Eos-server-5.0.2-1.el7 and new failover mechaninms

Dear Elvin and Andreas

Do you have a mini-guide or documentation ( beyond Master/Slave QuarkDB Configuration — EOS CITRINE documentation) for the setup of the new failover mechanisms amongst the
different MGM ?

thank you in advance
best
e.v.

HI Emmanouil,

We’ll try to add a section on this in the documentation, but to be honest there is not much to configure. You just need two MGMs using the same QDB backend and make sure you have the following env variable set for both EOS_USE_QDB_MASTER=1. That should be it.

Cheers,
Elvin

hello Elvin
can I make this with 3 MGM nodes ?
they is a need ? for a (simply) round-robin dns conf like

eostest.grif.fr has address 192.54.206.179
eostest.grif.fr has address 134.158.132.132
eostest.grif.fr has address 134.158.72.173

FYI
best

e.v.

You can do with as many MGMs as you want. The DNS alias needs to point to the current master. This will be improved (relaxed) soon with some functionality from the XRootD 5 client, but it’s not yet in the 5.0.4 release.

Cheers,
Elvin

hello Elvin
thank you for your reply
I fix the path in dynamic libraries
(e.g. mgmofs.macaroonslib /opt/eos/xrootd/lib64/libXrdMacaroons.so /usr/lib64/libEosAccSciTokens.so)
and now the authedication works nice for xroot or https protocol

a) for the quarkdb still I use the general xroot5 executable
/usr/bin/xrootd -l /var/log/quarkdb/xrootd.log -c /etc/xrootd/xrootd-quarkdb.cfg -k fifo -s /var/run/quarkdb/xrootd.pid
I had some issues with the eos xrootd ( did not start, complains for the number or arguments)

b) for the failover dns, could you send the script that you use in order to tag a mgm as Master for the dns failover ?
c) they is a simple way to get the master direct from quarkdb with redis-cli ?

I need to know which is the master in order to feed the DNS load balance with correct server
thank you in advance
best regads
e.v.
p.s.
now I use something like
status=/usr/bin/timeout 10 /usr/bin/eos root://localhost:1094 ns|grep "^ALL Replication"
but I would like not to use eos cli command ( migth be complecate for periodic cron job)

Hi Emmanouil,

Glad to hear that things are working fine for your and really sorry for the late reply on this! Let me answer your questions one by one:

a) We did not have any issues with this. Could you let me know exactly what is the problem and maybe I can provide some advice?

b) We don’t have yet such a script as most of the instances are still running EOS4 and the ones moved to EOS5 are not yet handling the DNS changes.

c) Yes, you can use the following call to determine which is a master and which not:

$ xrdfs root://eosaliceo2-ns-01.cern.ch// query opaquefile "/?mgm.pcmd=is_master"                                                                                                                                                                     
OK
$ xrdfs root://eosaliceo2-ns-00.cern.ch// query opaquefile "/?mgm.pcmd=is_master"
[ERROR] Server responded with an error: [3011] Unable to find master file [ENOENT] ; No such file or directory

The first response is from an MGM master and the second from an MGM slave.

Let me know if you need more info.

Cheers,
Elvin