`perf(supervisor): sequential inspect loop in DockerResourceMonitor blocks task scheduling thread`

[KanhaiyaBagul]

Bug Description

The DockerResourceMonitor.getNodeResources() method calculates system resource consumption by iterating through all containers. During each iteration, it performs a sequential inspect call to retrieve CPU and memory limits:

for (const container of stats) {
  if (container.State === "running") {
    const c = this.docker.getContainer(container.Id);
    const { HostConfig } = await c.inspect(); // Sequential blocking call

    const cpu = this.resourceParser.cpu(HostConfig.NanoCpus ?? 0);
    const memory = this.resourceParser.memory(HostConfig.Memory ?? 0);

    cpuUsed += cpu;
    memoryUsed += memory;
  }
}

Because getNodeResources is called frequently to make container placement decisions during scheduling, this sequential inspect loop acts as a synchronous bottleneck. On docker hosts running many containers, performing sequential inspect round-trips to the Unix socket blocks the scheduling thread for seconds and creates workload placement latency.

Root Cause File

apps/supervisor/src/resourceMonitor.ts

Steps to Reproduce

1. Start the Trigger.dev supervisor daemon on a machine hosting 50+ active docker containers.
2. Trigger multiple rapid task scheduling executions (wouldFit).
3. Notice scheduling latency and elevated CPU thread blocking inside the supervisor daemon due to sequential docker socket round-trips.

Proposed Fix

Use Promise.all to fetch container configurations concurrently rather than sequentially:

const runningContainers = stats.filter(container => container.State === "running");
const inspections = await Promise.all(
  runningContainers.map(async (container) => {
    const c = this.docker.getContainer(container.Id);
    const { HostConfig } = await c.inspect();
    return {
      cpu: this.resourceParser.cpu(HostConfig.NanoCpus ?? 0),
      memory: this.resourceParser.memory(HostConfig.Memory ?? 0),
    };
  })
);

for (const inspect of inspections) {
  cpuUsed += inspect.cpu;
  memoryUsed += inspect.memory;
}