Qdrant Scaling
Route first, then answer. Match the user's symptom in the table, Read that file, and answer from it.
Do not answer from this page alone: it contains routing only, not the guidance. If two rows match, read both.
| The user says | Read |
|---|---|
| Data does not fit on a single node, running out of disk or memory as the dataset grows | scaling-data-volume/SKILL.md |
| Need to shard the collection across more nodes, data outgrew one node | scaling-data-volume/SKILL.md |
| Cannot handle enough parallel queries, need higher QPS or throughput | scaling-qps/SKILL.md |
| Can't hold the request rate, CPU is pegged | scaling-qps/SKILL.md |
| A single query is too slow, need to cut the tail latency of individual requests | minimize-latency/SKILL.md |
| p99 or tail latency too high, but traffic/QPS is fine | minimize-latency/SKILL.md |
| Queries return very large result sets and slow down | scaling-query-volume/SKILL.md |
Large limit, top-1000 queries, pagination, scroll across shards | scaling-query-volume/SKILL.md |
| Many tenants or customers, one collection each, tenant isolation | scaling-data-volume/tenant-scaling/SKILL.md |
| Only recent data matters, retention, expiring old vectors, time-based rotation | scaling-data-volume/sliding-time-window/SKILL.md |
| Single node no longer fits the workload, before deciding to shard | scaling-data-volume/vertical-scaling/SKILL.md |
| Already vertically maxed out, need more nodes, resharding | scaling-data-volume/horizontal-scaling/SKILL.md |
Latency and throughput pull opposite ways on segment count.
For latency, increase segments toward the CPU core count (default_segment_number: 16).
For throughput, use fewer and larger segments (default_segment_number: 2).
Applying the wrong direction makes the reported problem worse.