-
Notifications
You must be signed in to change notification settings - Fork 242
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Farmer uses a lot of RAM #569
Comments
Looks like too many CPU cores for the amount of RAM causes too much parallel work happening on the farmer that leads to excessive RAM usage. @i1i1 please try to reproduce this and see what can be done about it. We have a bunch of places in the farmer that start parallel work without any backpressure. @mrSinisterE can you share if you're using HDD or SSD (SATA or NVMe) and how much space did you allocate for farmer? |
@nazar-pc HDD, 750G |
Thanks for report, @mrSinisterE! |
@mrSinisterE, thanks for all your input. It would be helpful if you could provide your cpu info. Just run |
You can see it is 20 cores with 16G of RAM, doesn't matter what CPU too much I think |
I just think that total number of cores/threads is hidden by the window? Anyway, will test for 20 cores setup |
10 cores, 20 threads |
Hmm, 🤔 can't reproduce this issue. Tried following:
I seted up 20 threads for rayon, so it should be close to that. 🤷. Also plot size was set to 750G. Will try syncing for a bit longer |
Architecture: x86_64 At the time the video was recorded, the size of the plot was almost completely full. |
@i1i1 I think this might be related to recommitments according to forum thread. Try to plot a bunch of stuff and they run recommitment over it (on proper plot just stopping farmer, deleting commitments databases and starting again will work). |
This is fairly easy to reproduce. Plot 750G fully, wait for salt update and RAM requirements slowly go up to 32G per farmer process. Iscpu output below. ]# dmesg lscpuArchitecture: x86_64 |
@nazar-pc tried recommitment bench on 750G plot, and it consumes around 750-900M of ram. Will try to sync node with such setup |
Here is a stack trace while it was consuming 25G
|
Found the reason, working on solution now |
Tested fix from upcoming release, used ~400M of RAM on the farmer with 300G plot size. I think this issue will be resolved. |
I'm not sure it is same issue, but farmer using 20G+ RAM with 2TB plot I found some discussion on Discord, said that Windows report wrong RAM usage. |
I'm fairly certain it is not. Please create a forum thead with additional information like screenshot of process list that shows how much RAM farmer uses. This resolved issue is not the best place for such discussions. |
Node is looped on one block and then fills all the RAM staying on the same block.
aa3.mp4
The text was updated successfully, but these errors were encountered: