Gigahorse Harvester unexpected stop (Windows)

2023-06-08T13:57:30.006 daemon chia.daemon.server : INFO Unexpected message type. Closing websocket with [‘chia_harvester’]. 257

any clue?

1 Like

Same observation today on one of my harvesters.
Version 1.8.1.giga13

There is a bug report related to the message

but in this case it was extremly high number of plots in my case less than 5k plots

I assume it happens if a compressed plots passes the filter needs to be inflated but you do not have a GPU being able to do so. My testsetup has no GPU yet included (need to change CPU cooler to get GPU in) but already some C8 plots. Only guessing but as soon as I remove the compressed plots from the config it works stable.

did you find any root cause?

several month the issue was gone on my farm now after upgrading to 2.1.1 I got the issue again. (with the same compute etc. setup I had working before)

Hi, back on those days, I update for newer version, and it solved the problem. it happened to me again on Linux with GH 2.1.1 too, restarting service didn’t work, but after restarting the server it fixed the problem so far. But, no idea how stable will it be.

I added a second remote compute server and it improved (compute on linux and window, farmer and harvester on linux GH 22). Reboot did not change anything. Since I added some C20 test plots it might be related to GPU power even though I have none too late plots.

I did not find any concrete reason for this issue except a hint related to dimensioning.

C20? nice, I play safe on C18, if you are harvesting, plotting, and recompute as I’m doing on same GPU, use -S 2 on plotter, it reduce GPU memory usage a lot, that can be a cause as well.
Nvidia tunning: Clock +150 / MEM -502 / Power 90%

1 Like

thanks - 30% uncompressed, 60% C8, 10% C20 currently. I will most likely go for C19 for all. Starting to replot rest of the uncompressed ones. C20 takes too much GPU for a bigger farm. Currently having a 3090 as plotter and recompute + 3060 for recompute only. Long term I will add a second 3060 to have the 3090 free again.

I just wrote the following script which I run every 5min on the harvester. It checks if the harvester is still running if not triggers a restart.

pidof chia_harvester  >/dev/null
if [[ $? -ne 0 ]] ; then
        echo "Restarting chia harvester:     $(date)" >> /root/chia_harvester_restarted.log
        /root/ &
touch /root/chia_harvester_lastcheck.log

in I start the harvester and set some env vars.

thanks for the script, will try it!