seawall stack limits

Set a stack's resource limits: memory, CPU, processes, open files.

Set a stack's resource limits: memory, CPU, processes, open files

A dry run until --yes, which writes the disk copy and deploys it.

Usage

seawall stack limits [OPTIONS] <STACK>

Changes the host

Without --yes it prints the plan and changes nothing.

Arguments

<STACK>

The stack, by Portainer name or directory

Required.

Options

--yes

Write the change and deploy it

--never

Stop proposing limits for this stack (remembered)

--again

Propose limits for a stack that skipped them

--cpu-cap

A hard CPU quota too, sized from use (by default a CPU weight alone)

--remove <SETTING>

Take limits out instead: cpus, mem_limit, pids_limit, ulimits, or all (comma-separate), whoever set them

--undo <SETTING>

Put back what seawall changed instead (cpus, mem_limit, ..., or all): a limit it raised goes back to the one before, one it added goes, others stay

--service <SERVICE>

With --remove or --undo: only this service's

--no-deploy

Write the disk copy only, don't deploy it

--untried

Deploy even when the trial copy can't run (else the change stops there)

--diff

Show the compose diff, by service, and the notes too

--wait

After a redeploy, wait out the 2-minute compare with before. Without it the facts timer runs it (SeawallDeployWorse)

Every command also takes the global options.

Examples

seawall stack limits ghost
seawall stack limits ghost --yes
seawall stack limits monitoring --remove cpus --service cadvisor --yes
seawall stack limits monitoring --undo all
seawall stack limits game --never

Notes

Limits are a backstop against a container running away. Memory is a hard limit at twice its observed peak (an alert when a container nears it); CPU a weight (cpu_shares), which only counts when containers compete; processes one fixed cap of 4096. --cpu-cap adds a hard CPU quota, four times the 99th percentile of each minute (at least one core, raised when throttled in more than 5% of its periods); collectors and proxies never get one. With --yes it writes the disk copy and deploys it, backed up, rolled back if it doesn't settle, and watched.