How to fix Firestore: Document exceeds 1 MiB maximum size in Firebase

Serialize the payload and count the UTF-8 bytes so you can stop short of Firestore's 1 MiB limit:

const payload = { /* your document */ };
const bytes = new TextEncoder().encode(JSON.stringify(payload)).length;
console.log(`Preparing to write ${bytes} bytes (limit: 1,048,576)`);
if (bytes >= 1_048_576) {
  throw new Error('Document is too large; break it into smaller pieces first.');
}

TextEncoder counts the same UTF-8 bytes that Firestore uses and includes field names plus Maps/arrays. Run this before every bulk write or inside Cloud Functions so you can log which document boundary is about to break the limit and avoid the failed write altogether.

Keep the root document lean by storing heavy arrays or maps in dedicated child documents:

const batch = db.batch();
batch.set(db.doc('offers/offerA'), { id: 'offerA', summary: '...', lastUpdated: Date.now() });
items.forEach((item, index) => {
  const ref = db.collection('offers/offerA/items').doc(String(index));
  batch.set(ref, item);
});
await batch.commit();

Subcollections (e.g. offers/offerA/items) do NOT count toward the parent document's 1 MiB limit. Use pagination or chunked keys when writing dozens of children so each child document stays small. Store only a lightweight pointer object back in the parent document so updates stay fast.

Avoid embedding base64 blobs in Firestore fields because the serialized string plus field name easily exceeds 1 MiB:

const storageRef = firebase.storage().ref();
const uploadTask = storageRef.child(`uploads/${uid}/large-file.bin`).put(fileBlob);
const snapshot = await uploadTask;
const downloadUrl = await snapshot.ref.getDownloadURL();
await firestore.doc(`uploads/${uid}/meta`).set({ url: downloadUrl, size: fileBlob.size });

Now only a few metadata fields live in Firestore, while the bytes stay in Cloud Storage where Firestore's document-size limit does not apply. Combine this with Firestore security rules that check the Storage path if you need to enforce ownership.

Field names and Maps count toward the document size, so verbose keys hurt when every document stores similar data. Instead:
1. Replace repeated metadata maps with references to a shared document (e.g. settings/{orgId}) and store only the ID in each record.
2. Trim long field names or rely on fixed keys so the same structure is reused rather than duplicated per document.
3. Use subcollections for optional attributes instead of inflating the root document.
4. Delete unused fields, and rely on Cloud Functions to compute derived data when needed, rather than storing it permanently.
These changes reduce the serialized byte count and keep you safely under the 1 MiB limit.