[performance] loadMissedMessages increased CPU load on all instances

Description:

Today I did a manual LDAP sync and during that sync all RC instances increased CPU and network load.
In that same time sporadically increased amount of loadMissedMessages method requests.
It look like some bug in server-client intercommunion (but maybe I'm wrong here)

Steps to reproduce:

Unknow

Expected behavior:

Normal load during rush hours

Actual behavior:

Server Setup Information:

• Version of Rocket.Chat Server: 3.16.3
• Operating System: CentOS7
• Deployment Method: docker
• Number of Running Instances: 20
• DB Replicaset Oplog: Emabled
• NodeJS Version: 12.22.1
• MongoDB Version: 4.2.15

Client Setup Information

• Desktop App or Browser Version: 3.2.3 and Chrome 93
• Operating System: Windows 10

Additional context

No

Relevant logs:

There was LDAP connection error, I don't know if it can be related

September 14th 2021, 09:30:59.000	s-rc-srv-03	server.js:204 LDAP ➔ Connection.error connection Error: read ECONNRESET  errno: 'ECONNRESET',  code: 'ECONNRESET',  syscall: 'read'}
September 14th 2021, 09:30:49.000	s-rc-srv-03	rocketchat_rocketchat_4_1	server.js:204 LDAPSync ➔ error MongoError: cursor id 7274881890252097063 not found  operationTime: Timestamp { _bsontype: 'Timestamp', low_: 400, high_: 1631601048 },  ok: 0,  code: 43,  codeName: 'CursorNotFound',  '$clusterTime': {    clusterTime: Timestamp { _bsontype: 'Timestamp', low_: 402, high_: 1631601048 },    signature: { hash: [Binary], keyId: [Long] }  }}

No source file, test, or entry point is named. Start by reviewing the loadMissedMessages requests and LDAP sync logs around the reported CPU and network spike; done requires reproducing the issue, identifying the cause, and confirming that synchronization no longer creates abnormal load.