You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Currently the miner keeps crunching through available tasks, and the moment resources free up due to some task completing, another takes its place. This makes it a bit hard to gracefully shut down the worker/miner without losing progress (there's always that one PC in 5h) and it also makes it a bit hard to diagnose leaks (e.g. I'm certain there's a GPU RAM leak, but it's hard to pinpoint because there's always a new task scheduled). Last but not least, I almost always end up with "bad file descriptors" if I pull te run from underneath a PC2 and need to go back all the way to PC1 to regenerate the layers.
Wondering if it would be possible to support suspending the scheduling of new tasks, without interrupting currently running ones (and of course resuming the whole thing).
The text was updated successfully, but these errors were encountered:
Alternatively, would be sweet if I could change the tasks allowed for a worker dynamically without needing to restart the worker. That way I could turn certain tasks off and then back on later.
Closing this issue since these features are implemented.
You can enable/disable tasks with: lotus-woker set
lotus-worker set -h
NAME:
lotus-worker set - Manage worker settings
USAGE:
lotus-worker set [command options] [arguments...]
OPTIONS:
--enabled enable/disable new task processing (default: true)
And you can change tasks that are allowed dynamically with: lotus-worker tasks
lotus-worker tasks
NAME:
lotus-worker tasks - Manage task processing
USAGE:
lotus-worker tasks command [command options] [arguments...]
COMMANDS:
enable Enable a task type
disable Disable a task type
help, h Shows a list of commands or help for one command
OPTIONS:
--help, -h show help (default: false)
Currently the miner keeps crunching through available tasks, and the moment resources free up due to some task completing, another takes its place. This makes it a bit hard to gracefully shut down the worker/miner without losing progress (there's always that one PC in 5h) and it also makes it a bit hard to diagnose leaks (e.g. I'm certain there's a GPU RAM leak, but it's hard to pinpoint because there's always a new task scheduled). Last but not least, I almost always end up with "bad file descriptors" if I pull te run from underneath a PC2 and need to go back all the way to PC1 to regenerate the layers.
Wondering if it would be possible to support suspending the scheduling of new tasks, without interrupting currently running ones (and of course resuming the whole thing).
The text was updated successfully, but these errors were encountered: