Conditional upsert in MongoDB
design-patterns, mongodb
Solution
Solution below should do the trick, you just need to ignore dup key errors. Example is given in Mongo shell:
> var lastUpdateTime = ISODate("2013-09-10")
> var newUpdateTime = ISODate("2013-09-12")
>
> lastUpdateTime
ISODate("2013-09-10T00:00:00Z")
> newUpdateTime
ISODate("2013-09-12T00:00:00Z")
>
> var id = new ObjectId()
> id
ObjectId("52310502f3bf4823f81e7fc9")
>
> // collection is empty, first update will do insert:
> db.testcol.update(
... {"_id" : id, "ts" : { $lt : lastUpdateTime } },
... { $set: { ts: lastUpdateTime, data: 123 } },
... { upsert: true, multi: false }
... );
>
> db.testcol.find()
{ "_id" : ObjectId("52310502f3bf4823f81e7fc9"), "data" : 123, "ts" : ISODate("2013-09-10T00:00:00Z") }
>
> // try one more time to check that nothing happens (due to error):
> db.testcol.update(
... {"_id" : id, "ts" : { $lt : lastUpdateTime } },
... { $set: { ts: lastUpdateTime, data: 123 } },
... { upsert: true, multi: false }
... );
E11000 duplicate key error index: test.testcol.$_id_ dup key: { : ObjectId('52310502f3bf4823f81e7fc9') }
>
> var tooOldToUpdate = ISODate("2013-09-09")
>
> // update does not happen because query condition does not match
> // and mongo tries to insert with the same id (and fails with dup again):
> db.testcol.update(
... {"_id" : id, "ts" : { $lt : tooOldToUpdate } },
... { $set: { ts: tooOldToUpdate, data: 999 } },
... { upsert: true, multi: false }
... );
E11000 duplicate key error index: test.testcol.$_id_ dup key: { : ObjectId('52310502f3bf4823f81e7fc9') }
>
> // now query cond actually matches, so update rather than insert happens which works
> // as expected:
> db.testcol.update(
... {"_id" : id, "ts" : { $lt : newUpdateTime } },
... { $set: { ts: newUpdateTime, data: 999 } },
... { upsert: true, multi: false }
... );
>
> // check that everything worked:
> db.testcol.find()
{ "_id" : ObjectId("52310502f3bf4823f81e7fc9"), "data" : 999, "ts" : ISODate("2013-09-12T00:00:00Z") }
>
The only annoying part are those errors, but they are cheap and safe.
Problem
I've answered this question on LinkedIn and I thought it's something useful and interesting to share. The question was: "Suppose we have documents like `{_id: ..., data: ..., timestamp: ...}`. Is there any way to write update criteria which will satisfy following rules: 1 If there is no documents with following `_id` then insert this document; 2 If there is exists document with following `_id` then 2.1 If new timestamp greater then stored timestamp then update `data`; 2.2 Otherwise do nothing"