add index on juju.relationscopes

Hi,

I was looking into our jujud/mongodb resources issues at the end of last week when I spotted a bunch of similar COLLSCANs, that sometimes got run ~1000 times / min, for example :

Mar 24 04:19:13 juju-4da59b22-9710-4e69-840a-be49ee864a97-machine-2 mongod.37017[10287]: [conn9272] command juju.relationscopes command: find { find: "relationscopes", filter: { key: { $regex: "^r#5#spark/3#provider#" }, departing: { $ne: true }, model-uuid: "282da984-55ea-4c54-832d-dc1f57c4afcb" }, skip: 0 } planSummary: COLLSCAN keysExamined:0 docsExamined:8717 cursorExhausted:1 keyUpdates:0 writeConflicts:0 numYields:68 nreturned:1 reslen:387 locks:{ Global: { acquireCount: { r: 138 } }, Database: { acquireCount: { r: 69 }, acquireWaitCount: { r: 2 }, timeAcquiringMicros: { r: 20436 } }, Collection: { acquireCount: { r: 69 } } } protocol:op_query 109ms

I spoke to axw, and he said that "those queries happen when workers in juju start watching relations". We ended up adding the following index :

db.relationscopes.createIndex({"key": 1, "departing": 1, "model-uuid": 1})

I'm filing this bug to make sure we don't forget making juju add this index automatically.

Thanks !