Query Engine

Cross-CRDT queries — scan, filter, and aggregate across multiple CRDTs in a single request.

The Query Engine lets you aggregate data across multiple CRDTs in a single HTTP request. Instead of reading CRDTs one by one, you can sum all page view counters, union all shopping carts, or find the latest config register — with a single query scoped to a namespace.

​

HTTP endpoint

POST /v1/namespaces/:ns/query
Authorization: Bearer <token>
Content-Type: application/json

​

Request

{
  "from": "gc:views-*",
  "type": "gcounter",
  "aggregate": "sum"
}

Field	Type	Required	Description
from	string	✓	Glob pattern matched against CRDT IDs. * matches any sequence of characters.
type	string	—	Filter by CRDT type. When omitted, inferred from the from prefix (e.g. gc: → gcounter).
aggregate	string	✓	Aggregation function. Must be compatible with the matched CRDT types.
where	object	—	Optional content filter applied before aggregation.

​

Response

{
  "value": 42,
  "matched": 5,
  "scanned": 5,
  "execution_ms": 0.8
}

Field	Description
value	Aggregated result. Shape depends on aggregate. null when nothing matched.
matched	Number of CRDTs that passed glob + type + where filters.
scanned	Total number of CRDTs scanned in the namespace.
execution_ms	Server-side wall-clock execution time in milliseconds.

---

​

Aggregations by CRDT type

CRDT type	type value	Supported aggregations
GCounter	gcounter	sum, max, min, count
PNCounter	pncounter	sum, max, min, count
ORSet	orset	union, intersection, count
LwwRegister	lwwregister	latest, collect
Presence	presence	union, count
CRDTMap	crdtmap	collect
RGA	rga	collect
Tree	tree	collect

Mixing CRDT types in a single query (e.g. from: "*" without a type filter) and using a type-specific aggregation returns 400 incompatible_aggregate.

---

​

Examples

​

Sum all page view counters

{
  "from": "gc:views-*",
  "aggregate": "sum"
}

{ "value": 10420, "matched": 12, "scanned": 12, "execution_ms": 1.2 }

​

Count how many carts contain a product

{
  "from": "or:cart-*",
  "aggregate": "count",
  "where": { "contains": { "id": "prod-42" } }
}

{ "value": 3, "matched": 3, "scanned": 47, "execution_ms": 2.1 }

​

Union all carts

{
  "from": "or:cart-*",
  "aggregate": "union"
}

{ "value": ["prod-1", "prod-2", "prod-42"], "matched": 47, "scanned": 47, "execution_ms": 3.4 }

​

Latest config value across all regions

{
  "from": "lw:config-*",
  "aggregate": "latest"
}

{ "value": { "featureFlags": { "newUI": true } }, "matched": 3, "scanned": 3, "execution_ms": 0.5 }

​

Config registers updated in the last minute

{
  "from": "lw:config-*",
  "aggregate": "collect",
  "where": { "updatedAfter": 1700000000000 }
}

---

​

where clause

Field	Applicable to	Description
contains	ORSet	Only include sets that contain this JSON value (exact match).
updatedAfter	LwwRegister	Only include registers updated after this Unix timestamp in ms.

---

​

SDK — client.query()

const result = await client.query({
  from: "gc:views-*",
  aggregate: "sum",
});

console.log(result.value);       // number
console.log(result.matched);     // number of CRDTs matched
console.log(result.execution_ms); // server latency

​

Fluent type inference

The aggregate field is typed as a string literal union — TypeScript will catch invalid combinations at compile time when you use the QuerySpec type directly.

import type { QuerySpec } from "meridian-sdk";

const spec: QuerySpec = {
  from: "or:tags-*",
  aggregate: "union", // TS error if you write "sum" here
};

---

​

React — useQuery()

import { useMemo } from "react";
import { useQuery } from "meridian-react";

function TotalViews() {
  // Stabilize the spec with useMemo to avoid re-fetching on every render.
  const spec = useMemo(
    () => ({ from: "gc:views-*", aggregate: "sum" as const }),
    []
  );

  const { data, loading, error } = useQuery(spec);

  if (loading) return <span>Loading…</span>;
  if (error) return <span>Error: {error.message}</span>;

  return <span>Total views: {data?.value}</span>;
}

useQuery re-runs the query when spec changes. Use useMemo to stabilize the spec object and avoid unnecessary requests on every render.

---

​

Key IDs convention

The query engine matches on the full CRDT ID as stored — whatever string you pass to client.gcounter(id), client.orset(id), etc. Using a prefix convention like gc:views-homepage makes glob patterns predictable and enables type inference from the pattern.

gc:views-homepage    →  matched by  gc:views-*
gc:views-dashboard   →  matched by  gc:views-*
pn:score-user-42     →  matched by  pn:score-*
or:cart-user-42      →  matched by  or:cart-*

Type inference from prefix:

Prefix	Inferred type
gc:	gcounter
pn:	pncounter
or:	orset
lw:	lwwregister
pr:	presence
cm:	crdtmap
rga:	rga
tree:	tree