Performance drop when node goes down in repl. set

Ops and Admin
1

Hi,

I have 3 node setup (2 data nodes + one arbiter). That works fine, secondary is promoted when primary fails. However, when one data node is down, there is sudden drop in performance of write operations. from around 15.000 updates/sec to exactly 100. I’m using write concern of 1. The only thing different I see in MongoDb compass is that after one data node is down QWRITES start to appear (in performance tab). Any help is welcomed.
In 3 data node setup performance stays the same. Am I missing anything obvious (as google search yields no result about this)?

All nodes v6.0.5 running on linux.

2

Apparantly the issue was caused by flow control. Since there are no data nodes that can keep up with replication, primary (only data node now) will throttle request so replicas are not too far out of sync. Thus suspiciously “round” number of requests per second (100).

changing flow control helped:
db.adminCommand( { setParameter: 1, enableFlowControl: false } )

3

Yeah, by default flow control is enabled, however i’m not very sure how it works with an arbiter only (when your secondary node is down).

Generally using an arbiter is not recommended unless really needed.