Diagnose document validation, type and unique-key failures

Understand the likely causes

• Validation can fail on missing required fields, wrong BSON types, ranges or the proposed post-update shape. Uniqueness can fail on existing indexed values, collation equivalence or optional null/missing keys. A partial index constrains only its declared subset. Mixed schema versions and driver serialization can create both visible failures and silent report omissions.

Preserve the structured failure and rejected intent

Keep the error code, named index or rule details, operation type and request identifier, with sensitive values redacted. Code 121 and code 11000 need different remedies; do not catch both as harmless duplicates. Inspect what the complete proposed document would look like after an update or replacement. Replacing a document can remove required fields that were not included, even when the original stored version was valid. Preserve the rejected business event so repair does not silently discard legitimate work.

Read the active collection and index contracts

An existing authorized training session can inspect collection options and indexes. Check validator, validationLevel and validationAction together with unique, partial, sparse and collation properties. These definitions describe rules, not proof that all old documents are valid. Adding a validator does not retroactively convert historical data. The methods below are read-only illustrations; no real definitions are changed and no production metadata is fetched during authoring. A missing privilege should be handled through the normal review process rather than broadening application access.

db.getSiblingDB('quay_lab').getCollectionInfos({ name: 'inventory' });
db.getSiblingDB('quay_lab').inventory.getIndexes();

Inspect actual BSON types, missing and null

Group the affected population by type and schema version before deciding how to repair it. The aggregation expression $type can label missing and null separately, unlike an equality-to-null filter that can match both. A validator requiring bsonType int need not accept a double or string that prints like the same quantity. Date strings likewise are not BSON dates. Define accepted conversions, source time zones and exact numeric ranges, and quarantine invalid values instead of converting every failure to zero or empty text.

db.getSiblingDB('quay_lab').inventory.aggregate([
  { $group: { _id: { version: '$schemaVersion', qtyType: { $type: '$qty' } }, count: { $sum: 1 } } },
  { $sort: { '_id.version': 1, '_id.qtyType': 1 } }
]);

Repair the correct rule without bypassing it

For validation failure, correct the incoming shape or stage a reviewed compatible schema migration. For duplicate keys, identify whether the request is an idempotent retry or a genuinely conflicting record; return a recorded result only when event identity proves it. Optional fields may need a deliberately designed partial unique index, not deletion of arbitrary duplicates. A unique multikey index does not necessarily forbid repeated elements inside one document, and sharded uniqueness has routing-key restrictions. Review those contexts before changing constraints or inventing cross-document guarantees.

Keep it from recurring

• Keep API, driver BSON and collection rules aligned; preserve structured errors and rejected-event evidence. Document optional-field uniqueness, schemaVersion transitions and conversion policy. Validate existing collisions before index rollout and avoid granting routine validation bypass to application roles merely to make imports appear successful.