Comment 12 for bug 1519473

Not sure how related this is, but:

One of the customers had very large juju database - around 15-17GB was the disk footprint. The txns collection was using around 9GB - this was juju 1.20.

mongod on that deployment was using around 10GB of memory, so was jujud - this behavior was observed on all state servers (they are running juju-ha).

Once we purged unreferenced transactions from the txns collection (and reduced the size of txns-queue fields in various documents), and run repairdatabase (so that the disk footprint is smaller - the database was shrunk to cca 2.6GB), both jujud and mongod are using around 300-500 megabytes of RAM.

Mind you, this is on juju v1.20 - these bugs should not be present in 1.25, but it might be wise to check the size of the database, and particularly the txns collection.

One can use the 'unofficial' mgopurge tool to clean up juju database: https://github.com/mjs/mgopurge

Of course, make sure you have a backup of the database prior running the tool!