Build1 publisher2 min readPublished
DocumentDB 0.117 ships index pushdown for scalar $group switched off by default
Microsoft-led DocumentDB 0.117 adds scalar $group index pushdown, so Postgres can sum an indexed field for a MongoDB query that carries no hint. Because the feature ships off, MongoDB users weighing DocumentDB swap a query hint for a server setting.
The Engineer · Build desk
What happened
- DocumentDB is an open-source PostgreSQL extension that exposes the MongoDB API and translates MongoDB operators into PostgreSQL access paths.
- With an index on a in place, MongoDB Atlas 8.0 answered the unhinted $sum by examining all 50,000 documents and zero index keys.
- The same documents, indexes and unhinted pipelines ran on MongoDB, DocumentDB 0.116 and DocumentDB 0.117, so 0.116 is the baseline for the new release.
- The 0.117 run enabled only documentdb.enable_scalar_aggregate_index_pushdown and left every other setting unchanged.
Compiled by The EngineerSomething wrong?How this is made
Why it matters
- cost On MongoDB 8.0 an unhinted scalar sum is paid for with a read of every full document, so collections with wide payloads pay most for a value a narrow index already holds.
- decision A DocumentDB evaluation on default settings will not exercise the pushdown, so a head-to-head against MongoDB has to set the flag before it tests this feature at all.
- constraint The evidence covers one pipeline shape, a scalar $group with _id null over one indexed field; grouped accumulators and filtered pipelines sit outside it.
MongoDB Atlas 8.0's planner produced a single plan: a GROUP stage over a COLLSCAN, with an empty `rejectedPlans` list [6]. The run took 12 milliseconds [7]. Every value the sum needed was in the `a_1` index [4].
The total can be checked by hand. The loader sets `a` to `(index + 1) % 100` [13], so each of the 100 values appears 500 times, and 500 times the sum of 0 through 99 (4,950) is 2,475,000 [1]. MongoDB returned exactly that [8]. To get there, the collection scan read past about 10 MB of 200-byte payloads that the sum never uses [2].
DocumentDB's answer comes from PostgreSQL. Version 0.117-0, released September 10, 2026, adds a path that serves the scalar accumulator from the index [1]. Microsoft leads development and builds on feedback from enterprise customers on Azure DocumentDB [3]. The post's author wrote that "the PostgreSQL query planner ensures optimal performance without requiring additional hints in the query" [10].
The claim holds for the query text, while the opt-in moves to the server as a setting the release ships switched off [5]. I think off is the right default for a first release. A change to the access path of a common aggregation belongs behind an opt-in until operators have read the explain plans on their own collections.
The test design is careful. Each DocumentDB gateway run split into setup, VACUUM (ANALYZE) and query phases [11]. Timings across separate containers would mostly measure the containers, so the author compared rows, heap fetches and buffers instead [9].
The excerpt of the post ends inside MongoDB's explain output, before the DocumentDB counters appear. In this record, the case for 0.117 is the author's statement set against a MongoDB plan shown in full [6][10]. For MongoDB's side of the result to carry over to a production collection, that collection needs what this one has: an index on the summed field, and documents much wider than that field [4].
What to watch
- Whether a later DocumentDB release enables scalar $group index pushdown by default.
- The heap-fetch and buffer counts for DocumentDB 0.116 and 0.117 from the same test, set against MongoDB's 50,000 examined documents.
- Whether Azure DocumentDB, whose enterprise customers' feedback shapes the extension, exposes the pushdown setting to them.