Storage memory issues on node on Raspberry Pi

post by Post_Singularity on Aug 22, 2020

Hi everyone,

I’m trying to run a node on MainNet on my Raspberry Pi 3 B+.

It all ran smoothly until approximately 5.5M blocks where I started to experience slowness of the whole system but the command top showed no shortage of RAM or CPU time. The only parameter off the charts was the wait time for I/O, from which I can only guess that access to my SD card was slow.

However, my SD card runs fine when the algod process is stopped and I’ve even run benchmarks, and it all looks good.

Could there be a reason why, as the block piles up, the usage of storage memory increases to the point that the whole system is unresponsive?

post by cusma on Aug 22, 2020

Hi,

I used to run my Main Net node on Raspberry Pi 4 and some time ago started experiencing some issues too.

I suggest you to take a look at this thread in which microSD speed has been considered as one of the possible causes of the issues.

Out of curiosity: do you have issues with Fast Catchup also?

post by Post_Singularity on Aug 22, 2020

I don’t think it’s the SD card speed because I’ve copied the data folder into an SSD and then shared it through samba with the raspberry and it was barely an improvement. It would go ahead without making the system unresponsive, but it would go really slowly like 1 block/s.

My best guess is that some file dimensions grow with the syncing up of the chain and eventually it reaches a point where algod has to keep open files that are too big.

My second-best guess is that we are seeing the limitation of a shared bus bandwidth between USB ports, Ethernet, Wifi, and possibly the SD card bus?

I do. It goes up to approximately 350,000 accounts and then crashes. Before crashing, RAM usage is near 99% and the swap file builds up until the app crashes and the system becomes responsive again. This feels more like a memory leakage than anything else.