Description

Hello!

I just investigated undocumented behavior of unordered bulk execution.

simple example (Node.js):

// 1. initialize
const bulk = initializeUnorderedBulkOp();

// 2. populating data
bulk.insert(doc1);
bulk.insert(doc2);
bulk.insert(doc3);
...
bulk.insert(doc10000);

// 3. execution
bulk.execute().catch(error => {
   // ignore duplicete errors because this decision works faster then find().upsert().update()
   // yes, it's crutch, but it works :)
   if (error.code === 11000) {
     return;
   }
   throw error;
});

So... Sometimes in the bulk object I see more then single batch (If it was large source in input). And if some batch has any document duplicate unique key - following batches won't be processed.

Right now i got three batches. First of them has 8547 operations. In bulkResult I find 7 errors (duplicates). So after execution the target collection had increased for 8540 documents. Usually I use 'getWriteErrors()' to collect problems but (surprize), operations from second and next batches didn't have errors because hadn't process. 

I tried to use 'continueOnError' flag, but it didn't has any effect.

Now I use 4.2.7 version but abut month ago I had upgraded from 3.x. I dont't remember this problem before upgrade, but after quick analyse of result collections I may think that this problem can be present in previous versions... 

Scope of changes

Impact to Other Docs

MVP (Work and Date)

Resources (Scope or Design Docs, Invision, etc.)