local:8443 check ssl verify none resolvers k8s init-addr last,libc,none |
There was a problem hiding this comment.
Question: Don't we have to use {{ $.Release.Name }} instead of pmm in here?
There was a problem hiding this comment.
These are not pod names this is just a prefix for Haproxy slots. Haproxy will keep 10 slots, then queries DNS for the headless service (monitoring-service.namespace.svc.cluster.local) and will use those IPs of the pods and assign them to a slot. The actual pod name does not matter for haproxy. if we want we can use something like this:
server-template {{ $.Release.Name }}- 1-{{ $.Values.maxReplicas | default 10 }} {{ $.Values.service.name | default "monitoring-service" }}.{{ $.Release.Namespace }}.svc.cluster.local:8443 check ssl verify none resolvers k8s init-addr last,libc,none
That will make hapoxy stats maybe more clearer but again these names will not correspond with the real pods, these are just reserved slots and haproxy will assign the pods randomly to the slots..
2 participants
