Hello,
Is it possible to choose wich node can be elected or not for the backup.To be more clear, in my replicaset, I have some big hardware server and some small.I want launch the backup only on big server because performances are too low on small server.Thanks for your help.PS: @AndrewPogrebnoi I did not try the restore with 1.2 to see if memory problem is always here.
Hi @j2geu No, currently there is no mechanism to appoint some particular node to do the backup or not. Now nodes decide among themselves which will take the job. The decision is being made in favour of a secondary node with proper replication lag. The user can’t influence this decision.
Speaking of memory issues. We made some tests and it appears that pbm-agent needs about 1.3-1.5Gb of RAM which is needed for decompression buffers. You can disable backup compression by running pbm backup --compression=none it’ll dramatically reduce memory consumption for restoring of such backup (to ~100Mb) but you’re going to pay with increased disk space and network load.
We expect a pbm-agent to be installed 1-to-1 with every mongod node, but that is to provide failover. Only one node per replicaset does a given backup operation.
If you deliberately shutdown a pbm-agent on server A then ones on B or C will do the backup operation. For preference not the one on a primary node, but if there is only one pbm-agent alive for a replicaset and it’s connected to the primary node, then that one. The behaviour is not configurable.
I hesitate to suggest a way to do it because I fear it would split the user population into two groups - one who to set it regardless of whether the pbm-agent is on the primary node or not, and ones who only want it to affect which secondary is more likely to be used.