Looking at changing low_watermarks for a customer and noticed a few things.
Statistics for these show up in bytes from the command line:
[root@Config349VM1 bin]# ./mbstats localhost:11211 all|egrep "ep_mem|mem_used"
Are set in megabytes with flushctl:
[root@Config349VM1 bin]# ./mbflushctl localhost:11211 set mem_low_wat 1480
setting flush param: mem_low_wat 1480
and show up as gigs in the web console.
When setting low_wat, the number itself seems to be a aggregate of the cluster capacity. For a smaller cluster this is a more simple operation but for a large cluster with lots of memory this can be a bit confusing.
Can we change mbflushctl to take a % instead of a megabyte value?