<aside> 🧱

Буллеты про NoSQL: MongoDB, Cassandra/ScyllaDB, DynamoDB, key-value/document stores, моделирование, консистентность и эксплуатация.

</aside>

MongoDB

• Designed document schemas for X collections, embedding hot relations to serve X% of reads in a single round trip.
• Cut p95 read latency from Xms to Xms by adding compound indexes and eliminating X collection scans.
• Migrated X million documents to a new schema version online with a background migrator and dual-read fallback.
• Built aggregation pipelines replacing X application-side joins, cutting report generation from Xs to Xms.
• Introduced change streams publishing X domain events/sec into downstream services without polling.
• Sharded a X TB collection on a compound shard key, keeping chunk distribution within X% across X shards.
• Reduced working-set memory X% by projecting only required fields and moving X blob fields to object storage.
• Configured majority write and read concerns for X critical flows, eliminating stale-read defects in production.
• Cut Atlas spend X%/month by right-sizing the cluster tier and adding TTL indexes on X ephemeral collections.
• Replaced unbounded arrays with the bucket pattern, keeping documents under X KB and stabilising update latency.
• Introduced JSON schema validation on X collections, rejecting X% malformed writes at the database layer.
• Moved X analytical queries off the primary to dedicated analytics nodes, removing X% of load from the write path.

Cassandra и ScyllaDB

• Modelled Cassandra tables query-first with X denormalized views, serving Xk reads/sec at sub-Xms p99.
• Tuned partition sizes below X MB using time-bucketed partition keys, eliminating X wide-partition timeouts per week.
• Migrated from Cassandra to ScyllaDB, cutting p99 latency from Xms to Xms and node count from X to X.
• Configured per-workload tunable consistency (QUORUM/LOCAL_ONE), cutting cross-region read latency by X%.
• Set up multi-datacenter replication across X regions with NetworkTopologyStrategy for disaster recovery.