Repository navigation
apps/rabbitmq: default resourcesPreset t1.nano (128Mi) OOMKills RabbitMQ 4.2 — default install never starts #4003
Description
What happened
A default apps.cozystack.io/RabbitMQ install does not come up. packages/apps/rabbitmq/values.yaml defaults to version: v4.2 (image rabbitmq:4.2.4-management) and resourcesPreset: "t1.nano", which resolves to 250m CPU / 128Mi memory (packages/apps/rabbitmq/charts/cozy-lib/templates/_resourcepresets.tpl). RabbitMQ 4.2's Erlang VM does not fit in 128Mi and the server container is OOMKilled during boot, so the pod never becomes ready.
Evidence
Observed on a dev cluster while working on #3990. The rabbitmq-<name>-server-0 container's last state:
terminated: { reason: OOMKilled, exitCode: 137 }
with repeated BackOff restarting failed container rabbitmq and readiness-probe failures. Bumping the same instance to resourcesPreset: c1.small (1Gi) boots cleanly with 0 restarts. The default (t1.nano) does not.
Why it matters
This is the out-of-the-box default for the managed RabbitMQ app on current main, so a user creating a RabbitMQ with no overrides gets a broker that never starts. It was noticed only in passing during a backup PR (#3990), which is why it is filed separately.
Suggested fix
Raise the default resourcesPreset for apps/rabbitmq to something that fits RabbitMQ 4.2 (e.g. c1.small = 1 CPU / 1Gi, or t1.small/s1.micro), and/or document a minimum. Someone should reproduce and pick the right floor per supported version.
/kind bug
Activity
dosubot commentedon Sep 1, 2026
Andrey Kolkov (@androndo) this is already being addressed in PR #3936 ("fix(rabbitmq): right-size the default resources preset to s1.nano"), currently open and authored by yankawai [1].
The PR raises the default
resourcesPresetfromt1.nano(250m CPU / 128Mi) tos1.nano(250m CPU / 512Mi) [2]. The rationale matches your findings exactly: a default RabbitMQ 4.2 broker was OOMKilled (exit 137, 10 restarts, never Ready), while a canary on a larger preset went 3/3 Ready with zero restarts and a working-set peak of 125.8Mi — 98% of the 128Mi ceiling [1].An earlier revision proposed
u1.nano(1Gi), but review pushed back as over-provisioned without a measured peak.s1.nanowas chosen instead as the smallest preset that clears the ceiling with a ~4× margin over the observed peak [1].One thing to be aware of once it merges: instances that never set
resourcesPresetexplicitly will pick up the new default on the next Helm reconcile, raising guaranteed memory reservation from 128Mi to 512Mi per replica (~1.1Gi extra for a 3-replica install). Instances that already pinresourcesPresetexplicitly are unaffected. A corresponding Terraform provider update is tracked in cozystack/terraform-provider-cozystack#34 [1].To reply, just mention @dosu.
Share context across your team and agents. Try Dosu.
lexfrei commentedon Sep 23, 2026
Fixed by #3936: the default
resourcesPresetfor rabbitmq is nows1.nano(512Mi). Please reopen if a default install still gets OOMKilled on a build with it.