Resource Management

Each VM consumes real system memory and CPU. If Provium launches too many VMs simultaneously, the host can run out of memory or become CPU-saturated. The resource pool prevents this by gating VM boot on available resources.

Auto mode (default) #

Without explicit --mem or --cpus flags, Provium uses auto-detection:

• Memory: Before each VM boot, Provium reads /proc/meminfo MemAvailable. If there is not enough free memory (accounting for the VM's configured size plus a 1.5 GB host reserve), the boot waits until other VMs shut down and release their memory.
• CPU: Provium limits parallelism to half the host CPU count (minimum 2) to avoid saturating KVM and vhost-vsock resources.

Auto mode adapts to the host's actual load. If other processes are consuming memory, fewer VMs run concurrently.

Fixed mode #

With --mem and/or --cpus, Provium uses a fixed resource budget:

provium test --mem 4G --cpus 8 tests/

VMs acquire resources from the declared pool:

• A VM configured with memory = "512M" and cpus = 1 takes 512 MB and 1 CPU from the pool
• If the pool does not have enough resources, the VM boot blocks until another VM shuts down
• When a VM shuts down, its resources are returned to the pool

Fixed mode is predictable — the same resource budget always produces the same scheduling behavior. It is useful on CI systems or shared machines where you want to cap Provium's resource usage.

provium test --mem 4G --cpus 8 tests/

resource pool: 4096M memory, 8 CPUs

  kacs_boot.lua ... ok (2.1s)
  kacs_token.lua ... ok (86 tests, 4.3s)
  ...

How resources are calculated #

Each VM's resource usage comes from the boot options in the Lua script:

vm:boot({memory = "1G", cpus = 2})
-- Uses 1 GB from the memory pool, 2 from the CPU pool

vm:boot()
-- Uses 512 MB (default) from memory, 1 (default) from CPUs

Resources are acquired at boot time and released when the VM is shut down or killed.

Parallelism interaction #

The -p flag controls how many test files run concurrently. The resource pool controls how many VMs run concurrently within those tests.

• Auto mode: The -p flag and memory pool both gate concurrency. A test file may be running but its VM boot may be waiting for memory.
• Fixed mode: When --mem or --cpus is set, -p defaults to the total number of test files (unlimited concurrency) since the pool handles all gating.
• Explicit -p: Always respected. Combines with the pool — the semaphore limits test concurrency, and the pool limits VM concurrency within those tests.