Resource limits
Since v0.10.0 every service can have resource limits, so one service that leaks memory or forks without end is stopped by its own limit, not by the server running out.
Where to set them
Section titled “Where to set them”- Sites and Compose services: the site’s Settings → Resources. A Compose project is limited per service; its
*entry is what services without their own limits get. - Workers and daemons: their form in the site’s Processes.
- Database containers: memory and CPUs in the database’s Limits & settings (see Create databases).
- Functions: their own memory and scaling settings.
Static sites have no limits. FrankenPHP sites run inside the shared edge, so they only take the restart policy, log caps and OOM preference.
The limits
Section titled “The limits”| Limit | Meaning |
|---|---|
| Memory limit | MB, at least 32. Bounded by the smallest server of the site |
| Memory reservation | MB, at most the limit; counted in the capacity view |
| CPUs | Cores, decimal (0.5, 2) |
Processes (pids_limit) |
The most processes and threads the service may run |
| Restart policy | always, unless-stopped or on-failure, with max restarts for on-failure |
| Log size and files | MB per log file and number of files kept |
| OOM | protect (killed last) or normal |
Leave a field empty for no limit.
How they are applied
Section titled “How they are applied”| Service | Mechanism | Takes effect |
|---|---|---|
| Docker sites | Container limits | Live (docker update); log caps, OOM and removed limits at the next deploy |
| Compose services | A generated compose.falak.yaml override (mem_limit, cpus, pids_limit and the matching deploy.resources, logging) |
Next deploy |
| Classic sites, workers, daemons | A systemd slice per service (MemoryMax, MemoryHigh, CPUQuota, TasksMax); the site’s web process, Octane and Horizon share the site’s slice |
Live |
The panel says which applies after you save: live, redeploy, or nothing to do.
Defaults per environment
Section titled “Defaults per environment”Limits are written on a service when it is created, never merged later, so changing a default or moving a service never changes the limits of services that exist.
| Environment | Memory | CPUs | Processes | Logs |
|---|---|---|---|---|
| Production | none | none | none | none |
| Other environments | 512 MB | 1 | 512 | 20 MB × 3 |
| Previews | 384 MB | 0.5 | 256 | 10 MB × 2 |
| Fork previews | 256 MB | 0.25 | 128 | 5 MB × 2 |
The operator changes the non-production defaults with FALAK_LIMITS_NONPROD_MEMORY_MB, _CPUS, _PIDS, _LOG_MB and _LOG_FILES (and FALAK_LIMITS_PREVIEW_*, FALAK_LIMITS_FORK_PREVIEW_*) in /opt/falak/custom.env.
Capacity per server
Section titled “Capacity per server”The server page’s Capacity card adds up what every service on it may use — sites, Compose services, workers, daemons, database containers (their whole memory) and functions (max_instances × their memory) — against the server’s RAM and cores, and warns when memory limits, reservations or CPU limits are overcommitted. The API has it as GET /api/v1/servers/{server}/capacity.
OOM kills and restart loops
Section titled “OOM kills and restart loops”Agents report out-of-memory kills and restarts of managed containers, slices and supervised programs. Two alerts follow:
| Alert type | When |
|---|---|
limits.oom_killed |
A service was killed for running out of memory |
limits.restart_loop |
A service restarted 5 times within an hour (FALAK_LIMITS_RESTART_LOOP) |
Both can trigger an automatic rollback while a release is watched after deploy.