MongoDB, WiredTiger: Need to update one field in all documents (700k). Which side-effects will occur?

Question

One of the collections has documents with a field "sc":[float]. I want to iterate all documents and change this to "sc": float, that is, remove the array and assign the float value directly to the key.

Isn't it true, that WiredTiger appends all updated documents instead of trying to do some in-place updates like MMap?

This would basically double the size of the database, with the initial half being obsolete data.

Do I need to call mongod -repair in order to discard those obsolete documents, or is there something else I need to do?

Answer 1

WT is always rewriting documents when updated, but instead of NMAP it is not using padding, and allocating documents in free blocks, so if we have document 1,2,3 - there is a chance that no1 will be moved to end of a file (or closest gap), and no2 will be allocated in previous no1 space.

what is important about WT:

WiredTiger does ongoing compaction and reuse of space automatically.

To ensure that file is used efficiency we can enforce WT to use smaller page size as per this discussion.

--wiredTigerCollectionConfigString="leaf_page_max=8KB"

Also compact could be use to reclaim space more here.