Open Smithx10 opened 6 years ago
verified that im getting the same behavior on Ubuntu LX
Same behavior on my Ubuntu and Debian LX
Partly related to https://github.com/iputils/iputils/issues/129
@Teutone I saw there is a PR opened on August 16, is there anything stopping it from getting merged?
I don't think so - just normal merge process, I don't know how long it takes them. I tried the changes in lx-brand instances (self-build ping from iputils with mentioned fix) and there were no issues. Working as expected.
I'm also seeing this kind of error.
socket.c setsockopt(20, IP_RECVTOS) failed: Protocol not available
There are various issues here. E.g., we have no support for the IP_RECVTOS
option in the native kernel, but adding support for it probably wouldn't be too bad. As for the other issues: I've have to fire up some lx zones and look more closely. In any event I'll assign this to myself and see if I can't at least file proper issues away and maybe even fix some of them.
Quick Update,
iperf3 has stopped working for me in CentOS LX.
[root@7abb8f3f-697d-4846-f348-c88cf605257f ~]# iperf3 -c xx.xx.xx.xx Connecting to host xx.xx.xx.xx, port 5201 iperf3: error - unable to set TCP_CONGESTION: Supplied congestion control algorithm not supported on this host
On paltform SunOS ac-1f-6b-a5-af-5a 5.11 joyent_20190718T005708Z i86pc i386 i86pc
Using this image: 3dbbdcca-2eab-11e8-b925-23bf77789921
@Smithx10 - did you recompile iperf3 or update it from pkgsrc/other-upstream? I've seen (mis)behavior like this when a program assumes existence of X implies existence of Y. In your case Y == TCP_CONGESTION, but it's not at all clear what X is.
@danmcd
I used
yum install -y iperf3
In centos lx. Image 3dbbdcca-2eab-11e8-b925-23bf77789921.
Okay, so TCP_CONGESTION did show up v. recently. I'm curious about what parameter was passed for the LX socket option? Maybe Linux does that differently than us (we use a "struct cc_algo") and we have to account for that?
I receive the following error as well when attempting to start the bind
service on LX CentOS 7 (3dbbdcca-2eab-11e8-b925-23bf77789921
):
setsockopt(519, IP_RECVTOS) failed: Protocol not available
I receive the following error as well when attempting to start the
bind
service on LX CentOS 7 (3dbbdcca-2eab-11e8-b925-23bf77789921
):setsockopt(519, IP_RECVTOS) failed: Protocol not available
Sounds like a new issue. Please file and link here?
Sounds like a new issue. Please file and link here?
This is also happening to php-fpm at line https://github.com/php/php-src/blob/df4c27642efb2ee986ad7fb744f76ab380cf5a4c/sapi/fpm/fpm/fpm_sockets.c#L501
We need TCP_INFO. This is an illumos issue too...
https://www.illumos.org/issues/14744 tracks this. No guarantees on whether or not this makes it upstream. Once it does, we'll need to plumb it straight into LX. Anyone who's interested in tackling this should try and get it ready for -gate first, but if you wish to put it straight into illumos-joyent, I expect LX plumbing to arrive with it.
@Adel-Magebinary FWIW, php-fpm
works well under SmartOS LX if you configure it to communicate with the web server via a Unix domain socket (i.e. a filesystem socket, rather than an inet socket).
In the php-fpm
config:
[somePoolName]
listen = /var/run/php-fpm/somePoolName
listen.backlog = 256
…
In the httpd
config:
<VirtualHost …>
<Directory …>
SetHandler proxy:unix:/var/run/php-fpm/somePoolName|fcgi://localhost/
…
@smokris, thanks for that. For some single-used / locally used fpm services, we used the socket instead of TCP. However, in our case, we need fpm to be available for other nodes.
eg
upstream m2_fastcgi_backend {
ip_hash
server unix:/var/run/php-fpm.sock;
server fpm-01.cns backup resolve;
server fpm-02.cns backup resolve;
server fpm-03.cns backup resolve;
}
whereas fpm-01.cns needs to listen to TCP.
This also happens in MySQL clusters.
2022-06-17T04:08:08.424303Z 0 [Note] [MY-000000] [WSREP-SST] Cannot open netlink socket: Protocol not supported 2022-06-17T04:08:08.424355Z 0 [Note] [MY-000000] [WSREP-SST] Cannot open netlink socket: Protocol not supported 2022-06-17T04:08:08.424388Z 0 [Note] [MY-000000] [WSREP-SST] Cannot open netlink socket: Protocol not supported 2022-06-17T04:08:08.657658Z 0 [Note] [MY-000000] [WSREP-SST] Cannot open netlink socket: Protocol not supported 2022-06-17T04:08:08.659807Z 0 [Note] [MY-000000] [WSREP-SST] Cannot open netlink socket: Protocol not supported 2022-06-17T04:08:08.659869Z 0 [Note] [MY-000000] [WSREP-SST] Cannot open netlink socket: Protocol not supported 2022-06-17T04:08:08.660141Z 0 [Note] [MY-000000] [WSREP-SST] Cannot open netlink socket: Protocol not supporte
@Adel-Magebinary netlink sockets are a whole other world, and LX isn't necessarily designed to keep up there.
I wonder if MySQL (or mariadb) via pkgsrc on a native zone might suit you better? It is available via pkgsrc...
mysql-client-5.6.51nb1 MySQL 5, a free SQL database (client)
mysql-connector-c++-8.0.25 Standardized MySQL database driver for C++ development
mysql-server-5.6.51nb1 MySQL 5, a free SQL database (server)
I'd recommend percona-cluster on native, it's kept reasonably up-to-date and supports the wsrep/sst stuff.
Seems the following commands are behaving strangely on LX. nmap nslookup ping iperf3
I decided to test them all on the following lx branded zone images.
Platform:
Images:
VM's running those images:
The Debian Behavior:
The Ubuntu Behavior:
The CentOS Behavior
The Alpine Behavior