Multicast packets


(Szymon Gwóźdź) #1

Hi!

When I'm launching ElasticSearch I get logs like this:

[13:48:24,418][WARN ][jgroups.UDP ] send buffer of socket
java.net.DatagramSocket@46b29c9d was set to 640KB, but the OS only allocated
167.77KB. This might lead to performance problems. Please set your max send
buffer in the OS correctly (e.g. net.core.wmem_max on Linux)
[13:48:24,418][WARN ][jgroups.UDP ] receive buffer of socket
java.net.DatagramSocket@46b29c9d was set to 20MB, but the OS only allocated
524.29KB. This might lead to performance problems. Please set your max
receive buffer in the OS correctly (e.g. net.core.rmem_max on Linux)
[13:48:24,418][WARN ][jgroups.UDP ] send buffer of socket
java.net.MulticastSocket@7846a55e was set to 640KB, but the OS only
allocated 167.77KB. This might lead to performance problems. Please set your
max send buffer in the OS correctly (e.g. net.core.wmem_max on Linux)
[13:48:24,418][WARN ][jgroups.UDP ] receive buffer of socket
java.net.MulticastSocket@7846a55e was set to 25MB, but the OS only allocated
524.29KB. This might lead to performance problems. Please set your max
receive buffer in the OS correctly (e.g. net.core.rmem_max on Linux)

How can I change this buffer in java (not in system, I don't want to change
system values)? I've tried with something like this:

discovery:
jgroups:
config: udp
mcast_send_buf_size: 10000
mcast_recv_buf_size: 10000
ucast_send_buf_size: 10000
ucast_recv_buf_size: 10000

but it didn't work (i had the same values in logs as before) :frowning:

I have to get rid of these warnings, because network in my company is
rejecting multicast messages with too many packets - admins sent me logs
like this:

Discard IP fragment set with more than 24 elements: src = 10.1.112.214, dest
= 228.8.8.8, proto = udp, id = 18373

And probably because of this after few minutes of feeding elasticsearch with
data I have logs like this:

[12:28:16,346][WARN ][jgroups.pbcast.NAKACK ] test-4-16470: dropped
message from test-5-22451 (not in xmit_table), keys are [test-4-16470,
test-3-24681, test-2-34614, test-1-48736], view=[test-1-48736|9]
[test-1-48736, test-3-24681, test-4-16470, test-2-34614]


(Shay Banon) #2

This values are recommended values for jgroups when using multicast. You can
move to use unicast if you don't want to use multicast or change the values
on the os level.

cheers,
shay.banon

On Tue, Apr 20, 2010 at 3:15 PM, Szymon Gwóźdź sz.gwozdz@gmail.com wrote:

Hi!

When I'm launching ElasticSearch I get logs like this:

[13:48:24,418][WARN ][jgroups.UDP ] send buffer of socket
java.net.DatagramSocket@46b29c9d was set to 640KB, but the OS only
allocated 167.77KB. This might lead to performance problems. Please set your
max send buffer in the OS correctly (e.g. net.core.wmem_max on Linux)
[13:48:24,418][WARN ][jgroups.UDP ] receive buffer of socket
java.net.DatagramSocket@46b29c9d was set to 20MB, but the OS only
allocated 524.29KB. This might lead to performance problems. Please set your
max receive buffer in the OS correctly (e.g. net.core.rmem_max on Linux)
[13:48:24,418][WARN ][jgroups.UDP ] send buffer of socket
java.net.MulticastSocket@7846a55e was set to 640KB, but the OS only
allocated 167.77KB. This might lead to performance problems. Please set your
max send buffer in the OS correctly (e.g. net.core.wmem_max on Linux)
[13:48:24,418][WARN ][jgroups.UDP ] receive buffer of socket
java.net.MulticastSocket@7846a55e was set to 25MB, but the OS only
allocated 524.29KB. This might lead to performance problems. Please set your
max receive buffer in the OS correctly (e.g. net.core.rmem_max on Linux)

How can I change this buffer in java (not in system, I don't want to change
system values)? I've tried with something like this:

discovery:
jgroups:
config: udp
mcast_send_buf_size: 10000
mcast_recv_buf_size: 10000
ucast_send_buf_size: 10000
ucast_recv_buf_size: 10000

but it didn't work (i had the same values in logs as before) :frowning:

I have to get rid of these warnings, because network in my company is
rejecting multicast messages with too many packets - admins sent me logs
like this:

Discard IP fragment set with more than 24 elements: src = 10.1.112.214,
dest = 228.8.8.8, proto = udp, id = 18373

And probably because of this after few minutes of feeding elasticsearch
with data I have logs like this:

[12:28:16,346][WARN ][jgroups.pbcast.NAKACK ] test-4-16470: dropped
message from test-5-22451 (not in xmit_table), keys are [test-4-16470,
test-3-24681, test-2-34614, test-1-48736], view=[test-1-48736|9]
[test-1-48736, test-3-24681, test-4-16470, test-2-34614]


(Szymon Gwóźdź) #3

Yes, but could you explain me why my discovery.jgroups config isn't working?

Regards
Szymon Gwóźdź


(Shay Banon) #4

Yea, its because they are (sadly) not configured to be injected using system
properties (which is what I do when you pass them in the elasticsearch.yml
configuration). You can create your own copy of the udp.xml file, and place
it under config/jgroups location, and change it.

cheers,
shay.banon

2010/4/21 Szymon Gwóźdź sz.gwozdz@gmail.com

Yes, but could you explain me why my discovery.jgroups config isn't
working?

Regards
Szymon Gwóźdź


(system) #5