Background

When MongoDB receives an insert, it needs to decide whether to perform a regular insert or a timeseries insert. This decision currently happens independently in two places with duplicated logic.

Current code structure

Path 1: Direct inserts via CmdInsert

When a client sends an insert command, CmdInsert::typedRun() calls getCollectionPreConditionsAndIsTimeseriesLogicalRequest() to check if the target collection is timeseries. Based on the result it calls either performTimeseriesWrites() or performInserts().

Path 2: $out aggregation stage via MongoProcessInterface

The $out stage writes documents through MongoProcessInterface, which exposes two separate virtual methods: insert() and insertTimeseries(). The $out stage determines whether the output collection is timeseries during initialization, stores this in a _timeseries flag, and at flush time branches to call one or the other.

On the implementation side, NonShardServerProcessInterface::insertTimeseries() internally calls getCollectionPreConditionsAndIsTimeseriesLogicalRequest() and then performTimeseriesWrites() — the same logic that CmdInsert does. The sharded implementation ShardServerProcessInterface::insertTimeseries() simply delegates to insert(), since the timeseries dispatch happens on the receiving shard.

The problem

The timeseries-vs-regular branching logic is duplicated: once in CmdInsert::typedRun() and once split across NonShardServerProcessInterface::insert() / insertTimeseries(), with $out also carrying its own detection logic.

Proposal

1. Create a unified write function (e.g. in write_ops_exec) that encapsulates the timeseries branching: call getCollectionPreConditionsAndIsTimeseriesLogicalRequest(), then dispatch to performTimeseriesWrites() or performInserts() accordingly.
2. CmdInsert::typedRun() calls this unified function instead of branching manually.
3. Remove the insertTimeseries() virtual method from MongoProcessInterface entirely. The single insert() method on NonShardServerProcessInterface calls the same unified function. ShardServerProcessInterface is unchanged (it already delegates timeseries to regular insert).
4. Simplify $out — stop tracking _timeseries state for the write path. Always set the collection UUID on the insert command. Just call insert() and let the unified function handle timeseries detection.