[Bug] 2.0 RC2 Start Harvester Fails

[BrandtH22]

What happened?

Bug reported by Daryl in Discord (https://discord.com/channels/1034523881404370984/1099818017908592660/1135638423118549083)

Updated from the first 2.0 alpha release to the 2.0 rc2 release, but it keeps throwing an error when I try to start the harvester.

Attempted resolutions:

• Deleted config and started fresh
• enabled gpu decompression
• set parallel decompressor threads to 1.
• Threads were at 8, tried to bump them down to 1, no such luck
• updated and upgraded all my packages, then restarted to make sure something wasn't hanging on to the GPU or something. Just won't start the harvester
• Only way to get around the error was to rename it back to decompresser (from decompressor), but then GPU farming wasn't working

More info:
Nvidia Driver: 530.30.02
CUDA Version: 12.1
Ubuntu desktop, 40 cores with 256GB of RAM, RTX a4000

Steps to reproduce:

• Close gui
• Chia stop all -d
• backup config file to desktop
• delete config file
• sudo dpkg --install chia-blockchain_2.0.0rc2_amd64.deb
• chia init
• set parallel_decompressor_count to 1
• add plot directories
• switch use_gpu_harvesting to true
• run chia start harvester

Config file harvester section:

harvester:
  chia_ssl_ca:
    crt: config/ssl/ca/chia_ca.crt
    key: config/ssl/ca/chia_ca.key
  decompressor_thread_count: 0
  disable_cpu_affinity: false
  enforce_gpu_index: false
  farmer_peer:
    host: localhost
    port: 8447
  gpu_index: 0
  logging: *id001
  max_compression_level_allowed: 7
  network_overrides: *id002
  num_threads: 30
  parallel_decompressor_count: 1
  parallel_read: true
  plot_directories:
  - /mnt/j301
  plots_refresh_parameter:
    batch_size: 300
    batch_sleep_milliseconds: 1
    interval_seconds: 120
    retry_invalid_seconds: 1200
  port: 8448
  private_ssl_ca:
    crt: config/ssl/ca/private_ca.crt
    key: config/ssl/ca/private_ca.key
  recursive_plot_scan: false
  rpc_port: 8560
  selected_network: mainnet
  ssl:
    private_crt: config/ssl/harvester/private_harvester.crt
    private_key: config/ssl/harvester/private_harvester.key
  start_rpc_server: true
  use_gpu_harvesting: true

Version

2.0 RC2

What platform are you using?

Linux

What ui mode are you using?

CLI

Relevant log output

No response

[jmhands]

I reproduced on all 3 of my machines on Ubuntu server (22.04 & 23) with GPU harvesting enabled.

  File "chia/server/start_harvester.py", line 78, in <module>
  File "chia/server/start_harvester.py", line 74, in main
  File "chia/server/start_service.py", line 319, in async_run
  File "asyncio/runners.py", line 44, in run
  File "asyncio/base_events.py", line 616, in run_until_complete
  File "chia/server/start_harvester.py", line 66, in async_main
  File "chia/server/start_harvester.py", line 37, in create_harvester_service
  File "chia/harvester/harvester.py", line 108, in __init__
  File "chia/plotting/manager.py", line 106, in configure_decompressor
RuntimeError: Failed to preallocate memory for contexts with result 2
[3983716] Failed to execute script 'start_harvester' due to unhandled exception!

workaround is to reinstall nvidia driver and reboot system, then with no change to config.yaml the harvester was able to start

[wallentx]

We should try this with rc3, since we've added all architectures

[jmhands]

this has been fixed now, rc6