Query of one large (7MB) record by primary key from MongoDB runs CPU 100% for 5 seconds

Question

I have a replication set of three mongod processes running on my laptop (i7, SSD, 16GB RAM). I create an empty database and add a single 7MB record to it. I then query that record out from the command line:

echo "db.items.find({_id:'.......'})" | mongo mydb >tmp/junk

The client application "mongo" (not mongod or mongos) takes 100% CPU for several seconds before spitting out the response. If I read the record out using the Java client for MongoDB, my JVM process uses 100% CPU for several seconds before giving the response.

What could be going on here? How can I troubleshoot this?

Update: The replication set seems to be irrelevant; if I connect mongo directly to the master mongod, the performance is the same.

Update: The PHP client can get the data in some fraction of a second. Wha...?

Answer 1

Sounds like it takes mongo several seconds to decode the document's BSON into a Javascript object in C++, and several seconds for the Java driver to decode the BSON into a Java object. You could use mongodump to dump that one document to a bson file using mongodump's --query argument, then try:

time bsondump file.bson > /dev/null

That'll give you a basis for comparison: the time it takes a C++ program to decode the BSON document and format it as JSON. If that's lengthy, then the sheer size of the document is the main cost.