How much RAM does n8n need in production: workers, PostgreSQL and Redis
Published on 2026-11-14 by frfrand
How to size n8n based on executions, concurrency, persistence, workflow memory, and worker usage.
# How much RAM does n8n need in production: workers, PostgreSQL and Redis
Is a small VPS enough for n8n? It depends less on how many workflows you have saved than on how many run simultaneously and what they do. An automation that moves small JSON data and waits for an API behaves very differently from one that processes large files or thousands of items. Capacity should be estimated with real load and margin for spikes.
Components that consume memory
In a simple installation, n8n can run the editor, webhooks, and workflows in the same process. In queue mode, main processes and workers are separated, with Redis as the queue intermediary and a persistent database, usually PostgreSQL. If all reside on the same VPS, they compete for memory and CPU.
The operating system, reverse proxy, monitoring agent, and backups also consume resources. Do not attribute all available RAM to Node.js.
Concurrency matters
Two flows with the same number of nodes can have very different consumption. Measure executions per minute, average duration and p95, maximum number of simultaneous executions, payload size, and presence of binaries. An operational approximation of average concurrency is arrival rate per second multiplied by average duration in seconds; it does not replace peak measurement.
A workflow that makes ten small sequential calls is not sized the same as one that downloads and transforms a large file, runs custom code, or aggregates thousands of records in memory.
Estimate the VPS with a trial
1. Identify the five heaviest workflows and their frequency. 2. Run a representative test without triggering real effects on customers. 3. Measure RSS of processes, available memory, CPU, p95, OOM errors, and queue time. 4. Repeat with the expected peak plus an explicit margin. 5. Adjust concurrency limits and number of workers before scaling horizontally.
Avoid promising a universal figure like "1 GB per ten workflows." It is a rule without technical basis.
PostgreSQL and Redis are not the same
PostgreSQL stores metadata and execution data according to n8n configuration; it is advisable to control retention and cleanup to avoid uncontrolled growth. Redis is used to coordinate the queue in distributed execution architectures, not to replace database backups. Review compatibility and official queue mode configuration before implementing it.
In a sensitive environment, consider hosting the database outside the main process to isolate consumption and facilitate recovery. This adds network and operations, so it must be justified.
When to add workers
If pending executions increase and the main process remains stable, an additional worker can increase capacity, provided CPU, memory, external APIs, and database allow it. If the real limit is an API with rate limiting, multiplying workers can worsen 429 errors.
Configure OOM, disk, accumulated queue, and persistent failure alerts. Test process restart under load before trusting the system.
Frequently asked questions
### Is Redis mandatory in n8n? Not in all installations; it is part of the usual queue mode.
Does more RAM make workflows faster?
Only when there is memory pressure or processes that accumulate data.
When to use PostgreSQL?
When you seek production persistence in line with the architecture and product recommendations.
Resources
Check OOM Killer on Linux and connection pooling. At ArduMaker, size the n8n VPS based on measurements and not on the number of flows.