Summary

A Nextflow plugin that replaces the default /proc-based memory tracking with cgroup-aware tracking for containerized tasks.

Nextflow's default memory tracking reads /proc//status, which reports memory for the main process only. In containerized environments, this misses memory-mapped files (e.g. genome indices loaded by STAR, BWA, Bowtie2), page cache from I/O-heavy workloads, and container-wide memory enforced by the cgroup runtime. This plugin reads from the cgroup memory controller instead — the same accounting that the container runtime uses to enforce limits and trigger OOM kills.

Supports cgroup v1 and v2. Falls back to default /proc tracking if cgroups are unavailable.

Get Started

Requirements

• Nextflow >= 24.04.0
• Java 17 or later (for building from source)
• Linux with cgroup v1 or v2
• A containerized executor (Docker, AWS Batch, etc.)

Install from source

git clone https://github.com/watchmaker-genomics/nf-cgroup-metrics.git
cd nf-cgroup-metrics
make install

This installs the plugin to ~/.nextflow/plugins/nf-cgroup-metrics-1.0.0/.

Configuration

Add to your nextflow.config:

plugins {
    id 'nf-cgroup-metrics@1.0.0'
}

The plugin is enabled by default. Optional settings:

cgroup {
    metrics {
        enabled = true    // set false to disable
        verbose = false   // set true for debug logging
    }
}

The plugin only modifies containerized tasks. Tasks running on the local executor without a container use default /proc tracking.

Examples

AWS Batch

plugins {
    id 'nf-cgroup-metrics@1.0.0'
}

process {
    executor = 'awsbatch'
    queue = 'my-batch-queue'
}

aws.region = 'us-east-1'

Checking that it works

Enable verbose logging and run a pipeline:

cgroup {
    metrics {
        verbose = true
    }
}

nextflow run main.nf

The trace report fields vmem, rss, peak_vmem, and peak_rss will reflect container-wide memory usage rather than per-process values.

Verify cgroup access

# Check cgroup version
ls /sys/fs/cgroup/cgroup.controllers 2>/dev/null && echo "v2" || echo "v1 or none"
cat /proc/self/cgroup

# Check plugin installation
ls ~/.nextflow/plugins/nf-cgroup-metrics-1.0.0/

# Enable Nextflow debug mode
NXF_DEBUG=1 nextflow run main.nf

License

Apache License 2.0