client.scheduler_info()['workers'] only returns 5 workers when LocalCluster n_workers > 5

[Ajaypatel1234]

Describe the issue:

When creating a LocalCluster with more than 5 workers, all workers are successfully created and initially register with the scheduler, but client.scheduler_info()['workers'] only returns 5 workers. The remaining workers fail with CommClosedError immediately after registration and are removed from the scheduler.

This issue occurs regardless of the total number of workers requested (tested with 6, 8, 10+ workers) - only 5 workers ever appear in client.scheduler_info()['workers']. The issue occurs with both processes=True and processes=False.

Minimal Complete Verifiable Example:

import time
from distributed import Client, LocalCluster


if __name__ == "__main__":
    # Request 8 threaded workers
    cluster = LocalCluster(
        n_workers=8,
        threads_per_worker=1,
        processes=False,
        silence_logs=False
    )
    client = Client(cluster)

    print(f"Requested workers: 8")
    print(f"Scheduler address: {client.scheduler_info()['address']}")

    # Wait for workers to start
    time.sleep(3)

    workers = client.scheduler_info()['workers']
    print(f"Actual workers connected: {len(workers)}")
    print(f"Worker IDs: {sorted([w['id'] for w in workers.values()])}")

    for worker_key, worker_info in workers.items():
        print(f"  - Worker ID: {worker_info['id']} | Threads: {worker_info['nthreads']}")

    client.close()
    cluster.close()

Expected output:

Requested workers: 8
Actual workers connected: 8
Worker IDs: [0, 1, 2, 3, 4, 5, 6, 7]

Actual output:

Requested workers: 8
Actual workers connected: 5
Worker IDs: [Only 5 workers appear, specific IDs vary]
  - (5 workers shown, each with 1 thread)

Anything else we need to know?:

• When examining logs with silence_logs=False, all 8 workers are created and register with the scheduler, but 3 workers immediately encounter CommClosedError exceptions and are removed from the scheduler
• The specific workers that fail varies between processes=True and processes=False, but exactly 3 workers always fail, leaving 5 connected
• client.wait_for_workers(n_workers=8) returns successfully, claiming all 8 workers are ready, but client.scheduler_info()['workers'] only shows 5 workers
• The issue occurs with both processes=True and processes=False configurations
• Tested on two completely different systems (macOS ARM and Linux x86_64) with identical results
• The limitation appears to be hardcoded at 5 workers - requesting 6, 8, 10, or 20 workers always results in exactly 5 workers being reported by scheduler_info()
• Also tested with Dask version 2025.4.1 on both Python 3.13.3 and 3.12.12 - the issue persists in that version as well

Environment:

macOS System:

• Dask version: 2026.1.2
• Python version: 3.13.3 and 3.12.12 (both exhibit the same issue)
• Operating System: macOS (Apple M2 Pro, 10 cores)
• Install method: pip install "dask[complete]"

Linux System:

• Dask version: 2026.1.2
• Python version: 3.13.3 and 3.12.12 (both exhibit the same issue)
• Operating System: Linux (AMD EPYC 7702 64-Core Processor, 256 CPUs)
• Install method: pip install "dask[complete]"

Key dependencies installed:

bokeh            3.8.2
cloudpickle      3.1.2
distributed      2026.1.2
fsspec           2026.1.0
lz4              4.4.5
msgpack          1.1.2
numpy            2.4.2
pandas           3.0.0
psutil           7.2.2
tornado          6.5.4

[Ajaypatel1234]

The client.scheduler_info() has a default parameter n_workers=5 and I really lost the track of in which release this parameter was set.

I have set it to -1 to get all the workers' information.