Lighthouse: tokio-runtime segfault

[lightclient]

Description

I encountered some db corruption and I believe the root cause is lighthouse crashes regularly due to an issue with the tokio runtime. I'm continuing to debug. One concern is that the VPS I am running on doesn't have stellar stability reviews. I'm wondering if that is causing the runtime to go awry. However, I have managed to maintain 100% uptime on my go-ethereum nodes.

Version

I've run into this issue on both 5828ff1, v1.0, and v1.0.1.

Present Behaviour

The tokio-runtime segfaults after a period of time:

tokio-runtime-w[3460]: segfault at 470ba01feea ip 00005641f1424f1b sp 00007f59e8ee21f8 error 6 in lighthouse[5641ef75e000+1d9e000]

[lightclient]

Debugging this further, it seems that anytime I get close to maxing out the RAM it crashes.

[michaelsproul]

We haven't seen this issue on any other hardware, so it would be super interesting to know what's unique about your setup that's causing this.

It might be a bug in Tokio related to the specific (janky?) hypervisor your VPS provider uses. You could run something like slabbed-or-not to try and work out which hypervisor that might be https://github.com/kaniini/slabbed-or-not

Some of the mysterious database errors we've seen have also been from people running under a hypervisor, which might just be a coincidence, but I'm not sure.

The bug may happen to be fixed by the Tokio 0.3 change (which is almost ready for release in v1.0.2)

[lightclient]

I'm running on Contabo VPS.

$ ./slabbed-or-not
Not running under any known container type
Hypervisor: KVM

I'll keep an eye for the Tokio 0.3 release :)

[AgeManning]

The tokio 0.3 release has been merged. Let us know if the issue persists.

I believe one cause of this issue was running with a high peer count (likely more than the computer can handle).

[AgeManning]

I'm going to close this issue, assuming it has been resolved. Please re-open if the issue persists.

[lightclient]

It does appear a bit more stable, but still seeing the same issue on 2383bfe with 50 peers. I will try to ramp down to 30 to see if it improves.

[lightclient]

@AgeManning FYI, it doesn't look like I have the ability to reopen.

[AgeManning]

hmm are you running any strange hardware? What OS?

[AgeManning]

I only saw one reported issue of seg faults in tokio recently and looks like it's been fixed in 0.3: https://github.com/tokio-rs/tokio/pull/3019

[lightclient]

@AgeManning running Ubuntu 20.04 on a VPS. 8GB ram and 4 xeon vCores. The crash is relatively consistent, I've reprovisioned a few times only to find the same error. Maybe this week I can figure out how to get a proper core dump to share.

[pawanjay176]

@lightclient which lighthouse version are you running?

[lightclient]

@pawanjay176 I've run all of the following: 5828ff1, v1.0 (c6baa0e) , v1.0.1 (5a3b94cb), cut-v1.0.2 (2383bfe), and v1.0.2 (f718309).

[AgeManning]

Are you building the binary locally on the box?

Can you try running the portable version and see if it also happens there

make build-x86_64-portable

and use the binary at target/x86_64-unknown-linug-gnu/release/lighthouse

[lightclient]

@AgeManning okay, I'll give that a shot. I've been building locally and have been using the optimized version.