Hq-ts -

❌ “Isn’t TimescaleDB with PostgreSQL replication enough?” → Streaming replication is async by default. HQ requires synchronous commit to 3 nodes before ack.

| Test | Expected HQ behavior | |------|----------------------| | Kill leader node mid-write | Client gets retryable error, write succeeds on new leader within 2 sec | | Corrupt one disk’s WAL | Database detects checksum mismatch, repairs from other replicas | | Delay network 500ms between nodes | Writes still commit (higher latency but no loss) | | Replay same write 100x | Only 1 stored point, idempotent key wins | | Query across a 6-hour gap | Repair job finds gap, marks it as “inferred” vs “original” | If you’re building an HQ-TS system today:

| If regular TSDB is… | Then HQ-TS is… | |---------------------|----------------| | A notebook for temperatures | A flight recorder on a space shuttle | | 99.9% accurate | 99.999% with cryptographic proof | | “We’ll fix missing points later” | “Every nanosecond is accounted for” |