TPS falliing due to GC_active (native) taking too long
I've been experiencing quite some issues regarding tps instability on my server. recently i've been able to capture the issue while profiling the server:
GC_active getts stuck on libjvm.so.ThreadSafepointState::handle_polling_page_exception()
however i am not quite sure how to procede mitigating this issue.
I believe it is correlated to t:computerfire: :computerfire: ps drops due to its mspt being inversely proportional to the server's.
32 Replies
Thanks for asking your question!
Make sure to provide as much helpful information as possible such as logs/what you tried and what your exact issue is
Make sure to mark solved when issue is solved!!!
/close
!close
!solved
!answered
Requested by aikodev#0
here's some extra information on the server's performance itself
send the spark
ur using contabo
or g-portal
thats most likely the reason u have these issues.
Yes I am using contabo, but how does that cause the issue
they oversell a shit ton
spark
spark is a performance profiler for Minecraft clients, servers, and proxies.
❌ Processing Error
The bot cannot process this Spark profile. Please use an alternative Spark profile.
Requested by aikodev#0
and their hardware isnt good
I did realize that when I saw the CPU data...
The biggest issue regarding other providers is that I gotta keep it affordable
a cheap trusted mc host is gonna be alot better
The issue is that "cheap" mc Hosts Go for twice the price for half the stuff
At least most I've seen
what hosts?
I'd guess exaroton, some other stone mc Thing I can't remember either
exaroton is horrrrible
and pricing isnt good at all
Been working with contabo for almost two years so can't remember most pricing data
the billing system is ig nice, but usually costs more for a day or two of exaraton than a whole month at a good budget one
And it's exactly due to a tight budget that I work with contabo
whats ur budget
It's easier for me to deal with doing stuff myself than having a hole on my pocket
well with contabo all u get is oversold hardware
I dare say 10$/mo for 5+ people modded
that can easily be done on a budget host
Pebblehost or enviromc
or u could go oracle cloud
if u want free
The server needs to run well under a rather heavy load so I'd have to study that further
You've got any idea what could be causing that issue in special though
Because regardless of host, I need it fixed ASAP
And due to the logistics of it, we might not be able to switch hosts for a while
:contabo:
the memory is oversold
u could probably even be running on swap
I'll research into the issue
Honestly, i have absolutely no reason to believe that is the cause of the issue, since the server goes right back to its normal stable behavior on a restart
if it were a vps allocation issue, it shouuld stay regardless of jvm restart
however it is certain that something is causing the gc thread to fail, which is causing the issue at hand
Because it’s not going into swap
I'll disable swap entirely and see if the issue persists or not then
thanks
having read a bit on how swap behaves on high consumption apps like minecraft, it'd be reasonable to believe that swap is at fault here
tl;dr never activate swap on a game server i suppose
i'll wait to test the server under heavy load over a 24h period. if the issue doesn't manifest again after i remove access to swap, i'll mark it as solved
swap is only good for when your server is about to OOM
an issue which i can mitigate way before it happens
unlike the issue of swap getting burnt out due to possible memory leaks and slowly killing the tps
Seems that revoking access to swap has not affected the instability whatsoever
Do an allocation profiler to see what's allocating so much memory
To cause your GCs
Haven't been able to reproduce the issue in a while after readjusting JVM flags and removing swap, so I believe it was indeed a swap related issue
I've readjusted the flags to have the minimum allocated heap size at around the same size the server uses on lower load, so 3072M
Will now mark this issue as solved
Thank you, @snowznz.bak, for pointing out the issue
!solved
post closed!
The post/thread has been closed!
Requested by aikodev#0