Closed passionofvc closed 5 years ago
Where do you see the high memory usage, or rather... what do you consider "high"?
hi @MarkusTeufelberger last htop picture show 7.24G/7.64G, centos server 8G memory, and only run rippled daemon .
Set the node_size to tiny or use more RAM.
yes, rippled
is very hungry for memory.
even if you set node_size = tiny
it eats a lot of memory
also it loves to rape your I/O pretty much, so be prepared that your SSDs/HDDs will be constantly busy.
here is what I got with node_size = tiny
(VM is 16GB):
20828 ripple 20 0 9773.8m 8.579g 697320 S 15.7 53.6 18:42.80 rippled: main
New rippled v0.90.0
stopped working reliably (constant 100% cpu load and i/o) with RocksDB
engine with the same settings v0.81.0
worked.
I've switched to database engine NuDB
for now, but from iotop and top I can see that rippled is constantly using all available i/o there. Is there anything can be done about that?
seems a new version doesn't work at all on HDD.
after running for 24 hours it got stock with
{
"result" : {
"error" : "noNetwork",
"error_code" : 17,
"error_message" : "InsufficientNetworkMode",
"request" : {
"account" : "-",
"command" : "account_info",
"ledger_index" : "validated",
"method" : "account_info"
},
"status" : "error"
}
}
Rippled is with constant load 400% cpu (all cores) and 16GB memory.
The node is tiny
and uses NuDB
as you recommend.
[node_size]
#
# Tunes the servers based on the expected load and available memory. Legal
# sizes are "tiny", "small", "medium", "large", and "huge". We recommend
# you start at the default and raise the setting if you have extra memory.
# The default is "tiny".
tiny
[ledger_history]
256
[node_db]
type=NuDB
path=/home/ripple/.ripple/db/nudb
online_delete=2000
advisory_delete=0
I've downgraded to 0.81.0
for now and RocksDB which works just fine.
Could you look into it @wilsonianb ?
there is also another claim from some user - https://groups.google.com/forum/#!topic/ripple-server/4O3hk-OnMI0
for me v0.90.0
is not usable at all on hdds
NuDB can't be used on HDDs.
@MarkusTeufelberger as I said before new version v0.90.0
doesn't work with RocksDB
for some reason properly.
Constant load and high memory usage.
Is there any possibility the update broke something related to RocksDB
engine?
I've checked debug.log there are lots of lines like:
2018-Feb-27 02:59:33 Validations:WRN Unable to determine hash of ancestor seq=1 from ledger hash=5A944118AF413EEAE093C3BE15FE9657D144F4D0203A7AD2EEFAF8CB11A2AC33 seq=36855529
2018-Feb-27 02:59:33 Validations:WRN Unable to determine hash of ancestor seq=1 from ledger hash=5A944118AF413EEAE093C3BE15FE9657D144F4D0203A7AD2EEFAF8CB11A2AC33 seq=36855529
2018-Feb-27 02:59:33 Validations:WRN Unable to determine hash of ancestor seq=1 from ledger hash=80371445534F105BB12ED60CC5340C8BCB15F0DF89C36A5B0A93804044FCE657 seq=36860383
.....
On the version 1.0.1, ripled still eat more than 10Gb memory
There are several improvements that are planned to reduce the amount of memory used. Stay tuned under "Issues" for a new issue.
I run rippled with testnet for two week, everything is ok, but the memory is very in high usage. Does this is normal behavior for using memory . I set online_delete=2000, not other network error, and rpc response worked fine.
OS :centos7 mem:8G HDD:150G