[BUG] multiple new_signage_point spam, looks like malicious network attack #5410
Comments
|
To my understanding this happens on chain forks or if you fell behind in the network for some reasons (unlucky peer-connections or your own network issues). So far i've seen, there is some kind of fix coming for this in the next patch - at least as far as logging goes. If you had proper monitoring, you'd see, that this rarely affects the farmer/harvesters as far as being able to farm for this SPs goes. |
|
As far as I can tell, I never fall behind the chia network, always green "synced", always on latest height, same height my all other ~60 full node peers are at. It does affect my farming, because every burst forces my harvesters to search in plots for every extra signage point, effectively hanging my harvesters for ~30-60 seconds overall (as far as I can interpret logs). That's like 6 missed farm opportunities every time it happens. Now that you've mentioned it, I do see 1c808b6 "Fix duplicate signage point spikes" by @mariano54 , thanks for letting me know! If devs can confirm this is exact case that was fixed, then maybe we could close this bug. I'm a bit relieved now, thanks. |
|
I also did see those spikes coming in my logs, couldn't find any hard evidence for a specific issue so i just assumed it comes from forked chains -> where you get alot of signage points from the past seconds/minutes where a parallel chain reaches you. At least this is the conclusion i came to, might be wrong tho, so let's hope it gets fixed or at least improved in the next patch :) Ps. It would be nice to know in the logs for which specific block the signage points are for(logging also the block hash for example) |
|
I also see this problem, and like you it causes a metric ton of proofs in a very short space of time, which also locks up new proofs from working properly. We discussed it in this thread: #1796 (comment) Anecdotal evidence suggests that the times of the problems are the same for more than just me, which suggests a central issue of some kind. I agree this could be a malicious attack vector. Not seen a single comment from a developer on preventing this though, it could be worked around by caching all recent signage hashes and not attempting one it's already done, that won't prevent the signage spam, but it would prevent it from actually responding to it. |
|
It is gone with 1.1.6 |
Describe the bug
Every ~10 seconds I should be getting a new signage point to test my plots against it. Usually it works fine. But sometimes I get tens (10x+) signage points at the same time with no apparent reason. It leads to extra load and lookup timeouts.
To Reproduce
Steps to reproduce the behavior:
Expected behavior
Signage points should come one at a time, not in multiple bursts.
Screenshots
Desktop (please complete the following information):
Additional context
I have created a Discussion post #4803 , investigated it from "maybe my drives are slow" into this current state "full node is sending bursts of signage points with no apparent reason".
Currently it kind of looks like a malicious attack on network, maybe one or some of my peers or their peers are sending this. Maybe there's something with distinguishing unique messages? (doesn't look like it, since this messages of a burst batch trigger different "plots filter passed", different plots).
Discussion mentioned got exactly zero replies, but now my investigation came far enough to warrant a bug. It looks not right at all, and it also looks like I have no way to influence it, multiple new_signage_point (queries to check my plots) just happen to come in bursts sometimes. I would think it's signage points I've missed, but it's not the case, timings do not match. It's just some extra malicious signage points.
Please investigate and help. Most probably this is happening not just for me.
The text was updated successfully, but these errors were encountered: