yagna
yagna copied to clipboard
Provider agent memory leak
Investigate and fix a memory leak in the provider agent.
host: office_8_67
net: devnet-beta
Max 1 day for investigating
This is still happening, although so far only observed on devnet-beta providers, running yagna 0.10.0.
I suspect this is related to #1925.
I saw this correlation here: https://github.com/golemfactory/golem-internal/issues/497
and here:
The rate of memory leaking is quite different, but in both cases I can see it's burning 100% CPU (one vcore), while looping on an event.
The issue #1925 is already resolved, but remaining question is: can someone intentionally generate a lot of events to perform a DoS attack?
This issue was not related to someone generating a lot of events, but to problem with sqlite timestamps comparison, which caused this effect. So even if it is possible to cause DoS in some way, it is not by exploiting problems discussed here.
Can you confirm, that this issue isn't happening anymore?
I haven't noticed it in recently in hybrid network (it's not a 100% firm "no", but it's also not a "yes, it's still happening"). Some nodes are filling their disk space, but due to smoltcp::socket::tcp
and ya_net::hybrid::service
debug messages.
Edit: also provider_datadir/exe-unit/work
takes a lot of space.