This is the cursed knowledge I have learned throughout my career and home labbing. The kind of knowledge that makes you pull out your hair.
-
AWS Glue Schema Registry Has No Bulk API
2026/02While building an indexer to cache all schemas locally to eliminating slow on-demand AWS API calls from the UI. The hard truth became clear: there is no bulk API.
Every full index requires a waterfall of calls per schema/per schema version:
ListSchemas- paginated list, limited fields only (no version info, no schema definition)GetSchema- full metadata for one schema at a timeListSchemaVersions- again, limited fieldsGetSchemaVersion- full definition for one version at a time
For 1,000 schemas with 10 versions each, that’s ~12,000 API calls per index
The only escape: compare
updatedTimefromListSchemasagainst stored state and skip schemas that haven’t changed. But still requires to do a full scan every 15 mins to catch schema version changes.AWS / Kafka -
Synology's Fake Authentication
2025/10Synology’s Hyper Backup forces you to enter username and password for rsync daemon connections, but completely ignores whatever you type. The UI screams “AUTHENTICATION REQUIRED!” while the backend whispers “actually, I don’t care.” Just type anything and move.NAS/Backup -
Prometheus Doesn't Deal with Exactness
2025/07Prometheus’srate()andincrease()functions use extrapolation instead of actual data, losing half your measurements and multiplying results incorrectly. Your “precise” business metrics are actually systematic lies.Monitoring -
CEL Java's Null Paradox
2025/06In Google’s CEL Java, Javanullisn’t actually null - it becomesCelUnknownSet, makingvalue == nullreturn unknown instead of true/false. You must useNullValue.NULL_VALUEinstead of actual null to represent “null” values.Expression Language