In a replica set setup (1 primary, 1 secondary, and 1 arbiter) of MongoDB (4.4), the number of ReplicationStateTransition locks is too high (about 3000 locks per second).
This is negatively affecting the performance of the aggregation queries as the number of locks (and also the yields are too high).
Ref.:
{
"locks": {
"Mutex": {
"acquireCount": {
"r": 1
}
},
"Global": {
"acquireCount": {
"r": 1225
}
},
"ReplicationStateTransition": {
"acquireCount": {
"w": 1225
}
},
"Database": {
"acquireCount": {
"r": 1225
}
},
"Collection": {
"acquireCount": {
"r": 1225
}
}
},
"numYields": 1224
}
In the execution plan of the aggregation queries (to fetch just 5000 documents), I can see that proper indexes are getting used for filtering. However, the index created does not cover the query (projected fields are not a part of the index).
Can someone guide me in terms of why MongoDB locks so frequently (specifically around ReplicationStateTransition locks)? And, what could be done to avoid such locks?