Progressive queries

The one sentence

A request with Progressive = true may be answered in pieces: a provider that reads several partitions emits its Initial as soon as it has rows, and delivers the rows of the partitions that answer later as Added changes. Without the flag nothing changes.

Why it exists

A query that names no partition reads every partition the caller can read. The PostgreSQL provider used to answer it with ONE statement over all of them, so the first row reached the caller only after the slowest partition had been read. On the public instance (2026-11-21) the search bar's free-text query read 173 partitions and its first row arrived after a median of 7 s, although most single partitions answer in tens of milliseconds (MeshWeaver.Plugins Hosting/SearchSuggestionLatency).

The contract

What the fan-in does with it

Nothing special, and that is the point. MeshQuery.MergeProviderObservables already gates its merged Initial on every provider's Initial and forwards later Added changes unchanged, and the snapshot surface (IMeshService.Query, CombineAnswers) already emits a frame per provider answer. A progressive provider's early Initial therefore reaches the caller at once, and its Added changes follow as further frames. ProgressiveQueryRequestTest pins this on both surfaces, and pins that a request without the flag still gets one complete Initial.

Who honours it

The PostgreSQL partitioned provider in MeshWeaver.Plugins (PostgreSqlPartitionedMeshQuery) for an unpinned fan-out: one statement per partition, a bounded number at a time, each through the I/O pool. Its first caller is the portal search bar (MeshSearch in MeshWeaver.Plugins).