avoid converting binary document to tree when possible on later mongo driver versions

[richardstartin]

Newer mongo driver versions allow more efficient reading of BSON documents. Unfortunately this isn't possible on older driver versions, so I added some instrumentation to extract the document's buffer and do the same thing manually.

[richardstartin]

This makes each BsonDocument 8 bytes larger, so may not be worth it

org.bson.BsonDocument object internals:
 OFFSET  SIZE               TYPE DESCRIPTION                               VALUE
      0    12                    (object header)                           N/A
     12     4      java.util.Map BsonDocument.map                          N/A
     16     4   java.lang.Object BsonDocument.__datadogContext$4           N/A
     20     4                    (loss due to the next object alignment)
Instance size: 24 bytes
Space losses: 0 bytes internal + 4 bytes external = 4 bytes total

But is probably ok for subclasses

com.mongodb.connection.ByteBufBsonDocument object internals:
 OFFSET  SIZE               TYPE DESCRIPTION                               VALUE
      0    12                    (object header)                           N/A
     12     4      java.util.Map BsonDocument.map                          N/A
     16     4   java.lang.Object BsonDocument.__datadogContext$4           N/A
     20     4   org.bson.ByteBuf ByteBufBsonDocument.byteBuf               N/A
Instance size: 24 bytes

[bantonsson]

So what do you think that the average retained weight of a BsonDocument is? Won't the 8 bytes be dwarfed by the Map and its contents?

[richardstartin]

@bantonsson yes, and the majority of them (in my experience with this driver) tend to be the ones we'll instrument (the ones sent to and from the database) which have alignment shadow to absorb the field. The problem is that for the document type where this adds eight bytes, what you're offsetting it against also contains BsonDocuments... It's possible this will cause small problems with mapping frameworks using a little more memory.

[tylerbenson]

If this api is only available in newer versions, why didn't muzzle require an updated version range?

[richardstartin]

If this api is only available in newer versions, why didn't muzzle require an updated version range?

@tylerbenson the API is available in v4 of the driver (we have two mongo instrumentation modules)

[tylerbenson]

Ok. I was confused what you were referring to since you made changes to both 3.x and 4.x modules.
Now I understand better though. You're taking advantage of the new API in 4.x and adding instrumentation to expose similar behavior to 3.x since it doesn't have the needed API.
👍

[richardstartin]

Yes the changes were made in separate commits