Conditional grouping with $exists inside $cond

I have two keys A and B and their existence in the document is mutually exclusive. I have to group by A when A exists and group by B when B exists. So I am $projecting the required value into a computed key called MyKey on which I'll perform a $group. But it looks like I'm making a mistake with the syntax. I tried writing $project in two ways:

{$project: {MyKey: {$cond: [{$exists: ["$A", true]}, "$A", "$B"]}}}

and

{$project: {MyKey: {$cond: [{"A": {$exists:true}}, "$A", "$B"]}}}

But I keep getting the error:

{ "errmsg" : "exception: invalid operator '$exists'", "code" : 15999, "ok" : 0 } ...

What's going wrong?

Use $ifNull instead of $cond in your $project:

{ $project: {MyKey: {$ifNull: ['$A', '$B'] }}}

If A exists and is not null its value will be used; otherwise the value of B is used.

if one wants to check $exists with in $cond an alternative approach is to use $not with $cond

{$project: {MyKey: {$cond: [{$not: ["$A"]}, "$B", "$A"]}}} 

and truth table for $not is as

Hopes that Helps

You can simulate exists with

$ne : [$var_to_check, undefined]

This returns true if the var is defined

I found your questions while looking for a similar problem, but insted of a key, I was looking for my parameters. I finally solved the issue.

This is what I used for my $_id.status parameter, to check that if it exists inside the cond.

$cond: [{
     $or: [{
          $ne: ["$_id.status", null]
     }]
}, 1, null]

$or is not needed. I keep it there... just for fun. I don't think it affects the query that much for the moment. I will test the speed later.