design - Collection data modeling in mongoDB -

i want design model profiles interaction, example <-> interact <-> b , interaction contains common fields , b.
lets have collection called interactions, have few thoughts in mind , looking best practice solution.

1. separate interaction 2 different documents, 1 each profile
   

   {   pid:"a id"   commonfield1:""   commonfield2:""   ..  } {   pid:"b id"   commonfield1:""   commonfield2:""   .. } 

   pros: fast read
   cons: each update common field should performed on both of documents

2. maintain 1 document interaction

   {  pids:['a id','b id']  commonfield1:""  commonfield2:""  .. } 

   pros: update common field once
   cons: tricky read

the thing there lot of reading lot of updating , collection should designed lot of millions of documents.

common queries in scenarios:

• retrieve profile interactions
• update specific profile interaction

i leaning second choice relying on multikey index on pids fast document lookup , enjoying in single update in each frequent change.

i have no experience in sharded collections have noticed multikey index not supported sharding key, should show stopper second choice?
does reads fast enough kind of index? , other choices use case?

your answer highly appreciated.

i think latter format makes more sense avoid duplication of updates.

for interaction pairs should using compound index rather array. compound index can be used _id , shard key (arrays not valid either).

so document might like:

{     _id: { pid1: 'a', pid2: 'b' },     commonfield1: '',     commonfield2: '', } 

if want avoid duplicate pairs sort ids in predictable order. example, pid1 might lesser of 2 values.

the default _id index allow efficiently either (pid1,pid2) or (pid1) interactions, you'll want add index on {'_id.pid2': 1}.