You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Every date in a query result is formatted with the native Date.prototype.toISOString, directly in the Postgres adapter and through Parse._encode (which calls toJSON) in the MongoDB transform. That applies to createdAt and updatedAt of every object and to every Date field, so a large result formats a lot of dates.
The native method is slow for this. In V8 it formats each of the seven fields through a separate snprintf call. Measured on Node 20 and 24, formatting 60k dates (createdAt and updatedAt of 30k objects) takes about 35 ms with toJSON and about 26 ms with toISOString, against about 8 ms for a plain JavaScript formatter built from the getUTC* getters.
In a CPU profile of a MongoDB query for 10k objects with 20k included pointers, date formatting was about 10% of server CPU time.
Feature / Enhancement Description
Add a shared Utils.toISOString(date) that produces exactly the same output as the native method, using the getUTC* getters and lookup tables for zero padding. Years outside 1000-9999 use the extended year format and fall back to the native method, which also keeps the native RangeError for invalid dates.
Use it where query results are converted:
MongoDB: MongoTransform encodes dates through a local encodeDate, which keeps the output and invalid-date error of Parse._encode.
Postgres: postgresObjectToParseObject uses it in place of the native method.
Example Use Case
A client queries 10k objects, each with createdAt, updatedAt and a Date field.
The server formats 30k dates while converting the results.
With the faster formatter, the MongoDB request took about 68 ms instead of about 87 ms in a local measurement. On Postgres, conversion took about 9 ms less CPU per request, of which about 2 ms returned in JSON.stringify. The visible gain depends on how much of a request is spent formatting dates.
Alternatives / Workarounds
Removing only the Parse._encode dispatch around the native call was measured and made no measurable difference; the cost is the native formatting itself.
A faster toISOString in V8 would help all of Node, but would only reach future Node major versions, since each major keeps its V8 version.
3rd Party References
The fast-iso-string package takes the same approach in userland.
Activity
parse-github-assistant commentedon Oct 10, 2026
🚀 Thanks for opening this issue! We are excited about your ideas for improvement!